Output Power Allocation Control Method for a Hybrid Power System

Abstract

The present invention belongs to the field of hybrid power energy distribution, and specifically relates to a method for controlling the output power distribution of a hybrid power system, including: calculating the deviation between the predicted current required actual power and the currently collected actual output power of the PEMFC, the deviation between the currently collected actual output power and the previously collected actual output power, and the deviation between the currently collected and the previously collected life indexes of the PEMFC; calculating the weighted sum of the absolute values of the deviations, and taking the weight coefficients of the same sign; controlling the trained adaptive fuzzy neural network to perform supply distribution on the difference between the currently required actual power and the currently collected actual output power based on the weighted sum, and output the power value to be output by the PEMFC to the load, and this power value is inversely proportional to the weighted sum, and the difference between the above-mentioned difference and this power value is provided or received by other power sources, so that the PEMFC outputs a power value to the load that can delay the aging of the PEMFC, and at the same time, the hybrid power system meets the current required actual power demand of the load.

Description

Technical Field

[0001] The present invention belongs to the field of hybrid power energy distribution, and more specifically, relates to an output power distribution control method for a hybrid power system. Background Art

[0002] Since the early days of human civilization when resources such as wood were used as energy sources, through the explosive development of fossil fuels during the Industrial Revolution, to the present day when new-generation clean energy sources such as solar energy, wind energy, and hydrogen energy are gradually occupying a place in the energy market, the renewal of energy technologies is one of the important driving forces for the development of the world economy. Fuel cells powered by the new clean energy source of hydrogen energy are widely used in fields such as transportation, industrial production, and distributed energy storage due to their many advantages such as environmental protection, high efficiency, reliability, and low risk. Contrary to its broad application prospects is the durability problem of hydrogen fuel cells.

[0003] Especially in complex working conditions where the actual load power demand changes frequently, such as vehicle driving conditions, the vehicle-mounted PEMFC system needs to adjust the electrical energy output in real time according to the actual required power calculated based on the vehicle driving conditions. This is very likely to cause structural changes or performance degradation in some components inside the PEMFC system, such as the dissolution or deposition of platinum catalysts in the catalyst layer resulting in a reduction in the electrochemically active area or carbon corrosion of the gas diffusion layer, etc. These minor damages are difficult to directly observe. If not paid attention to, it may lead to a significant decline in the output characteristics of the fuel cell stack, and in severe cases, it may even affect the operation of the entire vehicle. Therefore, in order to maintain the continuous and stable working conditions of the PEMFC system, the prediction of PEMFC performance degradation and the control strategy for delaying aging are crucial.

[0004] At present, certain work has been done on the analysis of the degradation mechanism of fuel cells, but it is not comprehensive enough. For example, in the modeling of vehicle-mounted fuel cell systems, few factors related to fuel cell stack degradation are involved, and there are some problems in predicting the remaining life of vehicle-mounted fuel cells; due to the complexity of the degradation mechanism, current experiments and research are still in the exploratory stage, and it is very difficult to give a unified model reflecting fuel cell degradation. On this basis, the current control methods for hybrid power systems include control based on the state machine strategy of droop control, control based on the energy management strategy of frequency separation, control based on support vector machines, control based on double-Q reinforcement learning for energy management, and control based on deep neural networks, etc. However, these control strategies rarely consider the life extension of hybrid power systems. Designing a method that can achieve real-time control and life extension of hybrid power systems is very important. Summary of the Invention

[0005] In view of the defects and improvement requirements of the prior art, the present invention provides a method for controlling the output power distribution of a hybrid power system, aiming to propose a method for controlling the output power distribution of a hybrid power system to delay the aging of PEMFC while rapidly responding to load requirements in real time.

[0006] To achieve the above object, according to one aspect of the present invention, a method for controlling the output power distribution of a hybrid power system is provided, including:

[0007] Predict the actual power currently required by the load, and collect the current actual output power of the PEMFC in the hybrid power system;

[0008] Calculate the deviation between the currently required actual power and the current actual output power, the deviation between the current actual output power and the actual output power collected last time, and the deviation between the life index of the PEMFC under the current collection and the life index under the previous collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients are of the same sign.

[0009] Control the trained adaptive fuzzy neural network to perform supply distribution on the difference between the currently required actual power and the current actual output power through fuzzy inference based on the value of the weighted sum, and output the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the difference value and this power value is provided or received by other power sources other than the PEMFC, so that the PEMFC outputs a power value that can delay the aging of the PEMFC to the load, and at the same time, the hybrid power system meets the current actual power requirement of the load, completing the current control of the hybrid power system.

[0010] Further, a trained GRU prediction neural network is used to predict the actual power currently required by the load based on historical actual operating conditions.

[0011] Further, each deviation is specifically the sum of the squares of the differences.

[0012] Further, the life index is the remaining electrochemically active surface area.

[0013] Further, the life index is the ratio of the remaining electrochemically active surface area to the original electrochemically active surface area.

[0014] Further, when calculating the weighted sum, the weight coefficient of the deviation between the life index of the PEMFC under the current collection and the life index under the previous collection takes a value greater than other weight coefficients.

[0015] Further, the training samples of the adaptive fuzzy neural network adopt cyclic operating condition data.

[0016] The present invention provides an output power distribution control system for a hybrid power system, which is characterized by being used to execute an output power distribution control method for a hybrid power system as described above, including: a prediction unit, a collection unit, and a power distribution unit;

[0017] The prediction unit is used to predict the actual power currently required by the load;

[0018] The collection unit is used to collect the current actual output power of the PEMFC in the hybrid power system;

[0019] The attention enhancement unit is used to calculate the deviation between the currently required actual power and the current actual output power, the deviation between the current actual output power and the actual output power collected last time, and the deviation between the PEMFC life index under the current collection and the PEMFC life index under the previous collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients are of the same sign;

[0020] The power distribution unit is used to control the trained adaptive fuzzy neural network based on the value of the weighted sum, and through fuzzy reasoning, supply and distribute the difference between the currently required actual power and the current actual output power of the PEMFC, and output the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the difference value and this power value is provided or received by other power sources other than the PEMFC, so that the PEMFC outputs a power value to the load that can delay the aging of the PEMFC, and at the same time, the hybrid power system meets the current actual power demand of the load.

[0021] The present invention also provides a computer-readable storage medium, which includes a stored computer program. When the computer program is run by a processor, it controls the device where the storage medium is located to execute an output power distribution control method for a hybrid power system as described above.

[0022] Generally speaking, through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0023] (1) The present invention discusses the influence of operating conditions on the lifespan of a fuel cell, and delays the decay of the fuel cell by means of control. Specifically, a fuzzy adaptive neural network based on an attention enhancement mechanism is proposed. The attention enhancement mechanism is as follows: calculate the deviation between the currently 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 last time, and the deviation between the lifespan index of the PEMFC under the current collection and that under the previous collection; and calculate the weighted sum of the absolute values of the deviations. Among them, the weight coefficients are of the same sign. Take this weighted sum as an input of the fuzzy neural network. The fuzzy neural network performs supply allocation on the difference between the currently required actual power and the current actual output power of the PEMFC through fuzzy inference, and outputs the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum. The difference part between the difference value and this power value is provided or received by other power sources, so that the PEMFC outputs to the load a power value that can delay the aging of the PEMFC. That is, the above-mentioned weighted sum serves as an influencing factor for the output of the fuzzy neural network, making the power value allocated to the PEMFC inversely proportional to the weighted sum. Through multiple feedback in the form of rewards and punishments, the weighted sum approaches 0, achieving fast power tracking, small power fluctuations, and slow fluctuations in the ECSA of the PEMFC, thereby fully protecting the PEMFC.

[0024] (2) When calculating the weighted sum, the weight coefficient of the deviation between the lifespan index of the PEMFC under the current collection and that under the previous collection takes a value greater than other weight coefficients, so that the adaptive fuzzy neural network preferentially optimizes and considers ΔECSA, and better achieves the effect of extending the lifespan.

[0025] (3) The present invention predicts the actual working conditions of a hybrid power system including a PEMFC through a GRU prediction neural network, and transmits the predicted current load demand power P req to the adaptive fuzzy neural network, thereby enabling a real-time prediction control function. Description of the Drawings

[0026] Figure 1 is a schematic diagram of the framework of an output power allocation control method for a hybrid power system provided by an embodiment of the present invention;

[0027] Figure 2 is a structural diagram of a vehicle-mounted PEMFC system provided by an embodiment of the present invention;

[0028] Figure 3 is a flowchart of an output power allocation control method for a hybrid power system provided by an embodiment of the present invention;

[0029] Figure 4The result graph of prediction using the GRU prediction network under four working conditions of CLTC-P, EPA, NYCC, and WLTC provided by the embodiments of the present invention;

[0030] Figure 5 The graph of the change of the ECSA of PEMFC over time obtained by regulating the hybrid power system using a general 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 provided by the embodiments of the present invention;

[0031] Figure 6 The graph of the change of the SOC of the lithium battery over time obtained by regulating the hybrid power system using a general 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 provided by the embodiments of the present invention. Detailed implementation manners

[0032] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] Embodiment 1

[0034] An output power distribution control method for a hybrid power system, as Figure 1 shown, includes:

[0035] Predict the current actual power required by the load, and collect the current actual output power of PEMFC in the hybrid power system;

[0036] Calculate 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 last time, and the deviation between the life index of PEMFC under the current collection and the last collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients are of the same sign;

[0037] Based on the value of the weighted sum, the trained adaptive fuzzy neural network performs supply allocation on the difference between the currently required actual power and the currently actual output power through fuzzy inference, and outputs the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the said difference and this power value is provided or received by other power sources, so that the PEMFC outputs to the load a power value that can delay the aging of the PEMFC. At the same time, the hybrid power system meets the currently required actual power demand of the load, thus completing the current control of the PEMFC / lithium battery hybrid power system.

[0038] In this embodiment, a vehicle proton exchange membrane fuel cell (Proton Exchange Membrane Fuel Cell, PEMFC) / lithium battery hybrid power system is selected as an example to explain the present invention. Other hybrid power systems including PEMFC can also be used to achieve the effects of fast power tracking, small power fluctuations, and slow fluctuations in the ECSA of the PEMFC.

[0039] Starting from the durability problem of the vehicle proton exchange membrane fuel cell (Proton Exchange Membrane Fuel Cell, PEMFC) / lithium battery hybrid power system, this embodiment distributes the output electrical characteristics of the PEMFC from different perspectives of modeling, prediction, and control, so as to delay the aging of the PEMFC while quickly responding to the load demand in real time.

[0040] Specifically, first of all, in order to concisely and representatively display the remaining life of the PEMFC, the electrochemically active surface area (ECSA) caused by the dissolution and deposition of platinum in the catalyst layer of the PEMFC stack is selected in this embodiment. Preferably, the normalized ECSA is used as an index for the remaining service life of the PEMFC to facilitate the comparison of the life performance of different models of PEMFC.

[0041] Secondly, in order to be able to control the vehicle PEMFC hybrid power system in real time, the controller needs to respond more quickly. There are two ways to make the controller respond more quickly. One is from the perspective of hardware upgrade, and the other is from the perspective of software programming. However, compared with software, the upgrade cycle of hardware is longer, which is not a favorable method for users. Therefore, in order to be closer to real life, this embodiment selects to solve this problem from the perspective of software. Preferably, a neural network predictor is used to predict the future situation in advance, and the prediction information is transmitted as feedforward information to the adaptive fuzzy neural network.

[0042] Regarding the background of the research of the present invention, in actual traffic, there are various randomness problems, and it is often necessary to glimpse the future through a small amount of historical data in a very short time. In order to be able to predict the future working conditions quickly and efficiently in real time, this embodiment proposes a lightweight neural network to achieve this function. Therefore, preferably, a GRU prediction neural network is selected as the predictor to predict the actual power required by the load based on the historical actual working conditions. This method has a low mean square error, high accuracy, and since the neural network has been trained, it requires less computing power during actual operation, has a relatively low operation cost, and a fast operation speed.

[0043] Regarding the GRU prediction neural network, the GRU prediction neural network adopted in this embodiment has three gates: an update gate, a reset gate, and an output gate. With the assistance of the update gate and the reset gate, the GRU finally realizes the functions of adjusting the input value, the memory value, and the output value. In order to be able to better predict the real-time working conditions, through practice, a preferred example is that the number of neurons in the input layer is 1, the number of neurons in the hidden layer is 288, and the number of neurons in the output layer is 1. During training, the solver is set to adam, and the gradient threshold is set to 1. In order to avoid overfitting, the initial learning rate is specified as 0.005, and the learning rate is reduced by multiplying by a factor of 0.2 after every 500 rounds of training. The actual working conditions of the PEMFC / lithium battery hybrid system are predicted by the GRU prediction neural network, and the predicted current load demand power P req is transmitted to the adaptive fuzzy neural network, so as to be able to realize the real-time prediction control function.

[0044] In addition, in the control, in order to achieve both providing sufficient energy for the normal driving of the vehicle by the PEMFC hybrid system and extending the life of the PEMFC, this poses greater requirements on the control method. And the fuzzy neural network control has strong learning and reasoning abilities. Therefore, it can reasonably infer how to output in a complex environment according to prior experience, the inherent properties of the system, and the constraints of the system itself in order to achieve the extension of the life of the PEMFC. Therefore, it is more suitable for dealing with such problems.

[0045] The fuzzy neural network is the product obtained by combining the neural network and the fuzzy system, absorbing the advantages of both. It not only enables itself to have the abilities of learning, imagination, and self-adaptation, but also includes the abilities of reasoning calculation and logical thinking. The essence of the fuzzy neural network is to input fuzzy input signals and fuzzy weights into the conventional neural network. The input layer of the neural network is the input signal of the fuzzy system, the output layer is the output signal of the fuzzy system, and the hidden layer is the membership function and the fuzzy rule. From the aspects of the expression, storage, application, and acquisition of knowledge from the environment, the fuzzy neural network has the following characteristics:

[0046] (1) In terms of the knowledge representation method, fuzzy systems can express human empirical knowledge, which is easy to understand, while neural networks can only describe the complex functional relationships between a large amount of data and are difficult to understand.

[0047] (2) In terms of the knowledge storage method, fuzzy systems store knowledge in rule sets, and neural networks store knowledge in weight coefficients, both of which have the characteristics of distributed storage.

[0048] (3) In terms of the knowledge application method, both fuzzy systems and neural networks have the characteristics of parallel processing. The number of rules simultaneously activated in fuzzy systems is not large, and the computational amount is small, while neural networks involve many neurons and have a large computational amount.

[0049] (4) In terms of the knowledge acquisition method, the rules of fuzzy systems are provided or designed by experts and are difficult to obtain automatically. The weight coefficients of neural networks can be learned from input-output samples without human setting.

[0050] Although the fuzzy neural network is also a local approximation network, it is established according to the fuzzy system model, and each node and all parameters in the network have obvious physical meanings. Therefore, the initial values of these parameters can be determined according to the system or qualitative knowledge (in this embodiment, the physical meaning of the input includes power, and the physical meaning of the output is current). Then, using the above learning algorithm, it can quickly converge to the required input-output relationship, which is the advantage of the fuzzy neural network compared with the previous simple neural network. At the same time, because it has the structure of a neural network, the learning and adjustment of parameters are relatively easy, which is its advantage compared with a simple fuzzy logic system.

[0051] In addition, based on the complexity of the research problem of the present invention, not only is a control scheme with strong learning, self-correcting, and generalization abilities required to achieve the control function, but also this control scheme needs to fully protect the controlled object, specifically improve its performance, and a fuzzy adaptive neural network controller with an attention enhancement mechanism is proposed as the core controller.

[0052] The attention enhancement mechanism is as follows: calculate the deviation between the currently 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 last time, and the deviation between the life index of the PEMFC under the current collection and that under the previous collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients have the same sign, and take this weighted sum as an input of the fuzzy neural network. The fuzzy neural network performs supply allocation on the difference between the currently required actual power and the current actual output power of the PEMFC through fuzzy inference, and outputs the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum. The difference between the said difference value and this power value is provided or received by the lithium battery, so that the PEMFC outputs to the load a power value that can delay the aging of the PEMFC, thereby enabling the fuzzy neural network to learn how to regulate the output to most effectively delay the aging of the PEMFC, correcting the output, and enhancing the generalization ability of the fuzzy neural network. Therefore, the above-mentioned weighted sum serves as an influencing factor for the output of the fuzzy neural network, making the power value allocated to the PEMFC inversely proportional to the weighted sum. Through the method of rewards and punishments and multiple feedbacks, the weighted sum approaches 0, achieving fast power tracking (reflected in the deviation between the currently required actual power and the current actual output power of the PEMFC), achieving small power fluctuations (reflected in the deviation between the current actual output power of the PEMFC and the actual output power collected last time), and achieving slow fluctuations in the ECSA of the PEMFC (reflected in the deviation between the life index of the PEMFC under the current collection and that under the previous collection), which can fully protect the PEMFC.

[0053] That is to say, compared with the traditional fuzzy neural network, the fuzzy adaptive neural network based on the attention enhancement mechanism proposed in this embodiment requires the fuzzy adaptive neural network to achieve fast power tracking, small power fluctuations, and slow ECSA fluctuations to a greater extent. When the weighted sum has not yet approached 0, it is necessary to continue to use the above-mentioned weighted sum to reward and punish the neural network.

[0054] The attention enhancement mechanism can be expressed as:

[0055]

[0056] where P error represents the difference between the predicted current load demand power P req and the current actual output power P FC of the PEMFC, ΔP FC represents the change value of the PEMFC output power between two sampling points, ΔECSA represents the change value of the ECSA between two sampling points, and ξ1, ξ2, and ξ3 respectively represent P error , ΔP FCThe weight values of and ΔECSA are all positive numbers.

[0057] By changing the relative magnitudes of ξ1, ξ2, and ξ3, the priority relationship of the adaptive fuzzy neural network for optimizing P error , ΔP FC and ΔECSA can be changed. Preferably, during the weighted sum calculation, the weight coefficient value of the deviation between the PEMFC life index under the current acquisition and that under the previous acquisition is greater than other weight coefficients. At the same time, P error , ΔP FC and ΔECSA exhibit good performance in the hybrid power system.

[0058] Generally speaking, the method of this embodiment can perform real-time prediction on the actual working conditions through the GRU neural network prediction module, and uses the attention enhancement mechanism to correct the actions of the fuzzy neural network, that is, it has extremely strong real-time performance. In addition, during the process of power distribution in the hybrid power system, the degradation of ECSA in the PEMFC can be slowed down, and the life attenuation of other power sources can also be slowed down (for example, slowing down the SOC fluctuation of lithium batteries). That is, this control scheme realizes the real-time control and life extension of the PEMFC hybrid power system.

[0059] Preferably, the above-mentioned deviations are the sum of the squares of the differences, which is convenient for calculation. The training samples of the adaptive fuzzy neural network adopt cyclic working condition data.

[0060] Regarding the solution of ECSA, the following explanations are given:

[0061] The attenuation model of ECSA is divided into a sub-model of ECSA attenuation affected by water content, a sub-model of ECSA attenuation affected by voltage, and a sub-model of the gas flow channel of the catalyst layer.

[0062] Among them, the sub-model describing the ECSA attenuation affected by water content is an empirical model summarized by obtaining the influence of different membrane water contents λ m on ECSA attenuation. The membrane water content λ m is also related to the gas relative humidity at the proton exchange membrane. Therefore, the specific method for building the sub-model is to obtain their influences on ECSA attenuation respectively under the conditions that the gas relative humidity at the proton exchange membrane is 50% and 100%, and then use the linear interpolation method to obtain the empirical formula of water content on ECSA attenuation. The inputs of the sub-model describing the ECSA attenuation affected by water content include: intake humidity, gas flow rate, current, etc. The output of the sub-model describing the ECSA attenuation affected by water content is the water content influence factor. Through this sub-model, it is possible to ensure the health state of the PEMFC by controlling the water content of the PEMFC within a certain range, achieving the purpose of its health management and control.

[0063] Among them, the membrane water content λ m Generally cannot be obtained by direct measurement. In this embodiment, it is obtained by indirect calculation. And indirect calculation requires knowing the relative humidity of the inlet gas. Assume that in the gases at the anode and cathode outlets of the PEMFC, the relative humidity RH = 0.5 in the initial state, and the humidity RH of the inlet gas is set as an adjustable quantity. Then the water vapor partial pressures at the inlets of the anode and cathode are as shown in Equation (1):

[0064] P v,i = RH × P sat,i (1)

[0065] P sat,i Represents the ambient atmospheric pressure at i. For example, P sat,m Represents the atmospheric pressure at the proton exchange membrane; P v,i Represents the water vapor partial pressure at i. For example: P v,in Represents the water vapor partial pressure at the inlet, P v,out Represents the water vapor partial pressure at the outlet, P v,m Represents the water vapor partial pressure at the proton exchange membrane. P sat,i Can be calculated according to the empirical formula (2) recommended by Emanuel, and its unit is Pa:

[0066]

[0067] Among them, T represents the temperature at i. Assume that the water vapor partial pressure from the inlets of the anode and cathode to the outlets shows a linear increasing relationship. Then the water vapor partial pressures in the anode and cathode gases are as shown in Equation (3):

[0068]

[0069] P v,in Represents the water vapor pressure at the inlet, P v,out Represents the water vapor pressure at the outlet. Since the distance from the inlets to the outlets of the anode and cathode is relatively short, the present invention uses P v To replace P v,m . After obtaining P v,m , the water content λ of the proton exchange membrane m Is obtained by using the empirical formulas (4) and (5):

[0070]

[0071]

[0072] λ m Represents the water content at the proton exchange membrane, a m Represents the relative humidity at the proton exchange membrane, P sat,m Represents the atmospheric pressure at the proton exchange membrane.

[0073] The membrane water content λ is obtained therefrom. m After that, the membrane water content λ is calculated. m The influence on the ECSA decay is considered.

[0074] The sub-model describing the influence of voltage on ECSA decay is an empirical model summarized by obtaining the influence of different voltages and voltage change rates on ECSA decay.

[0075] The sub-model describing the gas flow channels in the catalyst layer is an empirical model that numerically describes the physics of the catalyst layer, specifically including the set total surface area of Pt particles, the oxygen flow rate input to the flow channel, the hydrogen flow rate input to the flow channel, and the maximum current density that can pass through the catalyst layer.

[0076] Derived from the empirical formula, when the water content is smaller, the time consumed for the ECSA to decay to the minimum value is longer; when the voltage is smaller and the voltage change rate is smaller, the time consumed for the ECSA to decay to the minimum value is longer. Therefore, in order to maintain the healthy state of the PEMFC in the hybrid power system, it is necessary to maintain a low water content, a low voltage, and a low voltage change rate.

[0077] A relational expression is constructed using the remaining active surface area S(N) and the initial total active surface area S0, as shown in Equation (6):

[0078]

[0079] k represents the influence factor, N represents the number of cycles, and it can also be replaced by time / s. When the humidity is 50%, through fitting for the actual data, k = 2.05×10 -4 min -1 When the humidity is 100%, through fitting for the actual data, k = 3.72×10 -4 min -1 . Suppose When the ECSA is continuously degraded, there is a minimum ECSA, denoted as ECSA min , and ECSA min is defined as the minimum platinum surface area, with a value of 0.2. The relationship between the two is as shown in Equation (7):

[0080]

[0081] k total = k SC × k T × k RH × k UPL × k dwell (8)

[0082] In Equation (7), k totalRepresents the total degradation rate of all influencing factors, k SC Is the degradation rate of the standard voltage cycle, k T Is the ECSA decay rate factor caused by temperature, k RH Is the ECSA decay rate factor caused by relative humidity, k UPL Is the ECSA decay rate factor caused by the upper limit voltage value, k dwell Is the ECSA decay rate factor caused by the voltage duration. The effects of temperature, humidity, voltage, etc. on the ECSA degradation in PEMFC can be studied independently. Define k UPL 、k dwell The calculation formulas are shown in Eqs. (9) and (10):

[0083] k UPL =e C×(UPL-0.95) (9)

[0084] k dwell =0.38 + 0.29×t dwell (10)

[0085] In the formula, C is a constant value, and its value is 0.00152 mV -1 , UPL represents the local maximum value of the voltage within the sampling period, and t dwell Represents the duration during which the voltage remains at the local minimum value of the voltage within the sampling period. In the present invention, only the relationship between relative humidity, voltage and the ECSA degradation rate in PEMFC is studied, that is, it is assumed that k SC =k T =1.

[0086] To better illustrate the present invention, an example verification is carried out on an experimental test platform based on a proton exchange membrane fuel cell system. Among them, the test platform is mainly composed of a PEMFC stack and three subsystems: an air supply system, a hydrogen supply system, and a cooling water circuit system, as Figure 2 shown.

[0087] The output power distribution control method proposed in this embodiment and the control scheme based on a general fuzzy neural network are run on the test platform, and simulations are carried out under four different working conditions of WLTP, CLTC-P, EPA, and NYCC, and the simulation results are compared. Among them, the data of each working condition are obtained by converting the original speed data of each working condition into power data through an empirical formula.

[0088] The output power distribution control method proposed in this embodiment, as Figure 3 shown, the execution process can be as follows:

[0089] (1) The historical information of the actual load demand power is input into the load power prediction module to obtain the load power demand P at this momentreq 。

[0090] (2)P req On the one hand, it makes a difference with the output power P of the PEMFC FC to obtain P error , and on the other hand, it is input to the controller as one of the influencing factors for regulation.

[0091] (3)P error 、ΔP FC 、ΔECSA act on the attention enhancement mechanism together. After calculating the influence value J generated by it, it is input to the adaptive fuzzy neural network together with P req 、P error together.

[0092] (4) Through the control method deployed in the adaptive fuzzy neural network, the energy outputs of the PEMFC and the lithium battery are adjusted in real time. At the same time, the controller also has the functions of delaying the performance decay of the PEMFC and protecting the lithium battery, enabling them to work within the normal working range.

[0093] (5) The power output of the PEMFC / lithium battery hybrid system meets the actual load power demand at the current moment through feedback. When the demand power is higher than the output power of the PEMFC, the part where the demand power is higher than the output power of the PEMFC will be provided by the lithium battery, and when the demand power is lower than the output power of the PEMFC, the part where the output power of the PEMFC is higher than the demand power will charge the lithium battery.

[0094] (6) Return to step (1) and repeat the cycle like this, finally achieving the purpose of real-time response to the load power demand and delaying the performance decay of the PEMFC.

[0095] Figure 4 , which is the result of prediction using the GRU prediction network under the CLTC-P, EPA, NYCC, and WLTC working conditions. It can be found that the predicted load demand power is already very consistent with the actual load demand power. In order to more clearly present the gap between the prediction results of the GRU prediction network and the actual results, RMSE is used as its error index in this embodiment, and the recorded results are shown in the table. It can be seen that using the GRU prediction network to predict different working states can ensure that its prediction results are very close to the actual results. The RMSE value remains at a very low level, which indicates that the prediction error of the GRU prediction neural network is very small, and at the same time means that the method of using the predicted value of the GRU prediction network to estimate the future working conditions is accurate and feasible.

[0096] The gap between the GRU prediction results and the actual results under different working conditions

[0097]

[0098] Figure 5 , under the working conditions of CLTC-P, EPA, NYCC, and WLTC, the ordinary fuzzy neural network and the fuzzy neural network based on the prediction and attention enhancement mechanism are used to regulate the hybrid power system, and the change of the ECSA of PEMFC with time is obtained. The improved fuzzy neural network adopted in this embodiment, that is, the fuzzy adaptive neural network with attention enhancement mechanism, can better reduce the attenuation of ECSA in PEMFC compared with the ordinary fuzzy neural network. This shows that the attention enhancement mechanism adopted in this embodiment has a positive effect on suppressing the attenuation of ECSA.

[0099] Figure 6 , under the working conditions of CLTC-P, EPA, NYCC, and WLTC, the ordinary fuzzy neural network and the fuzzy neural network based on the prediction and attention enhancement mechanism are used to regulate the hybrid power system, and the change of the SOC of the lithium battery with time is obtained. The attention enhancement mechanism adopted in the present invention not only protects the PEMFC and slows down the attenuation of ECSA, but also slows down the change of SOC. Slowing down the change of SOC means that the life of the lithium battery is extended. At the same time, the attention enhancement mechanism adopted in this embodiment also ensures that the SOC is within the normal working range. Compared with the ordinary fuzzy neural network, the fuzzy neural network with attention enhancement mechanism adopted in this paper can better extend the service life of the PEMFC hybrid power system.

[0100] Generally speaking, this embodiment provides a control method for the PEMFC / lithium battery hybrid power system. This method is a fuzzy adaptive neural network control method based on prediction and attention enhancement mechanism, belonging to the field of hybrid power energy distribution. The fuzzy adaptive neural network control method based on prediction and attention enhancement mechanism uses the GRU neural network prediction module to predict the actual working conditions in real time and adopts the attention enhancement mechanism to correct the actions of the fuzzy neural network controller. At the same time, considering the predicted current load demand power, the current actual output power of PEMFC, and the influence value of the attention enhancement mechanism, the three factors interact through cross-coupling to more accurately and real-time regulate the PEMFC / lithium battery hybrid power system. The fuzzy adaptive neural network control method based on prediction and attention enhancement mechanism can slow down the degradation of ECSA in PEMFC and the SOC fluctuation of the lithium battery during the power distribution process of the hybrid power system. That is, this control scheme realizes the real-time control and life extension of the PEMFC hybrid power system.

[0101] Embodiment 2

[0102] A controller for a hybrid power system, which is used to execute an output power distribution control method for a hybrid power system as described above, includes: a prediction unit, a collection unit, and a power distribution unit.

[0103] Among them, the prediction unit is used to predict the actual power currently required by the load; the collection unit is used to collect the current actual output power of the PEMFC; the attention enhancement unit is used to calculate the deviation between the currently 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 last time, and the deviation between the life index of the PEMFC under the current collection and the life index of the PEMFC under the previous collection; and calculate the weighted sum of the absolute values of each deviation, where each weight coefficient has the same sign; the power distribution unit is used to control the trained adaptive fuzzy neural network based on the value of the weighted sum, through fuzzy inference, to supply and distribute the difference between the currently required actual power and the current actual output power of the PEMFC, and output the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the difference value and this power value is provided or received by other power sources, so that the PEMFC outputs a power value to the load that can delay the aging of the PEMFC, and at the same time the hybrid power system meets the current actual power demand of the load.

[0104] The related technical solutions are the same as those in Embodiment 1 and will not be elaborated here.

[0105] Embodiment 3

[0106] A computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein when the computer program is run by a processor, it controls the device where the storage medium is located to execute an output power distribution control method for a hybrid power system as described in Embodiment 1.

[0107] The related technical solutions are the same as those in Embodiment 1 and will not be elaborated here.

[0108] Those skilled in the art can easily understand that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for controlling the output power distribution of a hybrid power system, characterized in that, Including: Predict the current actual power required by the load and collect the current actual output power of the PEMFC in the hybrid power system; Calculate the deviation between the current actual power required and the current actual output power, the deviation between the current actual output power and the actual output power collected last time, and the deviation between the life index of the PEMFC under the current collection and the life index of the PEMFC under the previous collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients have the same sign; Control the trained adaptive fuzzy neural network to perform supply allocation on the difference between the current actual power required and the current actual output power through fuzzy inference based on the value of the weighted sum, and output the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the difference value and this power value is provided or received by other power sources other than the PEMFC, so that the PEMFC outputs a power value to the load that can delay the aging of the PEMFC, and at the same time the hybrid power system meets the current actual power requirement of the load, completing the current control of the hybrid power system.

2. The output power distribution control method according to claim 1, characterized in that Use the trained GRU prediction neural network to predict the current actual power required by the load based on historical actual operating conditions.

3. The output power distribution control method according to claim 1, characterized in that, Each deviation is specifically the sum of the squares of the differences.

4. The output power distribution control method according to claim 1, characterized in that The life index is the remaining electrochemically active surface area.

5. The output power distribution control method according to claim 1, characterized in that The life index is the ratio of the remaining electrochemically active surface area to the original electrochemically active surface area.

6. The output power distribution control method according to claim 1, characterized in that When calculating the weighted sum, the weight coefficient of the deviation between the life index of the PEMFC under the current collection and the life index of the PEMFC under the previous collection takes a value greater than other weight coefficients.

7. The output power distribution control method according to claim 1, characterized in that The training samples of the adaptive fuzzy neural network adopt cyclic operating condition data.

8. An output power distribution control system for a hybrid power system, characterized in that, A method for controlling the output power distribution of a hybrid power system as described in any one of claims 1 to 7, including: a prediction unit, a collection unit, and a power distribution unit; The prediction unit is used to predict the current actual power required by the load; The collection unit is used to collect the current actual output power of the PEMFC in the hybrid power system; The attention enhancement unit is used to calculate the deviation between the current actual power required and the current actual output power, the deviation between the current actual output power and the actual output power collected last time, and the deviation between the life index of the PEMFC under the current collection and the life index of the PEMFC under the previous collection; and calculate the weighted sum of the absolute values of the deviations, where the weight coefficients have the same sign; The power distribution unit is used to control the trained adaptive fuzzy neural network to perform supply allocation on the difference between the current actual power required and the current actual output power of the PEMFC through fuzzy inference, and output the power value that the PEMFC is to output to the load. This power value is inversely proportional to the weighted sum, and the difference between the difference value and this power value is provided or received by other power sources other than the PEMFC, so that the PEMFC outputs a power value to the load that can delay the aging of the PEMFC, and at the same time the hybrid power system meets the current actual power requirement of the load.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program is run by a processor, it controls the device where the storage medium is located to execute a method for controlling the output power distribution of a hybrid power system according to any one of claims 1 to 7.

Images