CN102951144A - Self-regulating neural network energy managing method based on minimum power loss algorithm - Google Patents

Abstract

The invention relates to a self-regulating neural network energy managing method based on a minimum power loss algorithm. The self-regulating neural network energy managing method comprises the following steps that (1) a vehicle control unit through a CAN (controller area network) bus from an automobile power accessory acquires an energy control strategy to compute required data, and synchronously collects a real time efficiency value of an existing energy source; (2) the vehicle control unit judges whether the complete data is received or not; (3) the vehicle control unit judges whether an instruction of updating a neural network is received or not; and if yes, the vehicle control unit updates the neutral network; (4) the vehicle control unit calculates the existing optimum power distribution through the neural network according to the received data needed by the energy control strategy calculation (5) the computed optimum power distribution of the neutral network is modified by utilizing a power gain coefficient; and (6) the vehicle control unit transmits a power distribution result to an energy controller through the CAN bus. Compared with the prior art, the self-regulating neutral network energy managing method based on the minimum power loss algorithm, disclosed by the invention, has the advantages of good fuel economy, capability of realizing real time regulation, good expansibility and the like.

Description

Self aligning neural network energy management process based on the minimum power loss algorithm

Technical field

The present invention relates to the control technology for electric motor car field, especially relate to a kind of self aligning neural network energy management process based on the minimum power loss algorithm.

Background technology

Extended-range electric vehicle has the basic characteristics of hybrid vehicle as a branch of hybrid vehicle.Be different from orthodox car and pure electric vehicle, extended-range electric vehicle can provide power by two kinds of energy sources.The feature in this dual energy source has increased the alerting ability of system, under the coordination control of car load energy management system, dual energy source and miscellaneous part cooperatively interact, and can carry out multiple optimal combination, form different power system mode of operations, to adapt to different driving cycles.

Simultaneously, the dual energy source has also increased the complexity of extended-range electric vehicle hybrid power system, how to carry out the switching of multiple-working mode, how to realize the optimal control of energy flow between the dual energy source, becomes a difficult point of hybrid power system field of research.In order to solve by the caused mode of operation switching problem in hybrid power system dual energy source, and the Optimal Control Problem of energy flow, have to increase an energy management system and solve the new problem that traditional combustion engine automobile and pure electric automobile are not had, namely must control by certain energy control strategy (Enerhy Control Strategy) coordination and the distribution of energy flow between the dual energy source.Control policy is the core of energy management system, is the key point that realizes the targets such as extended-range electric vehicle energy-saving and environmental protection.

At present research is four class hybrid vehicle energy management strategies the most widely: rule-based control policy, instantaneous optimization control policy, global optimization control policy and based on the ADAPTIVE CONTROL of optimized algorithm.

Select the control parameter according to component capabilities characteristic and engineering experience, in the variation range of control parameter, use one group of threshold value (logic threshold) that it is divided into different zones, at different control parameter regions, hybrid vehicle is according to different states and Power supply work pattern, thereby forms one group of control law.The control parameter has the demand power of the speed of a motor vehicle, running car, the power of driving engine, the state-of-charge (SOC) of battery etc. usually.Whether threshold value according to the control parameter is determined, can be divided into again static logic thresholding energy management strategy and fuzzy rule energy management strategy, the former comes the energy supply state of dividing system with accurate threshold value, and the latter carries out the afterwards energy supply state of decision systems of Fuzzy processing to threshold value.The advantage of the maximum of this class strategy is to be easy to Project Realization.But no matter whether rule-based energy management strategy carried out Optimization about control parameter, and still there is certain limitation in it aspect raising of fuel economy.

The instantaneous optimization strategy generally is to adopt " equivalent fuel oil consumption is minimum " method or " loss in efficiency is minimum " method, and the two principle is similar, and it is unified all to be that energy consumptive use ad hoc approach with two energy sources quantizes, so that calculate minimum total consumption.The instantaneous optimization strategy is optimum in each step-length, but can't guarantee optimum in the whole service interval, and needs a large amount of floating point operations and more accurate auto model, and calculated amount is large, realizes difficulty.This class energy management strategy is mainly used in having obtained good fuel efficiency in Computer Simulation in combustion engine-battery hybrid system at present, has also obtained some achievements on real vehicle is used.But this class strategy is used not extensively at real vehicle, because its collection for the real-time vehicle running state parameter, processing requirements are higher.

The global optimization control policy, in knowing in advance all processes of running car under the condition of all duty parameters, can realize the global optimization of energy management, the various control algorithm application is arranged in this energy management strategy, such as dynamic programming algorithm, Discrete Dynamic Programming algorithm, genetic algorithm etc.The global optimization pattern has realized optimization truly, but the algorithm of realizing this strategy more complicated all often, calculated amount is also very large, and need to obtain in advance all road informations, is difficult to be applied in the real-time control of actual vehicle.

Based on the ADAPTIVE CONTROL of optimized algorithm, can control parameter to adapt to the variation of driving cycle according to current driving conditions and the automatic adjustment in the automatic predict future of road conditions a period of time.So-called self adaptation is exactly at each time step, requires the adjustment component mode of operation according to current driving conditions and road conditions, by optimized algorithm, is guaranteeing under the optimized prerequisite of objective function torque demand reasonably to be distributed to driving engine and motor.Although the objective function model of ADAPTIVE CONTROL, optimized algorithm etc. are different, but because adaptive control is wanted a large amount of driving engine operating data of Real-time Collection, calculate fuel oil oil consumption and emission point, optimizing process is complicated, calculated amount is large, causes it can't be applied in practice at present.

Summary of the invention

Purpose of the present invention is exactly to provide the self aligning neural network energy management process based on the minimum power loss algorithm that a kind of fuel economy is good, can realize real-time adjustment for the defective that overcomes above-mentioned prior art existence.

Purpose of the present invention can be achieved through the following technical solutions:

A kind of self aligning neural network energy management process based on the minimum power loss algorithm, the method is used for the energy management of stroke-increasing electric automobile, described stroke-increasing electric automobile comprises entire car controller, CAN bus, energy source, energy controller and automobile power annex, described entire car controller connects energy controller and automobile power annex by the CAN bus respectively, described energy source is connected with energy controller, and described method may further comprise the steps:

1) entire car controller sends interrogation signal by the CAN bus to the automobile power annex, obtain energy management policy calculation desired data from the automobile power annex, described data comprise car load demand power and storage battery SOC, and entire car controller gathers the Real time Efficiency value of current energy source simultaneously;

2) entire car controller judges whether to receive complete data, if then execution in step 3), if not, then return step 1);

3) entire car controller judges whether to receive the instruction that needs to upgrade neural network, if then upgrade execution in step 4 after the neural network), if not, then execution in step 4);

4) entire car controller goes out current optimal power allocation according to the energy management policy calculation desired data that receives by neural computing;

5) use power gain factor that the optimal power allocation that neural computing goes out is revised;

6) entire car controller to energy controller transmit power assignment result, is finished entire car controller to the distribution control of each energy source horsepower output in the hybrid power system by the CAN bus.

Described automobile power annex comprises heat dissipation subsystem and air-conditioning subsystem.

Described step 3) instruction in comprises chaufeur appointment update instruction and the automatic update instruction of program.

Described step 3) concrete steps of upgrading neural network in comprise:

A) obtain related data by the CAN bus, calculate the loss in efficiency of different capacity under distributing according to following power system instantaneous power loss formula,

$\left\{\begin{array}{c}{P}_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{dis}})+P_b({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{RE}})/{\mathrm{\η}}_{\mathrm{DC}-\mathrm{DC}}{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{chr}}\\ P_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{chr}})+P_b{\mathrm{\η}}_{\mathrm{chr}}(1-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{dis}})\end{array}\right.$

In the formula, P _REBe distance increasing unit horsepower output, η _{PE_max}Be the most effective value of distance increasing unit, η _REBe efficiency value under the distance increasing unit current power,

Be distance increasing unit average effiiciency value, P _bBe storage battery horsepower output, η _DisBe battery discharging efficient, η _ChrBe battery charge efficient,

Be the average discharge efficiency of storage battery,

Be storage battery average charge efficient, η _DC-DCBe the DC-DC converter efficiency;

B) according to the result in line computation, neural network is trained;

C) the new neural network with the training gained substitutes former neural network.

Described energy source comprises distance increasing unit and storage battery, and described distance increasing unit is connected with storage battery by the DC-DC conv.

Described step 4) going out current optimal power allocation by neural computing in specifically may further comprise the steps:

A) will send neural network to by the data that the CAN bus obtains;

B) go out current optimal power allocation by neural computing, comprise the distance increasing unit horsepower output.

Described step 5) correction in is specifically: the distance increasing unit horsepower output that neural computing is gone out and the distance increasing unit power gain factor K distance increasing unit horsepower output after adjusted that multiplies each other; Described K calculates by following formula:

$K={\left(\frac{P_{\mathrm{adj}}}{P_{\mathrm{dis}}}\right)}^{\frac{{\mathrm{RE}}_{\max }-{\mathrm{RE}}_{\mathrm{cons}}}{{\mathrm{RE}}_{\max }}}$

P _DisMinimum power when discharging with storage battery for distance increasing unit, P _AdjBe coefficient adjustment power, RE _MaxBe distance increasing unit energy maximum reserves, RE _ConsBe the distance increasing unit energy consumption.

Compared with prior art, the present invention has the following advantages:

1) the present invention is based on instantaneous power min algorithm and in conjunction with the horsepower output of each energy source in the self aligning ANN (Artificial Neural Network) Control stroke-increasing electric automobile, have fuel economy good, be easy to the characteristics such as real vehicle realization;

2) the present invention can according to the change in use of power system efficient, make corresponding adjustment;

3) energy management method of the present invention can be applied to combustion engine---storage battery easily, and fuel battery engines---the hybrid power systems such as storage battery have good extendability.

Description of drawings

Fig. 1 is the schematic flow sheet of the inventive method;

Fig. 2 is the structural representation of stroke-increasing electric automobile among the embodiment.

The specific embodiment

The present invention is described in detail below in conjunction with the drawings and specific embodiments.

Embodiment

As shown in Figure 1, a kind of self aligning neural network energy management process based on the minimum power loss algorithm, the method is used for the energy management of stroke-increasing electric automobile.The method is based on the instantaneous power min algorithm in the instantaneous optimization energy management strategy, derive the mathematical formulae for the target hybrid power system, two major influence factors of energy distribution have been determined to affect: car load demand power and storage battery SOC, cooperate again real vehicle to collect the Real time Efficiency value of two energy sources, employing is the energy management rule of one group of dual input, single output in the method for line computation with the energy management Policy refinement.Adopt the BP neural network that the energy management rule that extracts is carried out match, re-use the nerve network controller that the neural network that trains is used for energy management.In addition, according to the continuous change of two energy source Real time Efficiency values, controller will adopt according to instruction and calculate in real time the also method of neural network training, constantly update the neural network of controller inside with the real source efficiency state of reflection real vehicle.Described instruction comprises the automatic update instruction of chaufeur appointment update instruction and program setting.The power stage that this controller is controlled hybrid power system in real time according to parameters such as the demand torque of hybrid power system and storage battery SOC distributes, to improve the fuel efficiency of hybrid power system.

As shown in Figure 2, stroke-increasing electric automobile in the present embodiment comprises entire car controller (VMS) 5, CAN bus, distance increasing unit 1, storage battery 2, distance increasing unit controller 3, battery controller 4 and automobile power annex 6, described entire car controller connects distance increasing unit controller 3, battery controller 4 and automobile power annex 6 by the CAN bus respectively, described distance increasing unit 1 is connected with distance increasing unit controller 3, described storage battery 2 is connected with battery controller 4, and described distance increasing unit 1 is connected with storage battery 2 by the DC-DC conv.The control parameter of energy management strategy is finished data interaction by the CAN bus between the controller of VMS and energy source.After VMS obtains energy management policy calculation desired data from the CAN bus, neural computing by inside goes out optimal power allocation, and the optimal power allocation correction of using power gain factor k that neural computing is gone out obtains final power division, by the CAN bus this distribution is sent to the controller of energy source again, to finish the output power to energy source.

Above-mentioned self aligning neural network energy management process based on the minimum power loss algorithm specifically may further comprise the steps:

In step 401, entire car controller (VMS) comprises demand power, storage battery SOC etc. to CAN bus inquiry energy management policy calculation desired data, and described energy management policy calculation desired data is according to obtaining based on the minimum power loss algorithm.Then execution in step 402.

In step 402, to choose energy management policy calculation desired data the data from each car accessories to the CAN bus transfer and send to VMS, entire car controller gathers the Real time Efficiency value of current energy source simultaneously by the CAN bus.Then execution in step 403.

In step 403, VMS judges whether to receive complete data, if complete, execution in step 404; If imperfect, return step 402.

In step 404, VMS judges whether to receive the instruction that needs to upgrade neural network, if then execution in step 405, if not, then execution in step 406.

In step 405, upgrade neural network, then execution in step 406.

The concrete steps of described renewal neural network comprise:

A) obtain related data by the CAN bus, calculate the loss in efficiency of different capacity under distributing according to following power system instantaneous power loss formula,

$\left\{\begin{array}{c}{P}_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{dis}})+P_b({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{RE}})/{\mathrm{\η}}_{\mathrm{DC}-\mathrm{DC}}{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{chr}}\\ P_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{chr}})+P_b{\mathrm{\η}}_{\mathrm{chr}}(1-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{dis}})\end{array}\right.$

In the formula, P _REBe distance increasing unit horsepower output, η _{RE_max}Be the most effective value of distance increasing unit, η _REBe efficiency value under the distance increasing unit current power,

Be distance increasing unit average effiiciency value, P _bBe storage battery horsepower output, η _DisBe battery discharging efficient, η _ChrBe battery charge efficient,

Be the average discharge efficiency of storage battery,

Be storage battery average charge efficient, η _DC-DCBe the DC-DC converter efficiency;

B) according to the result in line computation, neural network is trained;

C) the new neural network with the training gained substitutes former neural network.

In step 406, VMS goes out optimal power allocation according to the energy management policy calculation desired data of receiving by current neural computing.Then enter step 405.

Going out current optimal power allocation by neural computing specifically may further comprise the steps:

A) will send neural network to by the data that the CAN bus obtains;

B) go out current optimal power allocation by neural computing, comprise the distance increasing unit horsepower output.

In step 407, use power gain factor K that the optimal power allocation that neural computing goes out is revised; Then enter step 408.

Be to regulate the balance between distance increasing unit energy surplus and the electric energy surplus, introducing distance increasing unit power gain factor K, the distance increasing unit horsepower output that neural computing is gone out and the K distance increasing unit horsepower output after adjusted that multiplies each other.When COEFFICIENT K is wanted to make distance increasing unit energy surplus, emit relatively high power; When distance increasing unit energy surplus is less, emit original power.For realizing dynamic adjustment, the distance increasing unit power gain factor is designed to the power function relevant with distance increasing unit energy surplus, K calculates by following formula:

$K={\left(\frac{P_{\mathrm{adj}}}{P_{\mathrm{dis}}}\right)}^{\frac{{\mathrm{RE}}_{\max }-{\mathrm{RE}}_{\mathrm{cons}}}{{\mathrm{RE}}_{\max }}}$

P _DisMinimum power when discharging with storage battery for distance increasing unit, P _AdjBe coefficient adjustment power, RE _MaxBe distance increasing unit energy maximum reserves, RE _ConsBe the distance increasing unit energy consumption.

In step 408, VMS finishes control end VMS to the distribution of each energy source horsepower output of hybrid power system by the controller transmit power assignment result of CAN bus to each energy source.

Claims (7)

1. self aligning neural network energy management process based on the minimum power loss algorithm, the method is used for the energy management of stroke-increasing electric automobile, described stroke-increasing electric automobile comprises entire car controller, CAN bus, energy source, energy controller and automobile power annex, described entire car controller connects energy controller and automobile power annex by the CAN bus respectively, described energy source is connected with energy controller, it is characterized in that described method may further comprise the steps:

1) entire car controller sends interrogation signal by the CAN bus to the automobile power annex, obtain energy management policy calculation desired data from the automobile power annex, described data comprise car load demand power and storage battery SOC, and entire car controller gathers the Real time Efficiency value of current energy source simultaneously;

2) entire car controller judges whether to receive complete data, if then execution in step 3), if not, then return step 1);

3) entire car controller judges whether to receive the instruction that needs to upgrade neural network, if then upgrade execution in step 4 after the neural network), if not, then execution in step 4);

4) entire car controller goes out current optimal power allocation according to the energy management policy calculation desired data that receives by neural computing;

5) use power gain factor that the optimal power allocation that neural computing goes out is revised;

6) entire car controller to energy controller transmit power assignment result, is finished entire car controller to the distribution control of each energy source horsepower output in the hybrid power system by the CAN bus.

2. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 1 is characterized in that described automobile power annex comprises heat dissipation subsystem and air-conditioning subsystem.

3. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 1 is characterized in that described step 3) in instruction comprise that chaufeur specifies update instruction and the automatic update instruction of program.

4. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 1 is characterized in that described step 3) in upgrade neural network concrete steps comprise:

A) obtain related data by the CAN bus, calculate the loss in efficiency of different capacity under distributing according to following power system instantaneous power loss formula,

$\left\{\begin{array}{c}{P}_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{dis}})+P_b({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{RE}})/{\mathrm{\η}}_{\mathrm{DC}-\mathrm{DC}}{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{chr}}\\ P_{\mathrm{RE}}({\mathrm{\η}}_{\mathrm{RE}\_\max }-{\mathrm{\η}}_{\mathrm{RE}})+P_b(1-{\mathrm{\η}}_{\mathrm{chr}})+P_b{\mathrm{\η}}_{\mathrm{chr}}(1-{\stackrel{\‾}{\mathrm{\η}}}_{\mathrm{dis}})\end{array}\right.$

In the formula, P _REBe distance increasing unit horsepower output, η _{RE_max}Be the most effective value of distance increasing unit, η _REBe efficiency value under the distance increasing unit current power,

Be distance increasing unit average effiiciency value, P _bBe storage battery horsepower output, η _DisBe battery discharging efficient, η _ChrBe battery charge efficient,

Be the average discharge efficiency of storage battery,

Be storage battery average charge efficient, η _DC-DCBe the DC-DC converter efficiency;

B) according to the result in line computation, neural network is trained;

C) the new neural network with the training gained substitutes former neural network.

5. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 1 is characterized in that described energy source comprises distance increasing unit and storage battery, and described distance increasing unit is connected with storage battery by the DC-DC conv.

6. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 5 is characterized in that described step 4) in go out current optimal power allocation by neural computing and specifically may further comprise the steps:

A) will send neural network to by the data that the CAN bus obtains;

B) go out current optimal power allocation by neural computing, comprise the distance increasing unit horsepower output.

7. a kind of self aligning neural network energy management process based on the minimum power loss algorithm according to claim 6, it is characterized in that described step 5) in correction specifically: the distance increasing unit horsepower output that neural computing is gone out and the distance increasing unit power gain factor K distance increasing unit horsepower output after adjusted that multiplies each other; Described K calculates by following formula:

$K={\left(\frac{P_{\mathrm{adj}}}{P_{\mathrm{dis}}}\right)}^{\frac{{\mathrm{RE}}_{\max }-{\mathrm{RE}}_{\mathrm{cons}}}{{\mathrm{RE}}_{\max }}}$

P _DisMinimum power when discharging with storage battery for distance increasing unit, P _AdjBe coefficient adjustment power, RE _MaxBe distance increasing unit energy maximum reserves, RE _ConsBe the distance increasing unit energy consumption.

Images