CN107688343B - Energy control method of hybrid power vehicle - Google Patents
Energy control method of hybrid power vehicle Download PDFInfo
- Publication number
- CN107688343B CN107688343B CN201710645889.2A CN201710645889A CN107688343B CN 107688343 B CN107688343 B CN 107688343B CN 201710645889 A CN201710645889 A CN 201710645889A CN 107688343 B CN107688343 B CN 107688343B
- Authority
- CN
- China
- Prior art keywords
- working condition
- vehicle
- prediction
- vehicle speed
- current
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 86
- 238000005070 sampling Methods 0.000 claims abstract description 21
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 19
- 238000005457 optimization Methods 0.000 claims abstract description 17
- 230000008569 process Effects 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 230000001133 acceleration Effects 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 16
- 230000008859 change Effects 0.000 claims description 15
- 238000003064 k means clustering Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 238000003062 neural network model Methods 0.000 claims description 5
- 230000006399 behavior Effects 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 description 18
- 239000000446 fuel Substances 0.000 description 11
- 102100030462 Pleckstrin homology domain-containing family B member 1 Human genes 0.000 description 10
- 101150031793 Plekhb1 gene Proteins 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 7
- 230000007704 transition Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000006872 improvement Effects 0.000 description 5
- 210000002569 neuron Anatomy 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 101001126098 Homo sapiens Pleckstrin homology domain-containing family B member 2 Proteins 0.000 description 3
- 102100030463 Pleckstrin homology domain-containing family B member 2 Human genes 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 238000005096 rolling process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 240000008168 Ficus benjamina Species 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000010396 two-hybrid screening Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/02—Control of position or course in two dimensions
- G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
- G05D1/0257—Control of position or course in two dimensions specially adapted to land vehicles using a radar
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/02—Control of position or course in two dimensions
- G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
- G05D1/0276—Control of position or course in two dimensions specially adapted to land vehicles using signals provided by a source external to the vehicle
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/02—Control of position or course in two dimensions
- G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
- G05D1/0276—Control of position or course in two dimensions specially adapted to land vehicles using signals provided by a source external to the vehicle
- G05D1/0278—Control of position or course in two dimensions specially adapted to land vehicles using signals provided by a source external to the vehicle using satellite positioning signals, e.g. GPS
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Aviation & Aerospace Engineering (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Electric Propulsion And Braking For Vehicles (AREA)
- Hybrid Electric Vehicles (AREA)
Abstract
The invention discloses an energy control method of a hybrid vehicle, which comprises the following steps: (1) observing the current system state, including the vehicle running speed, the driver pedal information, the battery SOC, the engine rotating speed and torque and the like; (2) judging the current EVT mode state according to the vehicle running speed and the driver pedal information, updating the current system model and system constraints, and assuming that the EVT mode state is kept unchanged in the prediction time domain; (3) predicting the future vehicle speed in the prediction time domain to obtain the system observation input quantity in the prediction time domain; (4) constructing a prediction control optimization problem in a prediction time domain, and performing numerical solution on line through a dynamic programming algorithm; (5) calculating to obtain an optimal control sequence in a prediction time domain; (6) only the first group of optimal control quantity is adopted, the optimal control quantity acts on the system at the sampling moment, and the rest control quantity is abandoned; (7) this process is repeated at the next moment until the end of the trip.
Description
Technical Field
the present invention relates to a method for controlling energy of a vehicle, and more particularly, to a method for controlling energy of a hybrid vehicle.
Background
Hybrid vehicles are one of the effective ways to solve the problems of excessive vehicle energy consumption and air quality pollution at present. Compared with other hybrid power transmission schemes, the dual-mode transmission system can better meet the special requirements of a heavy off-road vehicle on wide speed regulation range, large driving power and the like, but the scheme is complex in structure and has higher requirements on an energy control method, and the design of the optimal energy control method capable of being used in real time becomes the core content for ensuring the normal and efficient operation of the dual-mode hybrid power vehicle.
at present, the most used method in the industry is a rule-based energy control method, the rule design mostly comes from heuristic discovery and engineer experience, and although the rule design is simple and easy to implement, the rule design has poor adaptability to different working conditions and cannot obtain the optimal control effect. In order to pursue a better control effect, a great amount of scientific research is carried out in academia to explore an energy control method based on optimization, and the main idea is to establish a system target cost function and constraint conditions and obtain an optimal control quantity through an optimization algorithm. The dynamic programming algorithm is most widely applied, but the global working condition is required to be known in advance, so that the dynamic programming algorithm can only be used for simulation. The equivalent fuel consumption strategy can be applied on line in real time, but has the defect that equivalent factors are difficult to set aiming at different working conditions. In recent years, a Predictive Control algorithm (MPC) developed in the past adopts the ideas of multi-step testing, rolling optimization and feedback correction, and obtains a good real-time Control effect. The method relies on predicting the future vehicle speed to a great extent, and the future vehicle speed is assumed to be kept unchanged in the prior art; or if the vehicle speed changes according to an exponential law, the methods are simple but not accurate; or obtaining the future running speed of the vehicle by means of a vehicle-mounted navigation system; or the future speed is predicted by identifying the repeated working conditions of special working vehicles, and the methods need to use a GPS system or prior working condition information and are not suitable for non-road vehicles without a positioning system and a perception radar.
disclosure of Invention
aiming at the problems, the invention provides an energy control method of a hybrid vehicle, which classifies working conditions by adopting a K-means (Kmeans) clustering algorithm, predicts the future vehicle speed by adopting a Markov chain or radial basis function neural network method aiming at different types of working conditions and realizes the improvement of the performance of the non-road dual-mode hybrid vehicle.
The main object of the present invention is to provide an energy control method of a hybrid vehicle.
The object of the present invention can be achieved by:
A method of controlling energy of a hybrid vehicle, the method comprising:
(1) observing the current system state, including the vehicle running speed, the driver pedal information, the battery SOC, the engine rotating speed and torque and the like;
(2) judging the current EVT mode state according to the vehicle running speed and the driver pedal information, updating the current system model and system constraints, and assuming that the EVT mode state is kept unchanged in the prediction time domain;
(3) predicting the future vehicle speed in the prediction time domain to obtain the system observation input quantity in the prediction time domain;
(4) Constructing a prediction control optimization problem in a prediction time domain, and performing numerical solution on line through a dynamic programming algorithm;
(5) calculating to obtain an optimal control sequence in a prediction time domain;
(6) Only the first group of optimal control quantity is adopted, the optimal control quantity acts on the system at the sampling moment, and the rest control quantity is abandoned;
(7) and (5) repeating the steps (1) to (6) at the next moment until the driving is finished.
The energy control method of a hybrid vehicle of the invention, further wherein the following characteristic parameters are used: the vehicle running conditions are classified by maximum acceleration (m/s2), maximum deceleration (m/s2), average acceleration (m/s2), vehicle speed standard deviation (km/h), difference between maximum vehicle speed and minimum vehicle speed (km/h), and acceleration standard deviation (m/s 2).
The energy control method of the hybrid vehicle further adopts a K-means clustering algorithm to classify data by calculating the degree of affinity and sparseness among samples, finally realizes that the data in the same class have larger characteristic similarity and the difference between different classes is larger, and the specific working condition judgment steps are as follows:
an off-line stage: (1) combining a plurality of standard cycle conditions to form a sample;
(2) Calculating the working condition characteristic parameters of 10 seconds in the past at each sampling moment in the cycle working condition to obtain characteristic parameter sample data [ x11, x 12.,. x1m ], [ x21, x 22.,. x2m ], … …, [ xn1, xn 2.,. xnm ], wherein m is the characteristic parameter ordinal number, and n is the cycle working condition length;
(3) Applying a K-means clustering algorithm, randomly selecting a clustering center c1 [ [ c11, c 12., c1m ], c2 [ [ c21, c 22., c2m ], calculating distances between all samples and the clustering center, grouping the samples according to a nearest neighbor rule, and attributing different theta m (K) clustering domains, wherein K is iteration times, nm is cycle working condition length contained in the mth class characteristic parameter, xi is a characteristic parameter sample value, and then adjusting the clustering center according to the following formula:
if cm (k +1) is not equal to cm (k), continuing to adjust the clustering center until the clustering center does not change any more, and considering that the classification is stable to obtain the clustering center c1 under a stable working condition and the clustering center c2 under a quick change working condition;
an online stage: (1) in the actual running process of the vehicle, working condition characteristic parameter values [ x1, x 1.., xm ] of 10 seconds in the past are calculated at the current sampling moment;
(2) The distance d of the characteristic parameter values [ x1, x 1.., xm ] to the two cluster centers c1 and c2 is calculated according to the following formula:
In the formula: j is 1,2 corresponds to two working conditions;
(3) And if d1 is not more than d2, judging that the current time is a stable working condition, and if d1 is more than d2, judging that the current time is a quick-change working condition.
the energy control method of the hybrid vehicle is characterized in that the acceleration of the vehicle at each moment is assumed to be irrelevant to historical information and is only determined by current information under the stable working condition, so that the acceleration change of the vehicle is considered to be a Markov process, a Markov chain model is used for simulating the change rule of the vehicle speed and the acceleration, and the future vehicle speed is predicted under the stable working condition.
The energy control method of the hybrid vehicle further comprises the step of predicting the future vehicle speed by learning the driving behavior of the driver on line by applying a radial basis function neural network theory according to the quick change working condition.
The energy control method of the hybrid power vehicle further comprises the steps that the radial basis function neural network model predicts the future vehicle speed according to the historical vehicle speed and the current pedal information of the driver, meanwhile, the current generated vehicle information is used as new historical information, and the self-adaptive online learning of the neural network is achieved through the method of selecting a self-organization center and determining a weight value through a pseudo-inverse method.
Drawings
FIG. 1 is a schematic block diagram of a two-mode hybrid vehicle system configuration.
FIG. 2 is a flow chart of the energy control method of the present invention.
FIG. 3 is a schematic of the combined cycle of the present invention.
Figure 4 is a diagram of a markov chain model transition probability matrix of the present invention.
FIG. 5 is a block diagram of a radial basis function neural network vehicle speed prediction architecture of the present invention.
FIG. 6 is a graph of online condition determination results of the present invention.
Fig. 7 is a vehicle speed prediction result map of the invention.
FIG. 8 is a graph comparing predicted error for partial cycle conditions.
FIG. 9 is a graph of vehicle speed, EVT mode, and battery SOC simulation results for the present invention.
Fig. 10 is a graph showing simulation results of rotational speed and torque of the engine, the motor a, and the motor B according to the present invention.
Fig. 11 is a diagram showing a simulation result of the engine operating point distribution of the present invention.
FIG. 12 is a comparison of engine operating point simulation results of the present invention: (a) predictive control, (b) hold control, (c) dynamic programming, (d) rules.
FIG. 13 is a comparison of battery SOC simulation results of the present invention: (a) predictive control, (b) hold control, (c) dynamic programming, (d) rules.
Detailed Description
the present invention will be described with respect to a particular embodiment of a two-mode hybrid vehicle. The system structure is shown in fig. 1, and the main parameters are shown in table 1. The system can realize the switching of two hybrid power operation modes, namely an EVT1 (EVT) mode and an EVT2 mode by operating a clutch and a brake, wherein the EVT1 mode is realized when the clutch is separated and the brake is engaged; EVT2 mode is selected when the clutches are engaged and the brakes are disengaged.
TABLE 1 two-mode hybrid vehicle principal parameters
In control-oriented modeling, inertia of the planet wheels and friction between elements are ignored, and assuming that the connections are rigid, a model of the transmission system can be obtained,
EVT1 modes:
EVT2 modes:
In the formula: k1, k2 and k3 are intrinsic parameters of the three planetary rows respectively, ω A and ω B are the rotating speed of the motor A, B, TA and TB are the torque of the motor A, B, ω i and ω o are the rotating speeds of the input end and the output end of the coupling mechanism, and Ti and To are the torque of the input end and the output end of the coupling mechanism respectively.
At the same time, due to the mechanical connection between the components, the system also satisfies the following relation:
ω＝iω (5)
V＝rω/i (6)
T＝T/i (7)
T＝iT (8)
In the formula: ω e is engine speed, Te is engine torque, V is vehicle speed, Tw is output torque on the wheels, iq is front drive ratio, rw is wheel radius, if is rear drive ratio.
the engine model uses a MAP model constructed from experimental data, assuming the engine has been fully warmed up and its specific fuel consumption is a static function of speed and torque:
In the formula: mf is engine fuel consumption and fe is MAP.
The SOC of the battery is an important variable in the energy control method of the hybrid vehicle, and the SOC is modeled by adopting an internal resistance model, and the influence of temperature is ignored to obtain the following SOC expression:
in the formula: voc is the open circuit voltage of the battery, Rbatt is the internal resistance of the battery, Cbatt is the battery capacity, η a, η B are the efficiencies of the motor A, B, respectively, and the indices kA, kB are equal to 1 when the motor charges the battery and equal to-1 when the motor discharges the battery.
And (3) regardless of the transverse and vertical motions of the vehicle and regardless of the gradient, obtaining a complete vehicle dynamic model according to a vehicle running balance equation:
In the formula: m is the total vehicle mass, rho is the air density, Cd is the air resistance coefficient, Af is the windward area of the vehicle, mu is the wheel rolling resistance coefficient, and g is the gravitational acceleration.
The energy control method of the dual-mode hybrid vehicle mainly aims to reasonably distribute required power on line and adjust the working point of an engine under the condition of meeting driving requirements and system constraints so as to obtain better fuel economy and maintain the SOC of a battery. The invention carries out real-time optimization of power distribution on line based on a predictive control algorithm, selects the rotating speed and the torque of an engine as a system control quantity u, defines the state quantity of the system as x, the observation input quantity of the system as v and the output quantity of the system as y, and can express a model facing a control system as follows:
in the formula: x ═ SOC ], u ═ Te ω e ] T, V ═ V Tw ] T,
At each sampling instant k, the optimal objective function in the prediction time domain is:
in the formula: ws and wm are respectively the weight coefficients of the corresponding items, SOCr is the reference value of the SOC of the battery, and P is the prediction time domain. At the same time, the following physical constraints need to be satisfied:
in the formula: and the values of max and min are the upper and lower limits of the corresponding items respectively.
When the optimization problem is solved, a system model is discretized, due to the fact that a prediction time domain is short, and the range of a feasible domain of the battery SOC at each moment is small, a dynamic programming algorithm can be used for solving the optimization problem on line in real time, and if U (k) ([ U (k) ], and
u(x(k))＝u(k) (15)
The core idea of prediction control is to solve an optimization problem in a finite prediction time domain at each sampling moment, calculate an optimal control sequence in the prediction time domain, only implement the optimal control of the sampling moment and abandon other control quantities, and repeat the process at the next sampling moment. The predictive control is used in the energy control method of the dual-mode hybrid vehicle, namely, the power distribution can be carried out through real-time optimization according to the current pedal information of a driver and vehicle information such as vehicle speed, battery SOC (state of charge), engine rotating speed torque and the like, and the fuel economy of the vehicle is improved. Because the whole cycle working condition cannot be predicted, the strategy cannot obtain a global optimal solution, but the strategy can be implemented on line, a global approximate optimal solution is obtained in a rolling optimization mode, and meanwhile uncertainty caused by factors such as model mismatch and interference can be considered, so that the control is kept to be actually optimal. At each sampling instant k, the energy control method flow chart is shown in fig. 2, and specifically, the following steps are performed:
(1) current system states are observed, including vehicle speed, driver pedal information, battery SOC, and the like.
(2) And judging the current EVT mode state according to the vehicle running speed and the driver pedal information, and updating the current system model and the system constraint. And assumes that the EVT mode state remains unchanged within the prediction domain.
(3) And predicting the future vehicle speed in the prediction time domain to obtain the system observation input quantity v in the prediction time domain, wherein the detailed method is described below.
(4) And constructing a prediction control optimization problem in a prediction time domain, and performing numerical solution on line through a dynamic programming algorithm.
(5) And calculating to obtain an optimal control sequence in a prediction time domain.
(6) only the first group of optimal control quantities is adopted, the system is acted on at the sampling moment, and the rest control quantities are abandoned.
(7) repeating steps (1) to (6) at the next time.
Under the condition of no prior information of the running condition, how to reasonably and accurately predict the future vehicle speed of the vehicle by using the vehicle history and the current data influences the optimization effect of the energy control method to a great extent. The invention classifies the working conditions into a stable working condition and a fast-changing working condition under an off-line state by utilizing a K-means clustering algorithm, and judges the type of the current working condition of the vehicle in real time at an on-line stage. And aiming at the stable working condition, a Markov chain-based vehicle speed prediction method is adopted, and aiming at the fast-changing working condition, a radial basis neural network-based vehicle speed prediction method is adopted, so that the advantages of the two methods are comprehensively utilized to achieve the optimal prediction effect. Meanwhile, the predicted vehicle speed is substituted into the formula (11) to calculate the required torque in the predicted time domain.
the main difference between the stable working condition and the fast-changing working condition is the fluctuation of the speed and the magnitude of the acceleration in the working condition, in order to distinguish two working condition types, the two working conditions need to be classified according to the characteristic parameters in the working condition, and table 2 shows the selected working condition characteristic parameters.
TABLE 2 characteristic parameters of the operating conditions
the method adopts a K-means clustering algorithm, carries out data classification by calculating the degree of affinity and sparseness among samples, finally realizes that the data in the same class have larger feature similarity, and the difference is larger among different classes, and the specific working condition judgment steps are as follows:
An off-line stage: (1) the multiple standard cycle conditions were combined to form a sample, as shown in FIG. 3.
(2) And calculating the working condition characteristic parameters of 10 seconds at each sampling moment in the cycle working condition to obtain characteristic parameter sample data [ x11, x 12.,. x1m ], [ x21, x 22.,. x2m ], … …, [ xn1, xn 2.,. xnm ], wherein m is the characteristic parameter ordinal number, and n is the cycle working condition length.
(3) applying a K-means clustering algorithm, randomly selecting a clustering center c1 [ [ c11, c 12., c1m ], c2 [ [ c21, c 22., c2m ], calculating distances between all samples and the clustering center, grouping the samples according to a nearest neighbor rule, and attributing different theta m (K) clustering domains, wherein K is iteration times, nm is cycle working condition length contained in the mth class characteristic parameter, xi is a characteristic parameter sample value, and then adjusting the clustering center according to the following formula:
And if cm (k +1) ≠ cm (k), continuing to adjust the clustering center until the clustering center does not change any more, considering that the classification is stable, and obtaining the clustering center c1 under the stable working condition and the clustering center c2 under the quick change working condition.
an online stage: (1) and during the actual running process of the vehicle, calculating the condition characteristic parameter values [ x1, x 1.., xm ] of the past 10 seconds at the current sampling time.
(2) The distance d of the characteristic parameter values [ x1, x 1.., xm ] to the two cluster centers c1 and c2 is calculated according to the following formula:
in the formula: j 1 and 2 correspond to two working conditions.
(3) And if d1 is not more than d2, judging that the current time is a stable working condition, and if d1 is more than d2, judging that the current time is a quick-change working condition.
The acceleration of the vehicle at each moment is assumed to be unrelated to historical information and is determined only by current information, so that the acceleration change of the vehicle is considered to be a Markov process, and the Markov chain model can be used for simulating the change rule of the vehicle speed and the acceleration and predicting the future vehicle speed under the stable working condition.
six groups of corresponding first-order Markov chain models are established according to the conditions that the pedal opening alpha of different drivers is less than or equal to 0, the alpha is more than 0 and less than or equal to 0.2, the alpha is more than 0.2 and less than or equal to 0.4, the alpha is more than 0.4 and less than or equal to 0.6, the alpha is more than 0.6 and less than or equal to 0.8, and the alpha is more than 0.8 and less than or equal to 1. Each group of Markov chain models form a discrete grid space by a vehicle speed V (0 to 30m/s) and an acceleration a (-1.5 to 1.5m/s2), the vehicle speed is defined as a current state quantity and is divided into p sections, the current state quantity is indexed by i epsilon { 1.. multidot.p }, the vehicle acceleration is defined as an output quantity at the next moment and is divided into q sections, and the current state quantity is indexed by j epsilon { 1.. multidot.q }. The transition probability matrix T for each set of markov chain models can be expressed as:
In the formula: n ∈ { 1., Np } is a target time at which the vehicle speed needs to be predicted in the prediction time domain, and Tij is a probability that the vehicle acceleration evolves to aj at the next time when the vehicle speed Vk + n at the current time is Vi.
In the initial state, selecting a typical stable working condition, calculating to obtain a Markov chain model transition probability matrix according to the following formula,
In the formula: nij is the number of times j occurs at the current time i and the next time j. FIG. 4 shows the Markov chain model transition probability matrix when the driver pedal opening is 0 < α ≦ 0.2.
in real-time operation, the Markov chain model needs online self-correction to adapt to the change of the working condition, at the current time k, if the vehicle speed Vk-1 at the previous time is Vi, and ak is aj at the time, the Markov chain transition probability matrix under the condition is adaptively corrected as follows:
T(k)＝T(k-1)+λ (20)
in the formula: s belongs to { 1.,. q }, s is not equal to j, and lambda is an adaptive coefficient. Equation (20) observes this event occurring at the current time, and corrects the probability of this event in the markov chain transition probability matrix, and equation (21) corrects the probability of other output values in this state when this event occurs. It can be noted that, in the actual adaptive correction process, only one column of probability data in the transition probability matrix at the current time is updated, and other probabilities are kept unchanged.
According to the Markov chain model, the acceleration of the vehicle at the next moment can be predicted at the current moment k, and the vehicle speed at the next moment can be obtained:
Similarly, the vehicle speed at each moment in the prediction time domain can be calculated from the vehicle speed at the previous moment:
in the formula: and P is more than or equal to n and is each target moment in the prediction time domain.
The prediction method based on the Markov chain can effectively predict the future vehicle speed under the stable working condition, but cannot effectively learn the behavior of the driver under the fast-changing working condition, so that the prediction precision is poor. Therefore, aiming at the quick change working condition, the invention applies the radial basis function neural network theory and carries out the prediction of the future vehicle speed by learning the driving behavior of the driver on line.
The radial basis function neural network is a local approximation network, has high convergence rate and small calculation amount compared with other forms of neural networks, and is most suitable for online speed prediction of the hybrid vehicle. Here, the inputs Nin defining the neural network model are the driver pedal information and the vehicle speed over a period of time:
In the formula: hh is the past vehicle speed vector length. The output Nout of the model is the predicted vehicle speed for a future period of time:
N＝V,V,...,V (25)
Neurons in the hidden layer adopt a gaussian function as a radial basis function:
in the formula: yj is the neural network output, ω ij is the output weight, bf is the neuron threshold preset by the developer, x is the neural network input, ci is the neuron node center, σ is the neuron radial basis function diffusion width, and h is the number of hidden layer nodes. Thus, a nonlinear neural network model for vehicle speed prediction can be obtained:
In the formula: fn is a radial basis function neural network map, the structure of which is shown in fig. 5.
If Hh is set to 9, that is, if the historical vehicle speed is 10 vehicle speeds in the past, the radial basis function neural network input amount is 11, and if the predicted vehicle speed is 5 seconds in the future, the neural network output amount is 5, and the number of neurons is equal to the number of input amounts, that is, h is 10. In the running process of a vehicle, the radial basis function neural network model predicts the future vehicle speed according to the historical vehicle speed and the current pedal information of the driver, meanwhile, the current generated vehicle information is used as new historical information, and the self-adaptive online learning of the neural network is realized by a method of selecting a self-organization center and determining a weight value through a pseudo-inverse method.
In order to verify the effectiveness of the energy control method provided by the invention, a simulation test is carried out in a Matlab environment. The sampling time interval of the energy control method is set to be 1 second in the simulation, so that not only can the stable control of the dynamic process of the system be ensured, but also larger control calculated amount can be allowed. Meanwhile, the prediction time domain is set to be P5 seconds, the initial value and the reference value of the battery SOC are both 0.65, and the simulation result is shown in fig. 6.
FIG. 6 shows the result of the determined working condition type for a typical integrated cycle working condition in the simulation online process. As can be seen from the figure, when the vehicle speed changes sharply, such as between 370s to 440s, 980s to 1030s and 1160s to 1220s, the working condition is judged to be a fast-changing working condition; and when the vehicle speed fluctuates in a small range or the vehicle slowly accelerates and decelerates, for example, between 600s to 700s and 1220s to 1900s, the working condition is judged to be a stable working condition, so that the effectiveness of the working condition type judgment method can be demonstrated.
fig. 7 is a visual display of the vehicle speed prediction result, and it can be seen that the vehicle speed prediction method provided by the invention can perform prediction more accurately in most of the time.
in order to further reasonably evaluate the prediction result through data comparison, Root Mean Square Error (RMSE) is introduced as an evaluation index. The RMSE represents the sample precision by calculating the standard deviation of the difference value between the sample value and the true value, is suitable for comparing the predicted value with the true value, and has the following calculation formula:
In the formula: RMSE (k) is a root mean square error value of a k-th sampling point in the cycle working condition in a prediction time domain, RMSE is a root mean square error value of the whole cycle working condition, Nc is the number of sampling points in the whole cycle working condition, and Vc (k + i) is the real vehicle speed of an ith sampling point after the k-th sampling point in the cycle working condition.
under the same cycle working condition, the prediction method participating in comparison comprises the following steps: keeping prediction, namely, keeping the predicted vehicle speed unchanged; markov chain prediction, namely vehicle speed prediction is carried out on the basis of the Markov chain in the whole process; predicting the vehicle speed based on the radial basis neural network in the whole course; and (3) comprehensive prediction, namely the comprehensive vehicle speed prediction method provided by the invention. A comparison of the simulation results is shown in Table 3.
TABLE 3 comparison of results of different prediction methods
as can be seen from the table, the retained prediction RMSE serving as the reference is higher, the prediction precision is poorer, the comprehensive prediction combines the advantages of Markov chain prediction and neural network prediction, the RMSE is minimum, the prediction precision is optimal, and the RMSE is improved by about 31 percent compared with the reference.
fig. 8 shows a comparison of the prediction error of the 5 th future second in the prediction time domain of the comprehensive prediction, the markov chain prediction and the neural network prediction at each sampling time in the partial cycle working condition, and the first column on the left side in the figure is the prediction between the cycle working conditions 980s and 1030s, so that the working condition is judged to be a fast-changing working condition at the moment, and the result of the comprehensive prediction is obviously superior to the markov chain prediction due to the neural network prediction. The second column in the diagram is prediction between the circulating working conditions 600s and 700s, at the moment, the vehicle speed fluctuation is small, the working conditions are judged to be stable working conditions, the Markov chain prediction is adopted for comprehensive prediction, and the result is obviously superior to the neural network prediction.
Fig. 9 shows simulation results of vehicle speed, EVT mode, and battery SOC, and it can be seen from the figure that the actual vehicle speed substantially coincides with the target vehicle speed, and the battery SOC can be maintained to fluctuate around 0.65, which illustrates that the energy control method can well maintain the battery SOC and supplement the energy less supplied by the engine to a certain extent through the consumption of electric energy to better adjust the engine operating point under the condition that the driver's demand is satisfied first.
fig. 10 shows the rotation speeds and torques of the engine, the motor a, and the motor B under this condition, and it can be seen from the figure that the fluctuation of the rotation speed of the engine is small because the road surface is decoupled from the engine and the speed regulation performance of the motors is far better than that of the engine, so that the fluctuation of the vehicle speed is mainly compensated by the rotation speed variation of the two motors. The torque of the electric machine is relatively small, so that engine torque fluctuations caused by road surface changes can only be compensated to a certain extent by the electric machine.
fig. 11 is a distribution diagram of the engine operating points under the operating condition, and it can be seen that the energy control method can better adjust the engine operating points, so that the engine can operate near the optimal fuel economy curve under most conditions, the engine operating efficiency is higher, and the vehicle economy is better.
In order to further verify the improvement of the predictive control-based energy control method on the vehicle performance, simulation test results of various energy control methods under different working conditions are compared, wherein the strategies participating in comparison comprise: predictive control, i.e. the energy control method proposed by the present invention; hold control, i.e., an energy control method assuming that the vehicle speed remains unchanged and employing predictive control; dynamic programming, namely, assuming that the working condition is known, a global optimal solution can be obtained; rules, i.e. rule-based energy control methods, are used as a reference to be lifted. The comparative results are as follows:
fig. 12 shows a comparison of simulation results of engine operating point distribution, and it can be seen that the predicted control and the hold control result similar to the dynamic programming, while the predicted control engine operating point distribution is more concentrated in the high efficiency region than the hold control, while the rule results are worse, and the engine is often operated in the low efficiency region.
Fig. 13 shows a comparison of simulation results of battery SOC, and it can be seen from the figure that the predictive control and the hold control are real-time optimization, and a local optimal solution is obtained, the result of which is similar to the global optimal solution obtained by the dynamic rule, and the result SOC of the rule is too stable, which is very different from the result of the dynamic planning, and the control effect is poor because the engine is not effectively adjusted by the fluctuation of the battery SOC.
Table 4 shows the comparison of simulation results of different energy control methods under different cycle conditions. Because the energy control method based on the predictive control is a real-time optimization, the battery SOC at the end time of the cycle condition cannot be ensured to be the same as the initial time, so for the fair comparison, the consumption of the electric energy needs to be considered at the same time, the battery SOC at the end time is converted according to the energy price to obtain the equivalent fuel consumption,
in the formula, Ec, s, Fc, s and Δ SOCc, s are respectively equivalent fuel consumption, fuel consumption and battery SOC variation values corresponding to the cycle condition and the optimization method, and are conversion factors for converting electric energy into fuel through energy price, the current price of diesel is 5.54 yuan/liter, and the price of electric energy is 0.9 yuan/degree. As can be seen from the table, for the improvement of the vehicle economy, the predictive control is close to the dynamic programming, has larger promotion than the rule and is superior to the maintenance control, and has the promotion of 18 percent at most compared with the benchmark.
TABLE 4 comparison of simulation results of different energy control methods
by analyzing and comparing simulation results, the effectiveness of the vehicle speed prediction method is verified, the accuracy is improved by 31%, meanwhile, the effectiveness of the energy control method based on prediction control is also verified, and the fuel economy is improved by 18% compared with a regular strategy.
It should be understood, however, that the above description is only one embodiment of the present invention, and it should be understood that a person skilled in the art may make several modifications and improvements without departing from the principle of the present invention, and the modifications and improvements are within the protection scope of the appended claims.
Claims (7)
1. a method of controlling energy of a hybrid vehicle, the method comprising:
(1) Observing the current system state, including the vehicle running speed, the driver pedal information, the battery SOC and the engine rotating speed and torque;
(2) judging the current EVT mode state according to the vehicle running speed and the driver pedal information, updating the current system model and system constraints, and assuming that the EVT mode state is kept unchanged in the prediction time domain;
(3) predicting the future vehicle speed in the prediction time domain to obtain the system observation input quantity in the prediction time domain;
(4) Constructing a prediction control optimization problem in a prediction time domain, and performing numerical solution on line through a dynamic programming algorithm;
(5) Calculating to obtain an optimal control sequence in a prediction time domain;
(6) only the first group of optimal control quantity is adopted, the optimal control quantity acts on the system at the current sampling moment, and the rest control quantity is abandoned;
(7) And (5) repeating the steps (1) to (6) at the next moment until the driving is finished.
2. The energy control method of claim 1, wherein the future vehicle speed is predicted according to the following classification prediction method:
Classifying the running working condition of the vehicle into a stable working condition and a fast-changing working condition under an off-line state by utilizing a K-means clustering algorithm, and judging the type of the current working condition of the vehicle in real time at an on-line stage;
And aiming at the stable working condition, a Markov chain-based vehicle speed prediction method is adopted, and aiming at the fast-changing working condition, a radial basis neural network-based vehicle speed prediction method is adopted, so that the advantages of the two methods are comprehensively utilized to achieve the optimal prediction effect.
3. The energy control method of claim 2, wherein the following characteristic parameters are used: the vehicle running condition is classified by maximum acceleration, maximum deceleration, average acceleration, vehicle speed standard deviation, difference between maximum vehicle speed and minimum vehicle speed and acceleration standard deviation.
4. the energy control method according to claim 2, wherein a K-means clustering algorithm is adopted to classify data by calculating the degree of affinity and sparseness among samples, and the specific working condition judgment steps are as follows:
an off-line stage: (1) combining a plurality of standard cycle conditions to form a sample;
(2) Calculating the working condition characteristic parameters of 10 seconds in the past at each sampling moment in the cycle working condition to obtain characteristic parameter sample data [ x11, x 12.,. x1m ], [ x21, x 22.,. x2m ], … …, [ xn1, xn 2.,. xnm ], wherein m is the characteristic parameter ordinal number, and n is the cycle working condition length;
(3) applying a K-means clustering algorithm, randomly selecting a clustering center c1 [ [ c11, c 12., c1m ], c2 [ [ c21, c 22., c2m ], calculating distances between all samples and the clustering center, grouping the samples according to a nearest neighbor rule, and attributing different theta m (K) clustering domains, wherein K is iteration times, nm is cycle working condition length contained in the mth class characteristic parameter, xi is a characteristic parameter sample value, and then adjusting the clustering center according to the following formula:
If cm (k +1) is not equal to cm (k), continuing to adjust the clustering center until the clustering center does not change any more, and considering that the classification is stable to obtain the clustering center c1 under a stable working condition and the clustering center c2 under a quick change working condition;
An online stage: (1) in the actual running process of the vehicle, working condition characteristic parameter values [ x1, x 1.., xm ] of 10 seconds in the past are calculated at the current sampling moment;
(2) The distance d of the characteristic parameter values [ x1, x 1.., xm ] to the two cluster centers c1 and c2 is calculated according to the following formula:
In the formula: j is 1,2 corresponds to two working conditions;
(3) And if d1 is not more than d2, judging that the current time is a stable working condition, and if d1 is more than d2, judging that the current time is a quick-change working condition.
5. the energy control method of claim 2, wherein the acceleration of the vehicle at each moment is assumed to be independent of historical information and is determined only by current information under the stationary condition, so that the acceleration change of the vehicle is considered to be a Markov process, and a Markov chain model is used to simulate the change rule of the vehicle speed and the acceleration and predict the future vehicle speed under the stationary condition.
6. The energy control method of claim 2, wherein the prediction of future vehicle speed is made through online learning of driver driving behavior for fast-changing conditions using radial basis neural network theory.
7. the energy control method according to claim 2, wherein the radial basis function neural network model predicts the future vehicle speed according to the historical vehicle speed and the current pedal information of the driver, and meanwhile, the current generated vehicle information is used as new historical information, and the adaptive online learning of the neural network is realized through a method of selecting a self-organization center and determining a weight value by a pseudo-inverse method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710645889.2A CN107688343B (en) | 2017-08-01 | 2017-08-01 | Energy control method of hybrid power vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710645889.2A CN107688343B (en) | 2017-08-01 | 2017-08-01 | Energy control method of hybrid power vehicle |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107688343A CN107688343A (en) | 2018-02-13 |
| CN107688343B true CN107688343B (en) | 2019-12-06 |
Family
ID=61153177
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710645889.2A Active CN107688343B (en) | 2017-08-01 | 2017-08-01 | Energy control method of hybrid power vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107688343B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510129B (en) * | 2018-04-12 | 2021-12-21 | 中南大学 | Intelligent prediction method and device for real-time electric quantity of unmanned vehicle |
| CN108510737B (en) * | 2018-04-12 | 2020-04-10 | 中南大学 | Unmanned vehicle power supply real-time monitoring method and device integrating wind environment |
| CN108572648B (en) * | 2018-04-24 | 2020-08-25 | 中南大学 | Unmanned vehicle power supply multi-source fusion prediction method and system |
| CN109733208B (en) * | 2019-01-24 | 2022-04-12 | 湖南普西智能科技有限责任公司 | Acceleration control method and system for pure electric vehicle |
| CN109910863B (en) * | 2019-04-17 | 2020-07-31 | 苏州赫行新能源汽车科技有限公司 | Control method based on special transmission for plug-in hybrid electric vehicle |
| CN110161854B (en) * | 2019-05-21 | 2021-06-29 | 吉林大学 | Method for controlling longitudinal driving of highway heavy trucks in formation |
| CN110991757B (en) * | 2019-12-10 | 2022-01-28 | 北京理工大学 | Comprehensive prediction energy management method for hybrid electric vehicle |
| CN111198501B (en) * | 2020-01-14 | 2022-05-13 | 浙江工业大学 | Method for determining fuel equivalent factor by RBF neural network |
| CN111267827B (en) * | 2020-02-13 | 2021-07-16 | 山东中科先进技术研究院有限公司 | Energy management method and system for hybrid electric vehicle |
| CN111597640A (en) * | 2020-05-22 | 2020-08-28 | 上海海事大学 | Method for predicting demand load of hybrid power ship under condition classification |
| CN112298155B (en) * | 2020-10-30 | 2022-04-08 | 江苏紫琅汽车集团股份有限公司 | Method for predicting energy consumption of hybrid power truck based on variable time domain model |
| CN112785845B (en) * | 2020-12-30 | 2022-11-01 | 桂林电子科技大学 | Vehicle speed prediction method based on K-means clustering and RBF neural network |
| CN113276829B (en) * | 2021-07-09 | 2022-11-01 | 吉林大学 | Vehicle running energy-saving optimization weight-changing method based on working condition prediction |
| CN113479187B (en) * | 2021-07-30 | 2023-03-17 | 南京航空航天大学 | Layered different-step-length energy management method for plug-in hybrid electric vehicle |
| CN113548035B (en) * | 2021-08-23 | 2024-03-12 | 北京理工大学 | Control method and device for vehicle power system |
| CN113655385B (en) * | 2021-10-19 | 2022-02-08 | 深圳市德兰明海科技有限公司 | Lithium battery SOC estimation method and device and computer readable storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102019926A (en) * | 2009-09-16 | 2011-04-20 | 通用汽车环球科技运作公司 | Predictive energy management control scheme for a vehicle including a hybrid powertrain system |
| CN102729991A (en) * | 2012-06-20 | 2012-10-17 | 浙江大学 | Hybrid bus energy distribution method |
| CN202703576U (en) * | 2012-06-14 | 2013-01-30 | 北京理工大学 | Hybrid vehicle energy management system |
| CN103914985A (en) * | 2014-04-25 | 2014-07-09 | 大连理工大学 | Method for predicting future speed trajectory of hybrid power bus |
| CN104627185A (en) * | 2013-11-08 | 2015-05-20 | 三菱电机株式会社 | Vehicle energy management device |
| CN104925063A (en) * | 2015-04-12 | 2015-09-23 | 北京理工大学 | Model predictive control method for electromechanical composite transmission vehicle |
| CN105416077A (en) * | 2015-12-01 | 2016-03-23 | 北京理工大学 | Energy management system and method for electric automobile |
| JP2016055759A (en) * | 2014-09-09 | 2016-04-21 | トヨタ自動車株式会社 | Hybrid electric vehicle control unit |
| CN106427987A (en) * | 2015-08-04 | 2017-02-22 | 现代自动车株式会社 | System and method for controlling hybrid vehicle |
| CN106494388A (en) * | 2016-09-28 | 2017-03-15 | 中国科学院深圳先进技术研究院 | A kind of hybrid vehicle energy management and speed-regulating device and method |
-
2017
- 2017-08-01 CN CN201710645889.2A patent/CN107688343B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102019926A (en) * | 2009-09-16 | 2011-04-20 | 通用汽车环球科技运作公司 | Predictive energy management control scheme for a vehicle including a hybrid powertrain system |
| CN202703576U (en) * | 2012-06-14 | 2013-01-30 | 北京理工大学 | Hybrid vehicle energy management system |
| CN102729991A (en) * | 2012-06-20 | 2012-10-17 | 浙江大学 | Hybrid bus energy distribution method |
| CN104627185A (en) * | 2013-11-08 | 2015-05-20 | 三菱电机株式会社 | Vehicle energy management device |
| CN103914985A (en) * | 2014-04-25 | 2014-07-09 | 大连理工大学 | Method for predicting future speed trajectory of hybrid power bus |
| JP2016055759A (en) * | 2014-09-09 | 2016-04-21 | トヨタ自動車株式会社 | Hybrid electric vehicle control unit |
| CN104925063A (en) * | 2015-04-12 | 2015-09-23 | 北京理工大学 | Model predictive control method for electromechanical composite transmission vehicle |
| CN106427987A (en) * | 2015-08-04 | 2017-02-22 | 现代自动车株式会社 | System and method for controlling hybrid vehicle |
| CN105416077A (en) * | 2015-12-01 | 2016-03-23 | 北京理工大学 | Energy management system and method for electric automobile |
| CN106494388A (en) * | 2016-09-28 | 2017-03-15 | 中国科学院深圳先进技术研究院 | A kind of hybrid vehicle energy management and speed-regulating device and method |
Non-Patent Citations (1)
| Title |
|---|
| 基于随机动态规划的混联式混合动力客车能量管理策略;林歆悠 等;《汽车工程》;20121231;第34卷(第9期);第830-836页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107688343A (en) | 2018-02-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107688343B (en) | Energy control method of hybrid power vehicle | |
| WO2021103625A1 (en) | Short-term vehicle speed condition real-time prediction method based on interaction between vehicle ahead and current vehicle | |
| Liu et al. | Online energy management for multimode plug-in hybrid electric vehicles | |
| Huang et al. | Model predictive control power management strategies for HEVs: A review | |
| CN112776673B (en) | Intelligent network fuel cell automobile real-time energy optimization management system | |
| Zhang et al. | A deep reinforcement learning-based energy management framework with Lagrangian relaxation for plug-in hybrid electric vehicle | |
| CN111267831A (en) | Hybrid vehicle intelligent time-domain-variable model prediction energy management method | |
| CN112668799A (en) | Intelligent energy management method and storage medium for PHEV (Power electric vehicle) based on big driving data | |
| Lin et al. | Blending gear shift strategy design and comparison study for a battery electric city bus with AMT | |
| CN107516107A (en) | A kind of driving cycle classification Forecasting Methodology of motor vehicle driven by mixed power | |
| CN110406526A (en) | Parallel hybrid electric energy management method based on adaptive Dynamic Programming | |
| CN113479186B (en) | Energy management strategy optimization method for hybrid electric vehicle | |
| Zhu et al. | A comprehensive review of energy management strategies for hybrid electric vehicles | |
| CN113246958B (en) | TD 3-based multi-target HEV energy management method and system | |
| Yang et al. | Energy management of the power-split hybrid electric city bus based on the stochastic model predictive control | |
| Montazeri-Gh et al. | Driving condition recognition for genetic-fuzzy HEV control | |
| CN117408360A (en) | Method, device and equipment for generating vehicle management strategy based on cloud domain interaction | |
| CN115805840A (en) | Energy consumption control method and system for range-extending type electric loader | |
| CN115257695A (en) | Energy management control system and method for hybrid electric vehicle | |
| Kabir | Vehicle speed prediction based on road status using machine learning | |
| Zhang et al. | Integrated velocity optimization and energy management strategy for hybrid electric vehicle platoon: A multi-agent reinforcement learning approach | |
| Xia et al. | A predictive energy management strategy for multi-mode plug-in hybrid electric vehicle based on long short-term memory neural network | |
| CN111198501A (en) | Method for determining fuel equivalent factor by RBF neural network | |
| Yang et al. | A real time multi-objective optimization Guided-MPC strategy for power-split hybrid electric bus based on velocity prediction | |
| Zhang et al. | Deep reinforcement learning based multi-objective energy management strategy for a plug-in hybrid electric bus considering driving style recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |