CN115716469A

CN115716469A - Output power distribution control method of hybrid power system

Info

Publication number: CN115716469A
Application number: CN202211485333.9A
Authority: CN
Inventors: 李曦; 曾令鸿; 傅俊; 盛闯; 郭子昂; 李贝佳; 李冬
Original assignee: Huazhong University of Science and Technology
Current assignee: Huazhong University of Science and Technology
Priority date: 2022-11-24
Filing date: 2022-11-24
Publication date: 2023-02-28

Abstract

The invention belongs to the field of hybrid power energy distribution, and particularly relates to an output power distribution control method of a hybrid power system, which comprises the following steps: calculating the deviation between the predicted current required actual power and the current actual output power of the acquired PEMFC, the deviation between the current actual output power and the actual output power acquired last time, and the deviation between the service life indexes of the PEMFC acquired next time and the PEMFC acquired last time; calculating the weighted sum of the absolute values of all the deviations, wherein all the weight coefficients take the same sign; and controlling the trained adaptive fuzzy neural network to supply and distribute the difference value of the current required actual power and the current actual output power based on the weighted sum, and outputting the power value to be output to the load by the PEMFC, wherein the power value is inversely proportional to the weighted sum, and the difference value between the difference value and the power value is partially provided or received by other power sources, so that the PEMFC outputs the power value capable of delaying the aging of the PEMFC to the load, and meanwhile, the hybrid power system meets the current required actual power demand of the load.

Description

Output power distribution control method of hybrid power system

Technical Field

The invention belongs to the field of hybrid power energy distribution, and particularly relates to an output power distribution control method of a hybrid power system.

Background

From the beginning of the early civilization period of human beings by using resources such as wood as energy sources, the explosive development of fossil fuels is in the period of industrial revolution, and the new generation of clean energy such as solar energy, wind energy, hydrogen energy and the like gradually occupies the energy market today, and the updating of energy technology is one of the important power sources for promoting the economic development of the world. In the presence of increasingly tense international energy situation and increasingly serious current situation of environmental pollution, a fuel cell using novel clean energy hydrogen energy as power is widely applied to the fields of transportation, industrial production, distributed energy storage and the like by virtue of multiple advantages of environmental protection, high efficiency, reliability, low risk and the like. One point that is contrary to its broad application prospects is the durability of hydrogen-powered fuel cells.

Particularly, in complex conditions where the actual load power demand frequently changes, such as vehicle driving conditions, the PEMFC system for a vehicle needs to adjust the output of electric energy in real time according to the actual power demand calculated according to the vehicle driving conditions, which may cause structural changes or performance degradation of some components inside the PEMFC system, for example, dissolution or deposition of platinum catalyst on a catalyst layer may cause reduction of an electrochemical active area or carbon corrosion of a gas diffusion layer, and the like, and these minor damages are difficult to directly observe. Therefore, in order to maintain the PEMFC system in a stable operating condition, a control strategy for predicting the performance degradation of the PEMFC and delaying aging is important.

At present, a certain amount of work is done on the degradation mechanism analysis of the fuel cell, but the analysis is not comprehensive enough, for example, few factors related to stack attenuation exist in the modeling of the vehicle fuel cell system, and some problems exist in the prediction of the residual life of the vehicle fuel cell; due to the complexity of the fading mechanism, the current experiments and research are still in the exploration stage, and it is very difficult to provide a unified model reflecting the fuel cell attenuation. On the basis, the control methods adopted by the hybrid power system at present include control of a state machine strategy based on droop control, control of an energy management strategy based on frequency separation, control based on a support vector machine, control of energy management based on double-Q reinforcement learning, control based on a deep neural network and the like, but the control strategies rarely take the service life of the hybrid power system into consideration, and it is very important to design a method capable of realizing real-time control and service life extension of the hybrid power system.

Disclosure of Invention

In view of the defects and the improved requirements of the prior art, the invention provides an output power distribution control method of a hybrid power system, and aims to provide an output power distribution control method of a hybrid power system, so as to delay the aging of a PEMFC while rapidly responding to a load requirement in real time.

To achieve the above object, according to an aspect of the present invention, there is provided an output power distribution control method of a hybrid system, including:

predicting the actual power currently required by a load, and acquiring the current actual output power of a PEMFC (proton exchange membrane fuel cell) in a hybrid power system;

calculating the deviation between the current required actual power and the current actual output power, the deviation between the current actual output power and the actual output power acquired last time, and the deviation between the service life indexes of the PEMFC acquired at the current time and the PEMFC acquired at the last time; calculating the weighted sum of the absolute values of all the deviations, wherein all the weight coefficients take the same sign;

and controlling the trained adaptive fuzzy neural network to supply and distribute the difference value between the current required actual power and the current actual output power through fuzzy reasoning based on the value of the weighted sum, outputting a power value to be output to a load by the PEMFC, wherein the power value is in inverse proportion to the weighted sum, and the difference value between the difference value and the power value is partially provided or received by other power sources except the PEMFC, so that the PEMFC outputs a power value capable of delaying the aging of the PEMFC to the load, meanwhile, the hybrid power system meets the current required actual power requirement of the load, and the current control of the hybrid power system is completed.

And further, predicting the current required actual power of the load by adopting the trained GRU prediction neural network based on the historical actual working condition.

Further, each deviation is specifically a sum of squares of the differences.

Further, the lifetime indicator is a residual electrochemically active surface area.

Further, the lifetime indicator is a ratio of a residual electrochemically active surface area to an original electrochemically active surface area.

Further, during weighting and calculation, the value of the weight coefficient of the deviation between the life indexes of the PEMFC under the current collection and the PEMFC under the previous collection is larger than other weight coefficients.

Further, the training sample of the adaptive fuzzy neural network adopts circulation condition data.

The present invention provides an output power distribution control system of a hybrid system, characterized in that the output power distribution control method for executing the hybrid system as described above includes: the device comprises a prediction unit, an acquisition unit and a power distribution unit;

the prediction unit is used for predicting the actual power currently required by the load;

the acquisition unit is used for acquiring the current actual output power of the PEMFC in the hybrid power system;

the attention enhancing unit is used for calculating the deviation of the current required actual power and the current actual output power, the deviation of the current actual output power and the actual output power collected at the previous time, and the deviation of the service life indexes of the PEMFC under the current collection and under the previous collection; calculating the weighted sum of the absolute values of the deviations, wherein each weight coefficient takes the same sign;

the power distribution unit is used for controlling the trained adaptive fuzzy neural network to supply and distribute the difference value between the current required actual power and the current actual output power of the PEMFC through fuzzy reasoning based on the value of the weighted sum, and outputting a power value to be output to a load by the PEMFC, wherein the power value is inversely proportional to the weighted sum, and the difference value between the difference value and the power value is partially provided or received by other power sources except the PEMFC, so that the PEMFC can output a power value capable of delaying the aging of the PEMFC to the load, and meanwhile, the hybrid power system meets the current required actual power demand of the load.

The present invention also provides a computer-readable storage medium including a stored computer program, wherein when the computer program is executed by a processor, the apparatus on which the storage medium is located is controlled to execute the output power distribution control method of a hybrid system as described above.

Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:

(1) The present invention discusses the effect of operating conditions on fuel cell life by delaying fuel cell degradation by controlled means. Specifically, a fuzzy adaptive neural network based on an attention enhancement mechanism is proposed, wherein the attention enhancement mechanism is as follows: calculating the deviation between the current required actual power and the current actual output power of the PEMFC, the deviation between the current actual output power of the PEMFC and the actual output power acquired last time, and the deviation between the service life indexes of the PEMFC acquired at the current time and the service life indexes of the PEMFC acquired last time; and calculating a weighted sum between absolute values of the deviations, wherein each weight coefficient takes the same sign, the weighted sum is used as one input of a fuzzy neural network, the fuzzy neural network supplies and distributes a difference value between the current required actual power and the current actual output power of the PEMFC through fuzzy reasoning, a power value to be output to a load by the PEMFC is output, the power value is inversely proportional to the weighted sum, and the difference value between the difference value and the power value is partially provided or received by other power sources, so that the PEMFC outputs a power value capable of delaying the aging of the PEMFC to the load. That is, the weighted sum is used as an influence factor of the fuzzy neural network output, so that a power value distributed to the PEMFC is inversely proportional to the weighted sum, and the weighted sum approaches 0 through repeated feedback in a reward and punishment manner, thereby realizing rapid power tracking, realizing small power fluctuation, and realizing slow ECSA fluctuation of the PEMFC, and thus fully protecting the PEMFC.

(2) During weighting and calculation, the value of the weight coefficient of the deviation between the service life indexes of the PEMFC under the current collection and the service life indexes under the previous collection is larger than that of other weight coefficients, so that the self-adaptive fuzzy neural network preferentially optimizes and considers delta ECSA, and the effect of prolonging the service life is better achieved.

(3) The invention predicts the actual working condition of the hybrid power system comprising the PEMFC through the GRU prediction neural network and obtains the predicted current load demand power P _req And the data is transmitted to the self-adaptive fuzzy neural network, so that the real-time predictive control function can be realized.

Drawings

FIG. 1 is a block diagram of a hybrid power system according to an embodiment of the present invention;

fig. 2 is a structural diagram of a vehicle PEMFC system according to an embodiment of the present invention;

FIG. 3 is a flowchart of an output power distribution control method of a hybrid power system according to an embodiment of the present invention;

FIG. 4 is a diagram of the results of prediction performed by using a GRU prediction network under four working conditions of CLTC-P, EPA, NYCC, and WLTC according to the embodiment of the present invention;

FIG. 5 is a diagram of the change of ECSA of PEMFC with time obtained by using a conventional fuzzy neural network and a fuzzy neural network based on a prediction and attention enhancement mechanism to regulate and control a hybrid power system under four working conditions of CLTC-P, EPA, NYCC and WLTC according to an embodiment of the present invention;

fig. 6 is a time-varying SOC diagram of a lithium battery obtained by controlling a hybrid power system by using a common fuzzy neural network and a fuzzy neural network based on a prediction and attention enhancement mechanism under four working conditions of CLTC-P, EPA, NYCC and WLTC according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

Example one

An output power distribution control method of a hybrid system, as shown in fig. 1, includes:

calculating the deviation between the current required actual power and the current actual output power, the deviation between the current actual output power and the actual output power acquired last time, and the deviation between the service life indexes of the PEMFC under the current acquisition and the PEMFC under the previous acquisition; calculating the weighted sum of the absolute values of all the deviations, wherein all the weight coefficients take the same sign;

and controlling a trained adaptive fuzzy neural network to supply and distribute a difference value between the current required actual power and the current actual output power based on the value of the weighted sum through fuzzy reasoning, outputting a power value to be output to a load by the PEMFC, wherein the power value is in inverse proportion to the weighted sum, and the difference value between the difference value and the power value is provided or received by other power sources, so that the PEMFC outputs a power value capable of delaying the aging of the PEMFC to the load, and meanwhile, the hybrid power system meets the current required actual power requirement of the load, thereby completing the current control of the PEMFC/lithium battery hybrid power system.

In this embodiment, a Proton Exchange Membrane Fuel Cell (PEMFC)/lithium battery hybrid power system for a vehicle is taken as an example to explain the present invention, and other hybrid power systems including PEMFCs may also be used to achieve the effects of fast power tracking, small power fluctuation, and slow ECSA fluctuation of PEMFCs.

In this embodiment, the durability problem of a Proton Exchange Membrane Fuel Cell (PEMFC)/lithium battery hybrid power system for a vehicle is taken as a starting point, the output electrical characteristics of the PEMFC are distributed from different angles of modeling, prediction and control, and the aging of the PEMFC is delayed while the load demand is quickly responded in real time.

Specifically, firstly, in order to show the remaining life of the PEMFC in a concise and clear manner and with extreme representativeness, in the embodiment, an electrochemical active area (ECSA) caused by dissolution and deposition of platinum in the PEMFC galvanic catalytic layer is selected, which can be used as an optimal choice, and the normalized ECSA is used as an index of the remaining life of the PEMFC, so as to facilitate comparison of the life performance of PEMFCs of different models.

Secondly, in order to be able to control the vehicle PEMFC hybrid system in real time, the controller needs to respond more quickly. There are two ways to enable the controller to react more quickly, one from a hardware upgrade perspective and the other from a software programming perspective. However, the upgrade of hardware has a longer upgrade period compared to software, which is a method not good for users. Therefore, in order to be closer to real life, the embodiment chooses to solve the problem from the perspective of software, and preferably adopts a neural network predictor to predict the future situation in advance, and transmits the prediction information to the adaptive fuzzy neural network as feed-forward information.

Against the background of the research of the invention, various randomness problems exist in actual traffic, and the future is often required to be snooped by a small amount of historical data within a short time. In order to predict the future working condition quickly and efficiently in real time, the embodiment provides a light neural network to realize the function. Therefore, the GRU prediction neural network can be selected as a preferred predictor to predict the current required actual power of the load based on historical actual working conditions.

Regarding the GRU predictive neural network, the GRU predictive neural network employed in the present embodiment has three gates: an update gate, a reset gate, and an output gate. With the aid of the update gate and the reset gate, the GRU finally implements the function of adjusting the input values, the memory values and the output values. In order to better predict the real-time working condition, the practice can be taken as a preferred example, the number of input layer neurons is 1, the number of hidden layer neurons is 288, and the number of output layer neurons is 1. In training, the solver is set to adam, and the gradient threshold is set to 1. To avoid the over-fit condition, an initial learning rate of 0.005 was assigned and the learning rate was reduced by multiplying by a factor of 0.2 after every 500 rounds of training. Predicting the actual working condition of the PEMFC/lithium battery hybrid power system through a GRU prediction neural network, and predicting the current load required power P obtained through prediction _req And the data is transmitted to the self-adaptive fuzzy neural network, so that the real-time predictive control function can be realized.

In addition, in the control, in order to simultaneously satisfy the requirements of the PEMFC hybrid system on providing sufficient energy for the normal running of the vehicle and the long service life of the PEMFC, great requirements are put on the control method. The fuzzy neural network control has strong learning and reasoning capabilities. It can reasonably infer how output is needed to achieve PEMFC lifetime in a complex environment based on prior experience, inherent system properties, constraints on the system itself, etc. And is therefore better suited to address such problems.

The fuzzy neural network is a product obtained by combining the neural network and the fuzzy system, draws the advantages of the neural network and the fuzzy system, not only has the capabilities of learning, imagining and self-adaption, but also comprises the capabilities of reasoning calculation and logical thinking. The essence of the fuzzy neural network is to input a conventional neural network into the fuzzy input signal and the fuzzy weights. The input layer of the neural network is an input signal of the fuzzy system, the output layer is an output signal of the fuzzy system, and the hidden layer is a membership function and a fuzzy rule. The fuzzy neural network has the following characteristics from the aspects of expression, storage, application and acquisition of knowledge in the environment:

(1) From the expression mode of knowledge, the fuzzy system can express the empirical knowledge of people, so that the understanding is convenient, and the neural network can only describe the complex functional relation among a large amount of data and is difficult to understand.

(2) From the knowledge storage mode, the fuzzy system centralizes the knowledge existence rule, and the neural network has the characteristic of distributed storage in the knowledge existence weight coefficient.

(3) From the application mode of knowledge, the fuzzy system and the neural network both have the characteristic of parallel processing, the fuzzy system has few simultaneously activated rules and small calculation amount, and the neural network has many involved neurons and large calculation amount.

(4) From the knowledge acquisition mode, the rules of the fuzzy system are provided or designed by experts and are difficult to acquire automatically. And the weight coefficient of the neural network can be learned from the input and output samples without being set by a human.

Although the fuzzy neural network is also a local approximation network, the fuzzy neural network is established according to a fuzzy system model, and each node and all parameters in the network have obvious physical meanings, so initial values of the parameters can be determined according to system or qualitative knowledge (in the embodiment, the input physical meaning comprises power, and the output physical meaning is current), and then the fuzzy neural network can be converged to a required input-output relationship quickly by using the learning algorithm, which is the advantage of the fuzzy neural network compared with the simple neural network. Meanwhile, because the neural network has the structure of the neural network, the learning and the adjustment of the parameters are easier, which is the advantage of the neural network compared with a simple fuzzy logic system.

In addition, based on the complexity of the problem to be studied in the present invention, not only a control scheme with learning, self-correction and strong generalization capability is required to implement the control function, but also the control scheme is required to fully protect the controlled object, improve the performance thereof in a targeted manner, and propose a fuzzy adaptive neural network controller with an attention-enhancing mechanism as a core controller.

The attention-enhancing mechanism is: calculating the deviation between the current required actual power and the current actual output power of the PEMFC, the deviation between the current actual output power of the PEMFC and the actual output power collected at the previous time, and the deviation between the service life indexes of the PEMFC collected at the current time and the service life indexes of the PEMFC collected at the previous time; and calculating a weighted sum between absolute values of the deviations, wherein each weight coefficient takes the same number, the weighted sum is used as one input of a fuzzy neural network, the fuzzy neural network supplies and distributes a difference value between the current required actual power and the current actual output power of the PEMFC through fuzzy reasoning, a power value to be output to a load by the PEMFC is output, the power value is inversely proportional to the weighted sum, the difference value between the difference value and the power value is partially provided or received by a lithium battery, so that the PEMFC outputs a power value capable of delaying the aging of the PEMFC to the load, the fuzzy neural network learns how to regulate and control the output so as to delay the aging of the PEMFC most, correct the output and enhance the generalization capability of the fuzzy neural network, therefore, the weighted sum is used as an influence factor of the output of the fuzzy neural network, the power value distributed to the PEMFC is inversely proportional to the weighted sum, the weighted sum is fed back for multiple times in a manner, the weighted sum is close to 0, the power rapid tracking is realized (the deviation between the current required actual power and the current output power of the PEMFC) is reflected in the slow acquisition of the PEMFC), the power is small (the deviation between the current actual output power and the acquired power of the PEMFC), and the life of the PEMFC is realized by a slow acquisition), and the protection of the PEMFC is realized under the slow acquisition of the power can be realized under the slow acquisition of the power can be realized.

That is, compared with the conventional fuzzy neural network, the attention-enhancing mechanism of the fuzzy adaptive neural network based on the attention-enhancing mechanism provided in this embodiment requires more fuzzy adaptive neural network to achieve fast power tracking, achieve small power fluctuation, and achieve slow ECSA fluctuation. When the unweighted sum approaches 0, the weighted sum needs to be continuously adopted to carry out reward and punishment on the neural network.

The attention-enhancing mechanism can be expressed as:

wherein, P _error Representing predicted current load demand power P _req And the current actual output power P of the PEMFC _FC Difference of (a) P _FC Denotes the PEMFC output power variation between two sampling points, Δ ECSA denotes the ECSA variation between two sampling points, ξ ₁ 、ξ ₂ And xi ₃ Respectively represent P _error 、ΔP _FC And the weight values of delta ECSA, which are positive numbers.

By varying xi ₁ 、ξ ₂ And xi ₃ Can change the adaptive fuzzy neural network pair P _error 、ΔP _FC And the priority relation optimally considered with the delta ECSA, and preferably, during weighting and calculation, the value of the weight coefficient of the deviation between the service life indexes of the PEMFC under the current collection and the service life indexes of the PEMFC under the previous collection is larger than that of other weight coefficients. At the same time make P _error 、ΔP _FC And Δ ECSA exhibits better performance in hybrid systems.

In general, the method of the embodiment can predict the actual working condition in real time through the GRU neural network prediction module, and corrects the action of the fuzzy neural network by adopting an attention enhancement mechanism, i.e., has extremely high real-time performance. Additionally, degradation of ECSA in the PEMFC may be mitigated and life decay of other power sources may also be mitigated (e.g., mitigating SOC fluctuations of the lithium battery) during hybrid system power distribution. Namely, the control scheme realizes the real-time control and the service life extension of the PEMFC hybrid power system.

Preferably, each deviation is the sum of the squares of the differences, which is convenient for calculation. The training sample of the adaptive fuzzy neural network adopts circulation condition data.

With respect to the solution of ECSA, the following description is now given:

the ECSA decay model is divided into a water content-influenced ECSA decay sub-model, a voltage-influenced ECSA decay sub-model and a catalytic layer gas flow passage sub-model.

Wherein the submodel describing the ECSA decay affected by the water content isObtaining different membrane water content lambda _m Empirical models summarized for the case of ECSA attenuation effects. Water content of membrane lambda _m The method is related to the relative humidity of the gas at the position of the proton exchange membrane, so that the specific submodel building method is to obtain the influence of the relative humidity of the gas at the position of the proton exchange membrane on the ECSA attenuation under the conditions that the relative humidity of the gas at the position of the proton exchange membrane is 50% and 100%, and then obtain an empirical formula of the water content on the ECSA attenuation by adopting a linear interpolation mode. Inputs to the submodel describing the ECSA decay affected by water content include: inlet air humidity, air flow rate, current, etc. The output of the submodel describing the ECSA decay for water content impact is the water content impact factor. The sub-model can ensure the health state of the PEMFC by controlling the water content of the PEMFC in a certain range, thereby achieving the aim of health control.

Wherein the water content of the membrane is lambda _m Generally, the method cannot be obtained by a direct measurement method, and the method is obtained by an indirect calculation method. The indirect calculation requires the knowledge of the relative humidity of the gas introduced. Assuming that the relative humidity RH =0.5 in the gas at the cathode and anode outlet of the PEMFC in the initial state and the humidity RH of the inlet gas is set to be an adjustable amount, the inlet water vapor partial pressure of the cathode and anode is expressed by the following equation (1):

P _v,i ＝RH×P _sat,i (1)

P _sat,i indicating the ambient atmospheric pressure at i, e.g. P _sat,m Represents the atmospheric pressure at the proton exchange membrane; p is _v,i Denotes the partial pressure of water vapour at i, for example: p _v,in Denotes the partial pressure of water vapor at the inlet, P _v,out Indicating the partial pressure of water vapour, P, at the outlet _v,m Representing the partial pressure of water vapor at the proton exchange membrane. P _sat,i It can be calculated according to empirical formula (2) recommended by Emanuel, in Pa:

wherein T represents the temperature at i. Assuming that the water vapor partial pressure from the inlet to the outlet of the cathode and anode exhibits a linear increasing relationship, the water vapor partial pressure in the cathode and anode gas is represented by the following equation (3):

P _v,in denotes the water vapor pressure at the inlet, P _v,out Indicating the water vapor pressure at the outlet, the invention uses P due to the relatively short distance from the inlet to the outlet of the cathode and anode _v In place of P _v,m . In the formation of P _v,m Water content lambda of the proton exchange membrane _m The following empirical formulas (4) and (5) are adopted to obtain:

λ _m denotes the water content at the proton exchange membrane, a _m Denotes the relative humidity at the proton exchange membrane, P _sat,m Representing the atmospheric pressure at the proton exchange membrane.

Thereby obtaining the water content lambda of the film _m Then, the water content lambda of the film is calculated _m Impact on ECSA attenuation.

The submodel describing the voltage effect on ECSA decay is an empirical model summarized by taking the cases of different voltage and voltage change rates on the ECSA decay effect.

The submodel for describing the catalytic layer gas flow passage is an empirical model for digitizing the physical description of the catalytic layer, and specifically comprises the set total surface area of Pt particles, the flow rate of oxygen input into the flow passage, the flow rate of hydrogen input into the flow passage, and the maximum current density which can pass through the catalytic layer.

The time taken for ECSA to decay to a minimum is longer the lower the water content, as derived from empirical formulas; the smaller the voltage, the smaller the rate of change of voltage, the longer it takes for the ECSA to decay to a minimum. Therefore, in order to maintain the health of PEMFCs in hybrid systems, it is necessary to maintain a lower water content and a lower voltage change rate.

Using the residual active surface area S (N) and the initial total active surface area S ₀ Constructing a relation formula as shown in formula (6):

k denotes the influence factor and N the number of cycles, but it can also be replaced by time/s, when the humidity is 50%, k =2.05 × 10 by fitting to the actual data ^-4 min ^-1 When the humidity was 100%, k =3.72 × 10 was obtained by fitting to actual data ^-4 min ^-1 . Is provided with

When the ECSA is continuously degraded, there is a minimum ECSA marked as ECSA _min (ii) the ECSA _min It is defined as the minimum platinum surface area, and takes a value of 0.2. The relationship between the two is shown in formula (7):

k _total ＝k _SC ×k _T ×k _RH ×k _UPL ×k _dwell (8)

in the formula (7), k _total Total degradation rate, k, representing all influencing factors _SC Is the degradation rate of the standard voltage cycle, k _T Is the temperature induced rate factor of decay, k, of the ECSA _RH Is the relative humidity induced rate factor of ECSA decay, k _UPL Is the ECSA decay rate factor, k, caused by the upper voltage value _dwell Is the ECSA decay rate factor caused by the voltage duration. The influence of temperature, humidity, voltage and the like on the degradation of ECSA in the PEMFC can be independently studied. Definition k _UPL 、k _dwell The calculation formula is shown in formulas (9) and (10):

k _UPL ＝e ^{C×(UPL-0.95)} (9)

k _dwell ＝0.38+0.29×t _dwell (10)

wherein C is a constant value and takes the value of 0.00152mV ^-1 UPL denotes the local maximum of the voltage over the sampling period, t _dwell Indicating the duration of time during which the voltage is maintained at a local minimum of voltage during the sampling period. Only the relationship between relative humidity, voltage and the rate of degradation of ECSA in PEMFCs was studied in the present invention, i.e., assuming k _SC ＝k _T ＝1。

To better illustrate the invention, an example validation was performed on a proton exchange membrane based fuel cell system experimental test platform. The testing platform mainly comprises a PEMFC pile, an air supply system, a hydrogen supply system and a cooling water path system, and is shown in FIG. 2.

The output power distribution control method provided by the embodiment and the control scheme based on the common fuzzy neural network are operated on the test platform, simulation is carried out under four different working conditions of WLTP, CLTC-P, EPA and NYCC, and simulation results are compared. Wherein, each working condition data is obtained by converting the original speed data of each working condition into power data through an empirical formula.

As shown in fig. 3, the execution flow of the output power allocation control method provided in this embodiment may be as follows:

(1) Historical information of actual load demand power is transmitted into a load power prediction module to obtain the load power demand P at the moment _req 。

(2)P _req On the one hand with the output power P of the PEMFC _FC Making a difference to obtain P _error And on the other hand, the signal is input to the controller to be used as one of the influencing factors of the regulation.

(3)P _error 、ΔP _FC Delta ECSA acts on the attention enhancement mechanism together, and after calculating the influence value J generated by the mechanism, the value is combined with P _req 、P _error And the input signals are jointly input into the adaptive fuzzy neural network.

(4) And the energy output of the PEMFC and the lithium battery is adjusted in real time through a control method deployed in the self-adaptive fuzzy neural network. Meanwhile, the controller also has the functions of delaying the performance attenuation of the PEMFC and protecting the lithium battery, so that the PEMFC and the lithium battery can work in a normal working range.

(5) The power output of the PEMFC/lithium battery hybrid system meets the actual load power demand at the present moment in a feedback manner, when the demanded power is higher than the output power of the PEMFC, the part of the demanded power higher than the output power of the PEMFC will be provided by the lithium battery, and when the demanded power is lower than the output power of the PEMFC, the part of the PEMFC output power higher than the demanded power will charge the lithium battery.

(6) And (4) returning to the step (1), repeating the steps, and finally achieving the purposes of responding to the load power demand in real time and delaying the performance degradation of the PEMFC.

FIG. 4 shows the result of prediction by using GRU prediction network under the working conditions of CLTC-P, EPA, NYCC and WLTC. It can be seen that the predicted load demand power and the actual load demand power have been matched very well. In order to more clearly present the difference between the predicted result and the actual result of the GRU prediction network, the embodiment adopts RMSE as the error indicator, and the recorded result is shown in the table. Therefore, the prediction of different working states by utilizing the GRU prediction network can ensure that the prediction result is very close to the actual result. The RMSE value is kept low, which means that the prediction error of the GRU predictive neural network is small, and means that a method of estimating future operating conditions using the predicted values of the GRU predictive network is accurate and feasible.

Difference between predicted result and actual result of GRU under different working conditions

FIG. 5 shows the ECSA of PEMFC varying with time by using the common fuzzy neural network and the fuzzy neural network based on prediction and attention-enhancing mechanism to regulate the hybrid system under the CLTC-P, EPA, NYCC, WLTC working conditions. The improved fuzzy neural network, i.e. the fuzzy adaptive neural network of the attention enhancing mechanism, adopted in the present embodiment can better reduce the attenuation of ECSA in PEMFC compared to the conventional fuzzy neural network. This demonstrates that the attention-enhancing mechanism employed in this example has a positive effect on suppressing ECSA decay.

FIG. 6 shows the SOC of the lithium battery changing with time by using the ordinary fuzzy neural network and the fuzzy neural network based on the prediction and attention enhancement mechanism to regulate and control the hybrid power system under the working conditions of CLTC-P, EPA, NYCC and WLTC. The attention-enhancing mechanism adopted by the invention not only protects the PEMFC, slows down the attenuation of the ECSA, but also slows down the change of the SOC. And the slow change of the SOC means that the service life of the lithium battery is prolonged. Meanwhile, the attention enhancement mechanism adopted by the embodiment also ensures that the SOC is in a normal working range. The fuzzy neural network of the attention-enhancing mechanism employed herein is more capable of extending the lifespan of PEMFC hybrid systems than the general fuzzy neural network.

In summary, the present embodiment provides a control method for a PEMFC/lithium battery hybrid system, which is a fuzzy adaptive neural network control method based on prediction and attention-enhancing mechanism, and belongs to the field of hybrid power energy distribution. The fuzzy adaptive neural network control method based on the prediction and attention enhancement mechanism predicts the actual working condition in real time through the GRU neural network prediction module and corrects the action of the fuzzy neural network controller by adopting the attention enhancement mechanism, and simultaneously considers the predicted current load demand power, the current actual output power of the PEMFC and the influence value of the attention enhancement mechanism, and the three factors are in cross coupling combined action, so that the PEMFC/lithium battery hybrid power system is regulated and controlled more accurately and timely. The fuzzy adaptive neural network control method based on the prediction and attention enhancement mechanism can slow down the degradation of ECSA in the PEMFC and slow down the SOC fluctuation of the lithium battery in the process of power distribution of a hybrid power system. Namely, the control scheme realizes the real-time control and the service life extension of the PEMFC hybrid power system.

Example two

A controller of a hybrid system for executing an output power distribution control method of a hybrid system as described above, comprising: the device comprises a prediction unit, an acquisition unit and a power distribution unit.

The prediction unit is used for predicting the actual power currently required by the load; the acquisition unit is used for acquiring the current actual output power of the PEMFC; the attention enhancing unit is used for calculating the deviation between the current required actual power and the current actual output power of the PEMFC, the deviation between the current actual output power of the PEMFC and the actual output power collected at the previous time and the deviation between the service life indexes of the PEMFC under the current collection and the service life indexes of the PEMFC under the previous collection; calculating the weighted sum of the absolute values of the deviations, wherein each weight coefficient takes the same sign; the power distribution unit is used for controlling the trained adaptive fuzzy neural network to supply and distribute the difference value between the current required actual power and the current actual output power of the PEMFC through fuzzy reasoning based on the value of the weighted sum, and outputting a power value to be output to a load by the PEMFC, wherein the power value is inversely proportional to the weighted sum, and the difference value between the difference value and the power value is partially provided or received by other power sources, so that the PEMFC outputs a power value capable of delaying the aging of the PEMFC to the load, and meanwhile, the hybrid power system meets the current required actual power demand of the load.

The related technical solution is the same as the first embodiment, and is not described herein again.

EXAMPLE III

A computer-readable storage medium including a stored computer program, wherein the computer program, when executed by a processor, controls an apparatus in which the storage medium is located to perform an output power distribution control method of a hybrid system according to an embodiment.

It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. An output power distribution control method of a hybrid system, characterized by comprising:

calculating the deviation between the current required actual power and the current actual output power, the deviation between the current actual output power and the actual output power collected at the previous time, and the deviation between the service life indexes of the PEMFC under the current collection and the PEMFC under the previous collection; calculating the weighted sum of the absolute values of all the deviations, wherein all the weight coefficients take the same sign;

2. The output power distribution control method of claim 1, wherein the actual power currently required by the load is predicted based on historical actual conditions using a trained GRU prediction neural network.

3. The output power allocation control method according to claim 1, wherein each deviation is specifically a sum of squares of the differences.

4. The output power distribution control method according to claim 1, wherein the lifetime index is a remaining electrochemically active surface area.

5. The output power distribution control method according to claim 1, wherein the lifetime index is a ratio of a remaining electrochemically active surface area to an original electrochemically active surface area.

6. The output power distribution control method according to claim 1, wherein in the weighting and calculating, a value of a weighting coefficient of a deviation between the lifetime indexes of the PEMFC at the current collection and at the previous collection is larger than other weighting coefficients.

7. The output power allocation control method according to claim 1, wherein the training samples of the adaptive fuzzy neural network employ cyclic behavior data.

8. An output power distribution control system of a hybrid system for performing an output power distribution control method of a hybrid system according to any one of claims 1 to 7, comprising: the device comprises a prediction unit, an acquisition unit and a power distribution unit;

9. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored computer program, wherein the computer program, when executed by a processor, controls an apparatus on which the storage medium resides to perform an output power distribution control method of a hybrid system according to any one of claims 1 to 7.