CN110406526A - Parallel hybrid electric energy management method based on adaptive Dynamic Programming - Google Patents
Parallel hybrid electric energy management method based on adaptive Dynamic Programming Download PDFInfo
- Publication number
- CN110406526A CN110406526A CN201910717298.0A CN201910717298A CN110406526A CN 110406526 A CN110406526 A CN 110406526A CN 201910717298 A CN201910717298 A CN 201910717298A CN 110406526 A CN110406526 A CN 110406526A
- Authority
- CN
- China
- Prior art keywords
- network
- dynamic programming
- energy management
- adaptive dynamic
- hybrid electric
- 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.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W10/00—Conjoint control of vehicle sub-units of different type or different function
- B60W10/04—Conjoint control of vehicle sub-units of different type or different function including control of propulsion units
- B60W10/06—Conjoint control of vehicle sub-units of different type or different function including control of propulsion units including control of combustion engines
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W10/00—Conjoint control of vehicle sub-units of different type or different function
- B60W10/04—Conjoint control of vehicle sub-units of different type or different function including control of propulsion units
- B60W10/08—Conjoint control of vehicle sub-units of different type or different function including control of propulsion units including control of electric propulsion units, e.g. motors or generators
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W20/00—Control systems specially adapted for hybrid vehicles
- B60W20/10—Controlling the power contribution of each of the prime movers to meet required power demand
- B60W20/15—Control strategies specially adapted for achieving a particular effect
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2510/00—Input parameters relating to a particular sub-units
- B60W2510/06—Combustion engines, Gas turbines
- B60W2510/0657—Engine torque
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2510/00—Input parameters relating to a particular sub-units
- B60W2510/24—Energy storage means
- B60W2510/242—Energy storage means for electrical energy
- B60W2510/244—Charge state
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2540/00—Input parameters relating to occupants
- B60W2540/10—Accelerator pedal position
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2540/00—Input parameters relating to occupants
- B60W2540/12—Brake pedal position
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2710/00—Output or target parameters relating to a particular sub-units
- B60W2710/06—Combustion engines, Gas turbines
- B60W2710/0666—Engine torque
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2710/00—Output or target parameters relating to a particular sub-units
- B60W2710/08—Electric propulsion units
- B60W2710/083—Torque
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/62—Hybrid vehicles
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02T90/10—Technologies relating to charging of electric vehicles
- Y02T90/14—Plug-in electric vehicles
Abstract
The invention discloses a kind of parallel hybrid electric energy management methods based on adaptive Dynamic Programming, it is intended to carry out energy management optimization to the combination drive mode in several operating modes of PHEV, include: that the total demand torque of vehicle is obtained according to the order of driver and speed, this moment battery SOC and motor torque are obtained by vehicle's current condition;The comprehensive information of vehicles obtained, establishes the parallel hybrid electric energy management model based on ADP;Execution network in ADP is used to near-optimization control strategy, it evaluates network and is used to near-optimization performance index function, the combination of the two networks is equivalent to the dynamic change that an intelligent body is capable of intelligence learning and online response system, it is automatically adjusted it to the parameter in network structure, the energy management model of foundation is solved, obtains optimal distribution of torque to motor.The method of the present invention can in real time by demand torque reasonably, be optimally assigned to engine and motor.
Description
Technical field
The invention belongs to parallel hybrid electric energy management technical fields, relate to a kind of based on adaptive dynamic rule
The parallel hybrid electric energy management method drawn.
Background technique
Energy management strategies are the key that realize that vehicle energy demand is distributed between engine and motor, the warp of PHEV
Ji property, dynamic property and used energy management strategies are closely related.Energy management strategies are oil burning power system and electric drive
System realizes the tie of good combination, in order to improve energy efficiency to the maximum extent and reduce exhaust emission, needs efficient energy
It measures management strategy and realizes distribution of the demand power between engine and motor.The core of energy management sought to solve to the phase
Hope the control that energy is converted between target and the vehicle performance of acquisition.Desired performance indicator multilist reduces fuel consumption now, subtracts
Few noxious gas emission, increase comfort and extension battery life etc. are many-sided.The purpose of control strategy is exactly to improve fuel oil
Economy, energy-saving and emission-reduction and guarantee that system has preferable performance indicator.Control strategy determine what moment, which type of
Under load, how internal combustion engine and motor will be used.
The drive system of PHEV in such a way that engine and motor are in parallel, have engine and driving motor directly to
The ability of driving wheel offer mechanical output.Motor had not only been able to achieve the function of motor but also had been able to achieve the function of generator.PHEV two
Item driving path parallel connection increases driving power to enhance the dynamic property of hybrid vehicle, parallel hybrid electric from
Engine is into the power transmission process between wheel, converted without mechanical energy-electric energy-mechanic energy in addition to frictional dissipation
Journey, energy conversion efficiency are high.
Summary of the invention
(1) technical problems to be solved
The object of the present invention is to provide one kind to be based on adaptive Dynamic Programming (Adaptive Dynamic
Programming, ADP) parallel hybrid electric (Parallel Hybrid Electrical Vehicle, PHEV) energy
Quantity management method uses a kind of novel intelligent algorithm --- adaptive Dynamic Programming, and it is applied to parallel hybrid power
In the energy management control of automobile, PHEV can be promoted while the smooth change and efficient region for maintaining battery SOC work
Fuel economy.
(2) technical solution
Parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterised in that including walking as follows
It is rapid:
Step 1, ordering as the stroke and speed of accelerator pedal or brake pedal obtain vehicle always according to driver
Demand torque T obtains this moment battery charge state SOC and motor torque T by vehicle's current conditione;
Step 2, be based on adaptive dynamic programming method, with maintain battery SOC smooth change and efficient region work
The energy management model that the fuel economy of PHEV establishes parallel hybrid electric for target is promoted simultaneously;
Step 3 is learnt by executing network in adaptive dynamic programming method and evaluating the intelligent online of network come to energy
Amount administrative model is solved, and is obtained the optimal demand torque for distributing to motor, is then divided further according to total demand torque
The optimal demand torque of dispensing engine;Detailed process is as follows:
Step 3.1, initialization evaluation network and the weight for executing network;
Step 3.2, the battery pack SOC that each sampling instant is acquired, vehicle demand torque T and engine current time turn
Square TeInput executes network, and executing network output is motor demand torque Tm_req;
Step 3.3, the battery pack SOC that each sampling instant is acquired, vehicle demand torque T, engine current time turn
Square TeWith the motor demand torque T for executing network outputm_reqAs the input of evaluation network, the approximation of cost function J is obtained
Step 3.4, according to evaluation network right value update mode, the more weight of New Appraisement network, make its exportWith generation
The error E of valence function JcConstantly approach 0;
Step 3.5, according to execute network right value update mode, update execute network weight;
Step 3.6 updates and records above-mentioned execution network and evaluates the weight of network;
Circulation step 3.1~3.6, untilIt is approximately equal to cost function J, optimum control is completed and exports Tm_req。
The parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterised in that: step
In rapid 3, the cost function J is exactly the purpose function for finding optimal solution, is the pass of adaptive dynamic programming method operation
Key, is defined as:
In formula, γ is discount factor, and 0 < γ≤1;U is utility function;The purpose of adaptive Dynamic Programming is exactly to select
One control sequence u (i), i=k, k+1 ... so that the cost function J of definition is minimized;
Wherein, the quadratic form utility function U defined in conjunction with SOC is as follows:
U (k)=x (k) Ax (k)T+ε(SOC-τ)2;
In formula, A is the unit matrix for meeting this formula matrix operation;X is the state variable of system input;ε be discount because
Son;τ is to fluctuate lower limit according to the SOC that vehicle is chosen.
The parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterised in that: institute
The optimization object function of commentary valence network are as follows:
Wherein, ec(k) the prediction error of evaluation network is indicated;
Evaluate the output valve of networkIt can be by being realized with time minimum error below:
When having E for all kc(k)=0 when, above formula means
Obviously have,Therefore, error function defined in minimizing, will obtain a trained nerve
Network, the output valve of the evaluation networkIt is an estimation of the cost function J of definition.
The parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterized in that, institute
The right value update of commentary valence network is carried out using gradient descent method, is made:
Wc(k+1)=Wc(k)+ΔWc(k),
Wherein,
In formula, k indicates sampling period, WcRepresent the weight of evaluation network, Ec(k) the optimization aim letter of evaluation network is indicated
Number, lc(k) learning rate of evaluation network is indicated.
The parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterized in that, institute
Stating and executing the right value update mode of network is by using control signal u (k), to minimizeFor target;Specific steps are as follows:
Battery pack SOC, vehicle demand torque T and the engine current time torque T that each sampling instant is acquiredeInput
Network is executed, executing network output is motor demand torque Tm_req, and execute the right value update model of network are as follows:
Wa(k+1)=Wa(k)+ΔWa(k),
Wherein,
In formula, WaIndicate that the weight of execution network, u (k) indicate control variable, la(k) learning rate of execution network is indicated.
The parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterized in that: institute
Stating execution network and evaluating network to be all made of BP neural network is also three_layer planar waveguide, and the training of network is by positive meter
It calculates and reversed error propagation process forms.
(3) beneficial effect
Parallel hybrid electric energy management control method provided by the invention based on adaptive Dynamic Programming, overcomes
The control strategy based on optimization such as the rule-based control strategies such as electric auxiliary, logic threshold and conventional Dynamic Programming (DP)
Deficiency is worked as by the method for adaptive Dynamic Programming using vehicle total demand torque T, battery charge state SOC and engine
Preceding torque TeFor input, the fuel-economy of PHEV is promoted while work with the smooth change and efficient region that maintain battery SOC
Property for target establish the energy management model of parallel hybrid electric, wherein executing network and evaluation in adaptive Dynamic Programming
The combination of network is equivalent to the dynamic change that an intelligent body (Agent) is capable of intelligence learning and online response system, keeps its right
Parameter in network structure is automatically adjusted, and is solved, is obtained optimal to energy management model established by the present invention
Distribution of torque guarantees that PHEV can high-efficiency operation in different driving cycles to motor.
Detailed description of the invention
Fig. 1 is adaptive Dynamic Programming schematic diagram of the present invention.
Fig. 2 is the parallel hybrid electric energy management control strategy of the present invention based on adaptive Dynamic Programming
Structural schematic diagram.
Fig. 3 is the structural scheme of mechanism that network is evaluated in adaptive Dynamic Programming of the present invention.
Fig. 4 is the structural scheme of mechanism that network is executed in adaptive Dynamic Programming of the present invention.
Fig. 5 is adaptive dynamic programming algorithm flow chart (ADHDP) of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention do into
The detailed description of one step.
The thought of adaptive Dynamic Programming (ADP) is using approximation to function structure, to approach the property in Dynamic Programming Equation
Energy target function and control strategy, to meet the principle of optimization.To obtain optimum control and optimal performance index function.It is calculated
The leading of method is exactly selection to controlling network, evaluating network and prototype network, minimizes cost by the iteration of network weight
Function.It embodies an important development direction of artificial intelligence and control field, is to dynamic programming algorithm and intensified learning
The synthesis of algorithm, the advantage with both the above algorithm, not only with the optimizing characteristic of Dynamic Programming, but with intensified learning from
Learning characteristic.As shown in Figure 1, ADP is mainly consisted of three parts: dynamical system, execution (Action) function and evaluation
(Critic) function.Each part can be replaced by neural network, and the present invention uses BP neural network, and BP neural network is as three
Layer feedforward neural network by the training to sample come the feature of Learning Studies object, so that it be made to possess the energy of association and prediction
Power.Wherein dynamical system can be modeled by neural network, executed network and be used to near-optimization control strategy, evaluate network
For near-optimization performance index function.The combination of the two is equivalent to an intelligent body (Agent), control/execution afterwards
(Action) reward/punishment (Reward/Penalty) that dynamical system (or controlled device) is generated in different phase is acted on
To influence evaluation function.Approximation to function structure or neural network are recycled, realizes and execution function and evaluation function is approached,
But executing function is to carry out on the basis of estimation of score function, that is, evaluation function must be made minimum.Evaluation function
Parameter update be is carried out based on Bell's graceful optimum principle, can not only reduce the forward calculation time in this way, can also online
The dynamic change for responding unknown system, is automatically adjusted it to certain parameters in network structure.
In conjunction with Fig. 2~Fig. 5, the parallel hybrid electric energy management provided by the invention based on adaptive Dynamic Programming
Control method includes the following steps:
Step 1: total according to the order (stroke of accelerator pedal or brake pedal) of driver and speed acquisition vehicle
Demand torque T obtains this moment battery charge state SOC and motor torque T by vehicle's current conditione;
Step 2: be based on adaptive dynamic programming method, with maintain battery SOC smooth change and efficient region work
The energy management model that the fuel economy of PHEV establishes parallel hybrid electric for target is promoted simultaneously.
In the present embodiment, adaptive dynamic programming method contains BP neural network, dynamic programming algorithm and extensive chemical
The theory of habit derives from time forward dynamic programming method, is closed by the iteration of optimal policy, cost or their derivative
System constantly generates the general solution of forward Dynamic Programming, and target is to overcome " dimension calamity ", and guarantee to converge to an approximation at any time
Optimal solution.ADP has the advantage that
(1) adaptive Dynamic Programming is not relying on an accurate controlled device mathematical model sometimes, and controller can
With online " study " control.
(2) adaptive Dynamic Programming can avoid " dimension calamity " problem of Dynamic Programming (DP) by Approach by inchmeal.
(3) adaptive Dynamic Programming does not need precise definition system performance index.
(4) adaptive Dynamic Programming is to solve Control of Nonlinear Systems to open up a new way.
The present invention is using one of adaptive Dynamic Programming classification --- and it executes and relies on heuristic dynamic programming (Action
Dependent Heuristic Dynamic Programming, ADHDP), be heuristic dynamic programming (HDP) execution according to
Rely form.ADHDP does not need prototype network, and only comprising executing network and evaluation network, the two networks are according to real system
It is required that can choose suitable neural network structure, the present invention selects BP (Back Propagation) neural network.Wherein, it holds
The input of row network is the state variable x of system, and output is the current control variable u (k) of system, it will generate one it is optimal or
The control sequence u (i) of person's suboptimum, i=k, k+1 ..., so that defined performance index function J (i.e. cost) is minimized, cost
Function is exactly the purpose function for finding optimal solution, this is also the effect of cost function, it is adaptive dynamic programming method
The key of operation defines the quality of methods and results in a sense.State variable and control variable are all commented in ADHDP
The input of valence network, output are the approximations to cost function。
Step 3: being learnt by executing network in adaptive dynamic programming method and evaluating the intelligent online of network come to energy
Amount administrative model is solved, and is obtained the optimal demand torque for distributing to motor, is then divided further according to total demand torque
The optimal demand torque of dispensing engine.Detailed process is as follows:
(1) weight of initialization evaluation network and execution network;
(2) battery pack SOC, vehicle demand torque T and the engine current time torque T acquired each sampling instante
Input executes network, and executing network output is motor demand torque Tm_req;
(3) battery pack SOC, the vehicle demand torque T, engine current time torque T acquired each sampling instanteWith
Execute the motor demand torque T of network outputm_reqAs the input of evaluation network, the approximation of cost function is obtained
(4) according to the right value update mode of evaluation network, the more weight of New Appraisement network exports itIt is approximately equal to
Cost function J.
(5) according to the right value update mode for executing network, the weight for executing network is updated;
(6) it updates and records above-mentioned execution network and evaluate the weight of network;
(2)~(6) are recycled, untilIt is approximately equal to cost function J, optimum control is completed and exports Tm_req。
In step 3, evaluation network and the intelligent online for executing network learn, and specific training process is as follows:
1) the on-line training process of network is evaluated
It evaluates network and uses three layers of BP neural network, 4 input neurons, 25 hidden layer neurons and 1 output mind
Through member.The hidden layer for evaluating network uses bipolarity sigmoidal function, and output layer uses linear function purelin.Evaluate net
The training of network consists of two parts, and one is positive calculating process, the other is the error of more New Appraisement network weight matrix
Back-propagation process.
The input vector in definition evaluation network k stage is inputC (k), does not bring specific value into herein for convenience of description.
InputC (k)=[u1(k),…,um(k),x1(k),…,xn(k)]
Evaluation network positive calculating process be
In formula, ch1jIt (k) is the input for evaluating network concealed j-th of node of layer;ch2jIt (k) is the network concealed layer jth of evaluation
The output of a node.
The training for evaluating network uses gradient descent method, is realized by minimizing the error that following formula defines.
The right value update process derivation for evaluating network is as follows.
①Wc2(weight matrix of hidden layer to output layer).
Wc2(k+1)=Wc2(k)+ΔWc2(k)
②Wc1(weight matrix of input layer to hidden layer).
Wc1(k+1)=Wc1(k)+ΔWc1(k)
2) the on-line training process of network is executed
The BP neural network that network equally uses three layers is executed, hidden layer neuron quantity is 20, and neuron uses
Sigmoidal function calculates, and output layer uses purelin linear function.The input for executing network is that three state variables include
Battery SOC, vehicle demand torque T and engine current torque Te.Network is executed to play a significant role ADP algorithm, it is each
The control law of step provides the calculating for being decided by execute network.
The training for executing network is still made of positive calculating and reversed error propagation process.It is same to describe for convenience
Do not bring specific value into, positive calculating process is
In formula, ah1jIt (k) is the input for executing network concealed j-th of node of layer;ah2jIt (k) is to execute network concealed layer jth
The output of a node.
The training of network is executed to minimizeFor target.The training for executing network still uses gradient descent method.
The right value update process derivation for executing network is as follows.
①Wa2(weight matrix of hidden layer to output layer)
Wa2(k+1)=Wa2(k)+ΔWa2(k)
(2. the weight matrix of input layer to hidden layer)
In formula, Wa2j:=Wa2(j,:),Wa2(j :) it is common matrix representation forms in MATLAB, representing matrix Wa2's
Jth row.
Wa1(k+1)=Wa1(k)+ΔWa1(k)
3) selection of network of relation parameter
In the implementation procedure of ADHDP, relevant parameter mainly includes the hidden layer number of nodes evaluated network and execute network
And learning rate and discount factor.Specific method determines the number of hiding node layer currently not yet, in neural network
In, the number of hidden layer neuron is generally possible to reflect to nonlinear mapping ability, theoretically, the quantity of the node of hidden layer
Bigger, nonlinear mapping ability is better.But with becoming larger for neuronal quantity, learning rate can decline;Neuron number is few,
It is corresponding to approach function and die down.So the selection of hidden layer neuron number is needed through depending on overtesting virtual condition.This
It is 20 that invention, which is chosen and executes the hidden layer neuron number of network, and the hidden layer neuron number for evaluating network is 25.
Learning rate is one and is greater than 0 number less than 1, and learning rate is bigger, and pace of learning is faster, but excessive learning rate
It can cause to vibrate, too small learning rate makes pace of learning too slow again, causes the training time too long.Therefore, learning rate usually with
The time be reduced to a lesser value from a biggish initial value, to accelerate pace of learning and avoid vibrating.Discount factor
Usually one positive number no more than 1 also determines the value without specific method now, can only be by test result come really
It is fixed.In general, the easier success of discount factor more small test, and the bigger control effect of discount factor is more preferably.
4) algorithm flow chart
Fig. 5 show the ADHDP algorithm flow chart that uses of the present invention, i.e. while instructing using based on " parallel " training method
Practice and executes network and evaluation network to control system progress Strategies Training.
Claims (6)
1. the parallel hybrid electric energy management method based on adaptive Dynamic Programming, it is characterised in that including walking as follows
It is rapid:
Step 1 obtains the total demand of vehicle according to the order such as stroke and speed of accelerator pedal or brake pedal of driver
Torque T obtains this moment battery charge state SOC and motor torque T by vehicle's current conditione;
Step 2 is based on adaptive dynamic programming method, while the smooth change to maintain battery SOC and efficient region work
Promote the energy management model that the fuel economy of PHEV establishes parallel hybrid electric for target;
Step 3 is learnt by executing network in adaptive dynamic programming method and evaluating the intelligent online of network come to energy pipe
Reason model is solved, and is obtained the optimal demand torque for distributing to motor, is then distributed to further according to total demand torque
The optimal demand torque of engine;Detailed process is as follows:
Step 3.1, initialization evaluation network and the weight for executing network;
Step 3.2, the battery pack SOC that each sampling instant is acquired, vehicle demand torque T and engine current time torque Te
Input executes network, and executing network output is motor demand torque Tm_req;
Step 3.3, the battery pack SOC that each sampling instant is acquired, vehicle demand torque T, engine current time torque TeWith
Execute the motor demand torque T of network outputm_reqAs the input of evaluation network, the approximation of cost function J is obtained
Step 3.4, according to evaluation network right value update mode, the more weight of New Appraisement network, make its exportWith cost letter
The error E of number JcConstantly approach 0;
Step 3.5, according to execute network right value update mode, update execute network weight;
Step 3.6 updates and records above-mentioned execution network and evaluates the weight of network;
Circulation step 3.1~3.6, untilIt is approximately equal to cost function J, optimum control is completed and exports Tm_req。
2. the parallel hybrid electric energy management method according to claim 1 based on adaptive Dynamic Programming,
Be characterized in that: in step 3, it is adaptive Dynamic Programming side that the cost function J, which is exactly the purpose function for finding optimal solution,
The key of method operation, is defined as:
In formula, γ is discount factor, and 0 < γ≤1;U is utility function;The purpose of adaptive Dynamic Programming is exactly to select one
Control sequence u (i), i=k, k+1 ... so that the cost function J of definition is minimized;
Wherein, the quadratic form utility function U defined in conjunction with SOC is as follows:
U (k)=x (k) Ax (k)T+ε(SOC-τ)2;
In formula, A is the unit matrix for meeting this formula matrix operation;X is the state variable of system input;ε is discount factor;τ
It is that lower limit is fluctuated according to the SOC that vehicle is chosen.
3. the parallel hybrid electric energy management method according to claim 2 based on adaptive Dynamic Programming,
It is characterized in that: the optimization object function of the evaluation network are as follows:
Wherein, ec(k) the prediction error of evaluation network is indicated;
Evaluate the output valve of networkIt can be by being realized with time minimum error below:
When having E for all kc(k)=0 when, above formula means
Obviously have,Therefore, error function defined in minimizing, will obtain a trained nerve net
Network, the output valve of the evaluation networkIt is an estimation of the cost function J of definition.
4. the parallel hybrid electric energy management method according to claim 3 based on adaptive Dynamic Programming,
Feature is that the right value update of the evaluation network is carried out using gradient descent method, makes:
Wc(k+1)=Wc(k)+ΔWc(k),
Wherein,
In formula, k indicates sampling period, WcRepresent the weight of evaluation network, Ec(k) optimization object function of evaluation network, l are indicatedc
(k) learning rate of evaluation network is indicated.
5. the parallel hybrid electric energy management method according to claim 4 based on adaptive Dynamic Programming,
Feature is that the right value update mode for executing network is by using control signal u (k), to minimizeFor target;
Specific steps are as follows:
Battery pack SOC, vehicle demand torque T and the engine current time torque T that each sampling instant is acquiredeInput executes
Network, executing network output is motor demand torque Tm_req, and execute the right value update model of network are as follows:
Wa(k+1)=Wa(k)+ΔWa(k),
Wherein,
In formula, WaIndicate that the weight of execution network, u (k) indicate control variable, la(k) learning rate of execution network is indicated.
6. the parallel hybrid electric energy management method according to claim 5 based on adaptive Dynamic Programming,
Feature is: it is also three_layer planar waveguide, the instruction of network that the execution network, which is all made of BP neural network with evaluation network,
Practice and is made of positive calculating and reversed error propagation process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910717298.0A CN110406526A (en) | 2019-08-05 | 2019-08-05 | Parallel hybrid electric energy management method based on adaptive Dynamic Programming |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910717298.0A CN110406526A (en) | 2019-08-05 | 2019-08-05 | Parallel hybrid electric energy management method based on adaptive Dynamic Programming |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110406526A true CN110406526A (en) | 2019-11-05 |
Family
ID=68365797
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910717298.0A Pending CN110406526A (en) | 2019-08-05 | 2019-08-05 | Parallel hybrid electric energy management method based on adaptive Dynamic Programming |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110406526A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110888323A (en) * | 2019-11-26 | 2020-03-17 | 大连理工大学 | Control method for intelligent optimization of switching system |
CN111367172A (en) * | 2020-02-28 | 2020-07-03 | 华南理工大学 | Hybrid system energy management strategy based on reverse deep reinforcement learning |
CN112217198A (en) * | 2020-09-01 | 2021-01-12 | 广西大学 | Photovoltaic power generation management method for multi-step graph neural network self-adaptive dynamic planning |
CN112896161A (en) * | 2021-02-08 | 2021-06-04 | 杭州电子科技大学 | Electric automobile ecological self-adaptation cruise control system based on reinforcement learning |
CN113232651A (en) * | 2021-02-08 | 2021-08-10 | 杭州电子科技大学 | Fuel automobile ecological self-adaptive cruise control method based on ADHDP |
CN113255954A (en) * | 2020-12-03 | 2021-08-13 | 广西大学 | Charging space-time prediction method of hybrid isomorphic depth map neural network |
CN114089633A (en) * | 2021-11-19 | 2022-02-25 | 江苏科技大学 | Multi-motor coupling drive control device and method for underwater robot |
CN116280140A (en) * | 2023-04-13 | 2023-06-23 | 广东海洋大学 | Ship hybrid power energy management method, equipment and medium based on deep learning |
US11958365B2 (en) * | 2019-12-09 | 2024-04-16 | Jiangxi University Of Science And Technology | Method for dual-motor control on electric vehicle based on adaptive dynamic programming |
CN114089633B (en) * | 2021-11-19 | 2024-04-26 | 江苏科技大学 | Multi-motor coupling driving control device and method for underwater robot |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104809363A (en) * | 2015-05-25 | 2015-07-29 | 中国石油大学(华东) | Three-component combination drive optimization method based on BK approximate dynamic programming |
CN106842948A (en) * | 2017-03-01 | 2017-06-13 | 长春工业大学 | The method for optimally controlling of the HDP molecular distillation systems based on BP networks |
CN106873379A (en) * | 2017-03-31 | 2017-06-20 | 北京工业大学 | A kind of sewage disposal method for optimally controlling based on iteration ADP algorithms |
CN108333925A (en) * | 2018-01-18 | 2018-07-27 | 上海交通大学 | Consider that the ship thrust of input and end conswtraint distributes adaptive optimization method |
CN109143867A (en) * | 2018-09-26 | 2019-01-04 | 上海海事大学 | A kind of energy management method of the hybrid power ship based on ANN Control |
CN109449925A (en) * | 2018-10-29 | 2019-03-08 | 国网甘肃省电力公司 | A kind of adaptive dynamic programming method of multiple target joint optimal operation |
CN109591659A (en) * | 2019-01-14 | 2019-04-09 | 吉林大学 | A kind of pure electric automobile energy management control method of intelligence learning |
CN110046800A (en) * | 2019-03-14 | 2019-07-23 | 南京航空航天大学 | The satellite cluster formation adjusting planing method of space-oriented target cooperative observation |
CN110045761A (en) * | 2019-04-12 | 2019-07-23 | 东南大学 | A kind of intelligent rotating platform control system design method based on adaptive Dynamic Programming |
-
2019
- 2019-08-05 CN CN201910717298.0A patent/CN110406526A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104809363A (en) * | 2015-05-25 | 2015-07-29 | 中国石油大学(华东) | Three-component combination drive optimization method based on BK approximate dynamic programming |
CN106842948A (en) * | 2017-03-01 | 2017-06-13 | 长春工业大学 | The method for optimally controlling of the HDP molecular distillation systems based on BP networks |
CN106873379A (en) * | 2017-03-31 | 2017-06-20 | 北京工业大学 | A kind of sewage disposal method for optimally controlling based on iteration ADP algorithms |
CN108333925A (en) * | 2018-01-18 | 2018-07-27 | 上海交通大学 | Consider that the ship thrust of input and end conswtraint distributes adaptive optimization method |
CN109143867A (en) * | 2018-09-26 | 2019-01-04 | 上海海事大学 | A kind of energy management method of the hybrid power ship based on ANN Control |
CN109449925A (en) * | 2018-10-29 | 2019-03-08 | 国网甘肃省电力公司 | A kind of adaptive dynamic programming method of multiple target joint optimal operation |
CN109591659A (en) * | 2019-01-14 | 2019-04-09 | 吉林大学 | A kind of pure electric automobile energy management control method of intelligence learning |
CN110046800A (en) * | 2019-03-14 | 2019-07-23 | 南京航空航天大学 | The satellite cluster formation adjusting planing method of space-oriented target cooperative observation |
CN110045761A (en) * | 2019-04-12 | 2019-07-23 | 东南大学 | A kind of intelligent rotating platform control system design method based on adaptive Dynamic Programming |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110888323A (en) * | 2019-11-26 | 2020-03-17 | 大连理工大学 | Control method for intelligent optimization of switching system |
US11958365B2 (en) * | 2019-12-09 | 2024-04-16 | Jiangxi University Of Science And Technology | Method for dual-motor control on electric vehicle based on adaptive dynamic programming |
CN111367172A (en) * | 2020-02-28 | 2020-07-03 | 华南理工大学 | Hybrid system energy management strategy based on reverse deep reinforcement learning |
CN111367172B (en) * | 2020-02-28 | 2021-09-21 | 华南理工大学 | Hybrid system energy management strategy based on reverse deep reinforcement learning |
CN112217198B (en) * | 2020-09-01 | 2022-05-17 | 广西大学 | Photovoltaic power generation management method for multi-step graph neural network self-adaptive dynamic planning |
CN112217198A (en) * | 2020-09-01 | 2021-01-12 | 广西大学 | Photovoltaic power generation management method for multi-step graph neural network self-adaptive dynamic planning |
CN113255954A (en) * | 2020-12-03 | 2021-08-13 | 广西大学 | Charging space-time prediction method of hybrid isomorphic depth map neural network |
CN112896161A (en) * | 2021-02-08 | 2021-06-04 | 杭州电子科技大学 | Electric automobile ecological self-adaptation cruise control system based on reinforcement learning |
CN112896161B (en) * | 2021-02-08 | 2022-06-21 | 杭州电子科技大学 | Electric automobile ecological self-adaptation cruise control system based on reinforcement learning |
CN113232651A (en) * | 2021-02-08 | 2021-08-10 | 杭州电子科技大学 | Fuel automobile ecological self-adaptive cruise control method based on ADHDP |
CN114089633A (en) * | 2021-11-19 | 2022-02-25 | 江苏科技大学 | Multi-motor coupling drive control device and method for underwater robot |
CN114089633B (en) * | 2021-11-19 | 2024-04-26 | 江苏科技大学 | Multi-motor coupling driving control device and method for underwater robot |
CN116280140A (en) * | 2023-04-13 | 2023-06-23 | 广东海洋大学 | Ship hybrid power energy management method, equipment and medium based on deep learning |
CN116280140B (en) * | 2023-04-13 | 2023-10-10 | 广东海洋大学 | Ship hybrid power energy management method, equipment and medium based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110406526A (en) | Parallel hybrid electric energy management method based on adaptive Dynamic Programming | |
CN111731303B (en) | HEV energy management method based on deep reinforcement learning A3C algorithm | |
CN111267831B (en) | Intelligent time-domain-variable model prediction energy management method for hybrid electric vehicle | |
Wu et al. | Continuous reinforcement learning of energy management with deep Q network for a power split hybrid electric bus | |
Li et al. | Ecological adaptive cruise control and energy management strategy for hybrid electric vehicles based on heuristic dynamic programming | |
CN105644548B (en) | The energy control method and device of hybrid vehicle | |
CN103935360B (en) | Hybrid electric vehicle complete vehicle torque-split system and method thereof based on parallel control | |
Li et al. | Back-to-back competitive learning mechanism for fuzzy logic based supervisory control system of hybrid electric vehicles | |
Li et al. | Power management for a plug-in hybrid electric vehicle based on reinforcement learning with continuous state and action spaces | |
He et al. | Energy optimization of electric vehicle’s acceleration process based on reinforcement learning | |
CN109591659B (en) | Intelligent learning pure electric vehicle energy management control method | |
CN112668799A (en) | Intelligent energy management method and storage medium for PHEV (Power electric vehicle) based on big driving data | |
Bo et al. | A Q-learning fuzzy inference system based online energy management strategy for off-road hybrid electric vehicles | |
Pei et al. | Fuzzy energy management strategy for parallel HEV based on pigeon-inspired optimization algorithm | |
CN109143867A (en) | A kind of energy management method of the hybrid power ship based on ANN Control | |
CN115793445B (en) | Hybrid electric vehicle control method based on multi-agent deep reinforcement learning | |
CN113110052B (en) | Hybrid energy management method based on neural network and reinforcement learning | |
CN110936824A (en) | Electric automobile double-motor control method based on self-adaptive dynamic planning | |
Zhang et al. | Driving behavior oriented torque demand regulation for electric vehicles with single pedal driving | |
CN110509914A (en) | A kind of energy consumption optimization method of parallel hybrid electric vehicle | |
Yang et al. | Reinforcement learning-based real-time intelligent energy management for hybrid electric vehicles in a model predictive control framework | |
CN115107733A (en) | Energy management method and system for hybrid electric vehicle | |
CN114802180A (en) | Mode prediction system and method for hybrid electric vehicle power system coordination control | |
Li et al. | Distributed cooperative energy management system of connected hybrid electric vehicles with personalized non-stationary inference | |
CN104527637B (en) | Method for controlling hybrid power vehicle and system |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191105 |