CN116435557A - Fuel cell thermal management method, device and system based on neural network - Google Patents

Fuel cell thermal management method, device and system based on neural network Download PDF

Info

Publication number
CN116435557A
CN116435557A CN202310576482.4A CN202310576482A CN116435557A CN 116435557 A CN116435557 A CN 116435557A CN 202310576482 A CN202310576482 A CN 202310576482A CN 116435557 A CN116435557 A CN 116435557A
Authority
CN
China
Prior art keywords
neural network
fuel cell
data set
cell stack
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310576482.4A
Other languages
Chinese (zh)
Inventor
甘振宁
杜进桥
田杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Power Supply Bureau Co Ltd
Original Assignee
Shenzhen Power Supply Bureau Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Power Supply Bureau Co Ltd filed Critical Shenzhen Power Supply Bureau Co Ltd
Priority to CN202310576482.4A priority Critical patent/CN116435557A/en
Publication of CN116435557A publication Critical patent/CN116435557A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
    • H01M8/04298—Processes for controlling fuel cells or fuel cell systems
    • H01M8/04992—Processes for controlling fuel cells or fuel cell systems characterised by the implementation of mathematical or computational algorithms, e.g. feedback control loops, fuzzy logic, neural networks or artificial intelligence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/27—Regression, e.g. linear or logistic regression
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
    • H01M8/04007—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids related to heat exchange
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
    • H01M8/04298—Processes for controlling fuel cells or fuel cell systems
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
    • H01M8/00—Fuel cells; Manufacture thereof
    • H01M8/24—Grouping of fuel cells, e.g. stacking of fuel cells
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02E60/30—Hydrogen technology
    • Y02E60/50—Fuel cells

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Chemical & Material Sciences (AREA)
  • Electrochemistry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Manufacturing & Machinery (AREA)
  • Sustainable Development (AREA)
  • Sustainable Energy (AREA)
  • Chemical & Material Sciences (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Automation & Control Theory (AREA)
  • Medical Informatics (AREA)
  • Fuzzy Systems (AREA)
  • Fuel Cell (AREA)

Abstract

The invention provides a fuel cell thermal management method based on a neural network, which comprises the following steps: operating a fuel cell stack for testing on a test bench, obtaining an initial data set according to a time sequence, and forming a training data set and a testing data set; establishing an echo state neural network regression model, and training by adopting a training data set to obtain an echo state neural network prediction model; testing under a test data set to verify the model prediction accuracy and obtain a trained neural network prediction model; and acquiring data of the fuel cell stack in an operating state in real time, obtaining a prediction signal by using the trained neural network prediction model, and controlling the operating states of the cooling water pump and the radiator according to the prediction signal. The invention also discloses a corresponding device and a corresponding system. By implementing the invention, the accuracy and efficiency of the thermal management of the fuel cell can be improved.

Description

Fuel cell thermal management method, device and system based on neural network
Technical Field
The invention relates to the technical field of thermal management of proton exchange membrane fuel cells (Proton Exchange Membrane Fuel Cell, PEMFCs), in particular to a fuel cell thermal management method based on a neural network.
Background
The hydrogen energy is a secondary energy source which is rich in source, green, low in carbon and wide in application, and has important significance for constructing a clean, low-carbon, safe and efficient energy system and realizing the carbon-peak carbon neutralization target. The hydrogen fuel cell automobile is widely focused on the advantages of high efficiency, cleanness and the like, and the proton exchange membrane fuel cell (Proton Exchange Membrane Fuel Cell, PEMFC) has the advantages of high energy conversion efficiency, low-temperature operation, high reliability, zero emission and the like, and has very wide application prospect in the field of new energy automobiles.
PEMFCs are devices that convert chemical energy into electrical energy, and the main components thereof include a proton exchange membrane, an anode, a cathode, and a bipolar plate. The PEMFC is a nonlinear complex system with multiple physical fields and multiple parameter coupling, the working temperature is a key factor influencing the output performance and the service life, and the excessive working temperature can evaporate liquid water to cause a film dry fault; too low a temperature can cause flooding of the cathode channels and oxygen cannot pass through the gas diffusion layer. The normal operating temperature range of the PEMFC is 60-80 ℃, but a large amount of heat is generated during the operation thereof, so that effective thermal management of the PEMFC is required. Improper thermal management can cause irreversible decline of the output voltage of the PEMFC, and accelerate the aging speed of the PEMFC.
The main methods of PEMFC thermal management at present are as follows: the cooling water flow rate is controlled on the temperature model, and the control method mainly comprises PID control, state feedback control, model prediction control, fuzzy control and the like. However, the control methods have the advantages of simple principle, convenient use, low response speed, long adjustment time and the like; the inherent nonlinear characteristic of the fuel cell, the uncertainty of parameters, and the characteristics of sensitive temperature change and large change amplitude of the high-power fuel cell stack for vehicles make the application of the control methods have certain difficulty.
With the intensive research, new control methods are presented, for example, a learner designs a fuzzy control method to be applied to PEMFC thermal management, and controls the temperature of the PEMFC by adjusting the rotation speed of a fan; an improved particle swarm optimization fuzzy PID fuel cell temperature control method is adopted, improved particle swarm optimization fuzzy PID control is utilized, and a control strategy is set according to a control experience rule; in addition, a model reference self-adaptive control method is designed, and the temperature of the fuel cell stack and the temperature of the circulating cooling liquid inlet are controlled by adjusting the mass flow of the cooling liquid and the opening coefficient of the bypass valve. In addition, a learner considers the defect of a single control method, and discloses a proton exchange membrane fuel cell temperature control method, wherein a PID controller is used as a general feedback controller to stabilize the temperature of a fuel cell stack, a preliminary control effect is achieved, then the variation trend of the temperature of a fuel cell system is obtained according to the advanced prediction function of a gray model, and a fuzzy controller compensates the uncertainty and the external interference of the system by using the obtained prediction information, so that the accuracy of the system control is further improved.
However, in practical application, especially in industrial process control, due to factors such as serious nonlinearity of a controlled object, uncertainty of a mathematical model, severe change of a system working point and the like, the existing control theory-based control theory has serious defects which are difficult to compensate, and the application effectiveness of the control theory is greatly limited; most of the methods adopt a step load signal mode to verify the control method, however, the hydrogen fuel cell automobile has processes of acceleration, uniform speed, deceleration and the like in actual running, the frequent change of working conditions can make the temperature control of the fuel cell more complex, the temperature can have extreme values, and the control theory-based method at the present stage has the following defects: the accuracy of controlling the temperature to the target value is insufficient, and the fluctuation is large at the control target temperature value.
Disclosure of Invention
The invention aims to solve the technical problem of providing a fuel cell thermal management method, a device and a system based on a neural network, which can efficiently and accurately stabilize the inlet and outlet temperature of a fuel cell stack of a PEMFC at a target temperature value.
The technical scheme adopted by the invention is that the invention provides a fuel cell thermal management method based on a neural network, which at least comprises the following steps:
step S10, operating a fuel cell stack for testing on a test bench, obtaining an initial data set according to a time sequence, and forming a training data set and a testing data set;
step S11, an echo state neural network (echo state network, ESN) regression model is established;
step S12, training the established echo state neural network regression model according to the training data set to obtain an echo state neural network prediction model;
step S13, testing the trained echo state neural network prediction model under a test data set to test model prediction accuracy and obtain a trained neural network prediction model;
step S14, collecting data of the fuel cell stack in an operating state in real time, and obtaining a prediction signal by using the trained neural network prediction model, wherein the prediction signal comprises: and controlling the running states of the cooling water pump and the radiator according to the prediction signals, and controlling the temperature of the fuel cell stack at a reasonable level.
Preferably, the step S10 further includes:
operating a fuel cell stack for testing on a test bench, and obtaining fuel cell target temperature, reaction gas humidity, reaction gas pressure, reaction gas flow rate, fuel cell current, cooling water flow and cooling air volume data at each time sequence through a plurality of sensors to form an initial data set;
preprocessing the initial data set, wherein the preprocessing comprises normalization and normalization;
the preprocessed initial data set is divided into a training data set and a test data set.
Preferably, the step S11 at least includes:
establishing an echo state neural network regression model, which is provided with an input layer, a reserve tank and an output layer; wherein the input of the input layer is time series data, and the output layer outputs another time series data related to the input;
initializing the echo state neural network regression model, comprising:
determining the number of neuron nodes of the reserve pool;
randomly generating an internal connection matrix W res The internal connection matrix W res Representing the connection status, including the direction and weight of the connection, between each neuron in the reservoir;
the weight size of the input connection matrix of the input layer and the spectrum radius of the internal connection matrix are determined.
At the end of the initialization process, the weight size of the input connection and the spectral radius of the internal connection matrix need to be set.
Preferably, the step S12 further includes:
preheating the initialized echo state neural network regression model to form an internal state with time sequence characteristics;
selecting data of a training data set, taking a target temperature of a fuel cell, a temperature of a reaction gas, a humidity of the reaction gas, a pressure of the reaction gas, a flow rate of the reaction gas and a current of the fuel cell as input, and taking a cooling water flow and a cooling air quantity as output according to a time sequence, and training the echo state neural network regression model;
when the training result of the echo state neural network regression model is that the fitting is not performed, the accuracy is high, that is, the root mean square error RMSE between the predicted value and the expected value is close to 0 and the fitting goodness R is satisfied 2 When the condition is close to 1, training is finished, and echo is obtainedA state neural network prediction model.
Preferably, the training the echo state neural network regression model specifically includes:
in the sampling phase, input samples need to be input into a reservoir of ESNs to produce the corresponding internal states; and after the sampling phase is completed, a set of reservoir states corresponding to each input sample can be obtained.
In the weight calculation stage, a linear regression method is used to calculate an output weight matrix of ESNs to map the internal states to the desired outputs.
Preferably, the weight calculation stage further includes:
determining an association between the interior of the reservoir and input at a previous time:
x(t+1)＝f[W res ·x(t)+W LR ·u(t)]
wherein W is LR The input connection weight matrix is used for representing the mapping relation input to the reserve pool; w (W) RO The output connection weight matrix is used for representing the mapping relation from the reserve pool to the output; f is an activation function, W res The internal connection weight matrix is used for representing the connection relation of each element in the vector x (t) of the reserve pool;
based on the system state matrix and the sample data collected during the sampling phase, an output connection weight matrix W is calculated RO The method comprises the steps of carrying out a first treatment on the surface of the Comprising the following steps:
the following objective function is obtained:
Figure BDA0004244460980000041
wherein, the state variable x (t) and the predicted output y (t) are in linear relation, y (t) is the predicted output,
Figure BDA0004244460980000042
is the desired output;
when the minimum mean square error of the system is satisfied, calculating to obtain an output connection weight matrix W RO ：
Figure BDA0004244460980000043
Wherein W is RO X (t) is the network output.
Preferably, in the step S13, the test data is the same as the training data in dimension, and is input in time series;
and further comprises: and optimizing parameters of the fuel cell stack, and optimizing the pool scale, the pool spectrum radius, the pool sparseness degree and the input unit scale parameters to meet the required fitting precision.
Preferably, the fuel cell stack is a fuel cell stack employing proton exchange membrane fuel cells.
Accordingly, as another aspect of the present invention, there is also provided a fuel cell thermal management device based on a neural network, including at least:
a data set acquisition unit for running the test fuel cell stack on the test bench, acquiring an initial data set according to time sequence, and forming a training data set and a test data set;
the regression model acquisition unit is used for establishing an echo state neural network regression model;
the training processing unit is used for training the established echo state neural network regression model according to the training data set to obtain an echo state neural network prediction model;
the test processing unit is used for testing the trained echo state neural network prediction model under a test data set so as to test the model prediction precision and obtain a trained neural network prediction model;
the prediction processing unit is used for collecting data of the fuel cell stack in an operating state in real time, and obtaining a prediction signal by using the trained neural network prediction model, wherein the prediction signal comprises the following components: and controlling the running states of the cooling water pump and the radiator according to the prediction signals, and controlling the temperature of the fuel cell stack at a reasonable level.
Accordingly, as still another aspect of the present invention, there is also provided a neural network-based fuel cell thermal management system, including: the fuel cell stack, the water tank, the water pump and the radiator are circularly connected;
a temperature sensor disposed at the inlet and outlet of the fuel cell stack;
and the controller is connected with the fuel cell stack, the water pump, the temperature sensor and the radiator, and is provided with an operation unit for executing the fuel cell thermal management method based on the neural network.
The embodiment of the invention has the following beneficial effects:
the invention provides a fuel cell thermal management method, a device and a system based on a neural network, which aim to reduce the error between the inlet and outlet temperature and the target temperature value of a fuel cell stack by improving the thermal management precision of the fuel cell and apply an echo state neural network to the field of thermal management of the fuel cell. And training the echo state neural network through a data set obtained by testing the fuel cell stack, controlling cooling water flow and cooling air quantity signals by a trained prediction model, and finally completing the optimization process of the thermal management of the fuel cell system. The accuracy and efficiency of the temperature control of the fuel cell stack are improved;
in this embodiment, since the hidden layer of the echo state neural network is a dynamic reserve pool structure, the echo state attribute is provided, so that not only is the stability of network prediction enhanced, but also the network output weight is obtained by using a linear algorithm, the training process is simplified, and meanwhile, the problem that the convergence speed of the traditional neural network is low and the local minimum is easy to fall into is solved.
Meanwhile, the regression model obtained through training has obvious improvement effect on stability and accuracy of temperature control compared with a method based on a control theory, has better temperature adjustment capability compared with a traditional neural network, can better resist disturbance of external load, and has smaller deviation from a set value.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a main flow chart of one embodiment of a neural network-based fuel cell thermal management method provided by the present invention;
FIG. 2 is a schematic view of an application environment of the present invention;
FIG. 3 is a schematic diagram of the overall principle of construction, training and application of the echo state neural network according to the present invention;
fig. 4 is a schematic diagram of a topology of an echo state neural network according to the present invention;
FIG. 5 is a schematic block diagram of the thermal management of a fuel cell stack in accordance with the present invention;
FIG. 6 is a schematic diagram of a more detailed flow scheme of the method according to the present invention;
fig. 7 is a main flow chart of an embodiment of a fuel cell thermal management device based on a neural network according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific embodiments in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
FIG. 1 is a main flow chart of one embodiment of a fuel cell thermal management method based on a neural network according to the present invention; as shown in fig. 2 to 7, in this embodiment, the method at least includes the following steps:
step S10, operating a fuel cell stack for testing on a test bench, obtaining an initial data set according to a time sequence, and forming a training data set and a testing data set;
in this embodiment, a test environment is built on a test bench according to the architecture of fig. 2, where the test environment includes at least: the fuel cell stack, the water tank, the water pump and the radiator are circularly connected, and the temperature sensor (and other sensors) are arranged at the inlet and the outlet of the fuel cell stack; and a controller connected with the components. The fuel cell stack is a fuel cell stack adopting PEMFC, heat generated by the fuel cell stack is firstly brought to the water tank by the cooling water pump through controlling the flow of cooling water, then the heat is brought to the radiator, and the radiator discharges the heat into the air through controlling the air quantity of the radiator.
In the embodiment of the invention, it is assumed that the temperature in the cooling water is uniform, and the fuel cell stack outlet cooling water temperature is taken as the temperature of the fuel cell stack. In view of the fact that the cooling water pump and the radiator cannot be started and stopped frequently in practical applications, in this embodiment, it is necessary to set minimum values of the cooling water flow rate and the cooling air volume.
More specifically, in this embodiment, for a tested fuel cell stack, basic parameters such as the number of cells, the reaction area of a single cell, and the like are determined, the fuel cell stack is tested under a circulation condition, and physical parameters are managed during the test;
wherein, in the test bench, each physical parameter needs to be controlled within the parameter range of the following table 1;
table 1 physical parameters of test stand
Parameters (parameters) Control range
Operating temperature 20–80℃
Gas temperature 20–80℃
Humidity of gas 0-100% (relative humidity)
Air velocity 0–100L/min
Hydrogen rate 0–30L/min
Gas pressure 0–2Bar
Electric current 0–300A
In a specific example, the step S10 further includes:
operating a fuel cell stack for testing on a test bench, and obtaining fuel cell target temperature, reaction gas humidity, reaction gas pressure, reaction gas flow rate, fuel cell current, cooling water flow and cooling air volume data at each time sequence through a plurality of sensors to form an initial data set;
preprocessing the initial data set, wherein the preprocessing comprises normalization and normalization;
the preprocessed initial data set is divided into a training data set and a test data set.
Step S11, establishing an echo state neural network regression model;
it can be understood that the echo state neural network regression model is constructed by adopting an echo state neural network, which is also called reservoir calculation, and is commonly used for solving the problem of time sequence prediction. The network belongs to a recurrent neural network, and a reserve pool formed by neurons obtained by an internal weight matrix connected in a random sparse mode is used as a hidden layer to represent input in a high-dimensional and nonlinear mode, and has better nonlinear learning capability. Hidden layer weights of the echo state neural network are generated in advance instead of training, the hidden layer weights and the weight training from the hidden layer to the output layer are carried out separately, after the neural network structure is fixed, the input weight matrix and the recursion weight matrix are not changed, and only the output weight matrix is subjected to linear regression optimization, so that the calculation efficiency of the echo state neural network is greatly improved. The general principles of construction, training and application of the echo-state neural network according to the present invention may be seen with reference to fig. 3.
In a specific example, the step S11 includes at least:
establishing an echo state neural network regression model, as shown in fig. 4, which is provided with an input layer, a reserve tank and an output layer; wherein the input of the input layer is time series data, and the output layer outputs another time series data related to the input; in this embodiment, the inputs are fuel cell target temperature, reactant gas humidity, reactant gas pressure, reactant gas flow rate, fuel cell current; the output is cooling water flow and cooling air quantity;
initializing the echo state neural network regression model, comprising:
determining the number (scale) of neuronal nodes of the reservoir; it can be appreciated that the more ganglion points in the reservoir, the more fitting capacity, but the more computational effort; in this embodiment, the number of neuronal nodes of the reservoir may be 10 times the total amount of data collected by the sensor.
Randomly generating an internal connection matrix W res The internal connection matrix W res Representing the connection status, including the direction and weight of the connection, between each neuron in the reservoir;
the weight size of the input connection matrix of the input layer and the spectral radius (spectral radius) of the internal connection matrix are determined.
It can be understood that in the initialization process, the weight of the input connection weight matrix corresponding to the input layer needs to be set, and the weight of the input connection determines the influence of the transmission of the input signal in the network. If the weights are too large, the input signal will likely have too much effect on the output, resulting in an overfitting of the model; if the weight is too small, important features of the input signal may not be captured. Therefore, the selection of the weight value is required to be moderate; for the weight of the input connection, the weight is initialized by random numbers, and then normalized to ensure that the characteristics of the input signal can be fully utilized.
The spectral radius of the internal connection matrix is an index for measuring the stability of the matrix, and represents the maximum eigenvalue of the matrix. The larger the spectral radius, the higher the degree of nonlinearity of the network, but the network may oscillate and diverge. Therefore, the spectral radius needs to be reasonably selected to ensure the stability and generalization capability of the network. In the initialization of the echo state neural network, the weight size of the input connection and the spectral radius of the internal connection matrix need to be set appropriately. For the spectrum radius of the internal connection matrix, a reasonable value can be determined by a trial and error method or an adjustment algorithm so as to ensure the stability and performance of the network. Common algorithms for adjusting the radius of the spectrum are eigenvalue-decomposition (eigenvalue decomposition) and power iteration (power iteration), etc.
Step S12, training the established echo state neural network regression model according to the training data set to obtain an echo state neural network prediction model;
in a specific example, the step S12 further includes:
step S120, preheating the initialized echo state neural network regression model to form an internal state with time sequence characteristics;
step S121, selecting data of a training data set, and training the echo state neural network regression model by taking a target temperature of a fuel cell, a temperature of a reaction gas, a humidity of the reaction gas, a pressure of the reaction gas, a flow rate of the reaction gas and a current of the fuel cell as input and a flow rate of cooling water and a cooling air volume as output according to a time sequence;
more specifically, the process of training the echo state neural network regression model includes a sampling stage and a weight calculation stage, and specifically includes:
in the sampling phase, input samples need to be input into a reservoir of ESNs to produce the corresponding internal states; and after the sampling phase is completed, a set of reservoir states corresponding to each input sample can be obtained.
For example, first, an initial state of the network is arbitrarily selected, and in general, the initial state of the selected network is 0, that is, x (0) =0. In the present embodiment, it is assumed that W OB At 0, the input-to-output and output-to-output connection weights are also assumed to be 0.
Given an input signal sequence:
u(0)，u(1)，...，u(P)
training samples (u (t), t=1, 2,..p.) go through the input connection weight matrix W LR Is added to the holding tank.
And sequentially completing the calculation and collection of the system state and the output y (t):
y(0)，y(1)，...，y(P)
in order to calculate the output connection weight matrix, it is necessary to sample the internal state variables from a certain time m and to use the vector (x 1 (i),x 2 (i)，...x N (i) (i=m, m+1..p.) the matrix B (P-m+1, n) is formed for the rows, while the corresponding sample data y (T) is also collected and forms a column vector T (P-m+1, 1).
In the weight calculation stage, a linear regression method is used to calculate an output weight matrix of ESNs to map the internal states to the desired outputs.
In a specific example, the weight calculation stage further includes:
determining an association between the interior of the reservoir and input at a previous time:
x(t+1)＝f[W res ·x(t)+W LR ·u(t)]
wherein W is LR The input connection weight matrix is used for representing the mapping relation input to the reserve pool; w (W) RO The output connection weight matrix is used for representing the mapping relation from the reserve pool to the output; f is an activation function, W res The internal connection weight matrix is used for representing the connection relation of each element in the vector x (t) of the reserve pool;
based on the system state matrix and the sample data collected during the sampling phase, an output connection weight matrix W is calculated RO The method comprises the steps of carrying out a first treatment on the surface of the Comprising the following steps:
the following objective function is obtained:
Figure BDA0004244460980000101
wherein, the state variable x (t) and the predicted output y (t) are in linear relation, y (t) is the predicted output,
Figure BDA0004244460980000103
is the desired output;
when the system mean square error is minimum, namely the expected target is reached, calculating to obtain the output connection weight matrix W at the moment RO ：
Figure BDA0004244460980000102
Wherein W is RO X (t) is the network output.
It will be appreciated that for making predictions in terms of time series, the output is typically re-used as input, thereby enabling a continuous backward prediction.
It will be appreciated that in one example, during training, where t=m, m=1, …, P represents the output of the first P time steps during training is used to train the output layer weight matrix. The time step here refers to different states in which the regression state neural network receives input data at different points in time and processes the input data.
Specifically, during the training process, the regression state neural network is preheated by the initial state and the input data, and some internal state values are generated during the process. The warm-up time is typically not used to train the output layer weight matrix.
After the warm-up process, the E regression state neural network begins to produce predicted output results and trains the output layer weight matrix based on the top P predicted output results and the actual output results in the training set. In this process, t=m, m=1, …, P represents that the prediction outputs of the first step (m=1) to the P-th step (m=p) from the beginning of the generation of the prediction output by the network are used to train the output layer weight matrix.
The training mode can enable the regression state neural network to learn and predict the predicted output results of the previous P time steps only, and discard the predicted output results of the subsequent time steps. Therefore, the influence of the past noise data on the network can be reduced, and the prediction accuracy and the robustness of the network are improved. Meanwhile, training only the output results of the first P time steps can also improve the training efficiency of the network. It should also be noted that in practical applications, the size of P needs to be carefully chosen to balance the stability and training effect of the network.
Step S122, calculating the root mean square error RMSE and the goodness of fit R-square (R) of the output features from the test dataset during the training test 2 )。
When the training result of the echo state neural network regression model is that the fitting is not performed and the accuracy is high, that is, the root mean square error RMSE between the predicted value and the expected value is close to 0 and the fitting goodness R-square (R 2 ) And when the condition is close to 1, training is finished, and an echo state neural network prediction model is obtained. After training the data in the training set in the process, the obtained echo state neural network prediction model can be used for modeling the specific problem of the time sequence.
Step S13, testing the trained echo state neural network prediction model under a test data set to test model prediction accuracy and obtain a trained neural network prediction model;
more specifically, in the step S13, the test data is identical in dimension to the training data and is input in time series;
and further comprises: and optimizing parameters of the fuel cell stack, and optimizing the pool scale, the pool spectrum radius, the pool sparseness degree and the input unit scale parameters to meet the required fitting precision.
Step S14, collecting data of the fuel cell stack in an operating state in real time, setting a target temperature value of an inlet and an outlet of the fuel cell stack, and obtaining a prediction signal by using the trained neural network prediction model, wherein the prediction signal comprises: and controlling the running states of the cooling water pump and the radiator according to the prediction signals, and controlling the temperature of the fuel cell stack at a reasonable level. It will be appreciated that the environment in which the fuel cell stack is operating is the same as that shown in fig. 2. The fuel cell stack may be, for example, applied in a vehicle.
It can be understood that in the method provided by the invention, according to the trained echo state neural network model, the predicted cooling water flow and cooling air quantity signals; and controlling the cooling water pump and the radiator fan according to the signals, further performing thermal management on the fuel cell stack, and finally completing control on the temperature of the cooling water inlet of the fuel cell stack and the temperature of the cooling water outlet of the fuel cell stack.
In the method provided by the invention, improvement is carried out on improving the thermal management precision of the fuel cell, and the echo state neural network is applied to the field of thermal management of the fuel cell by taking the smaller error between the inlet and outlet temperature of the fuel cell stack and the target temperature value as the target. And training the echo state neural network through a data set obtained by testing the fuel cell stack, controlling cooling water flow and cooling air quantity signals by a trained prediction model, and finally completing the optimization process of the thermal management of the fuel cell system.
Meanwhile, in the embodiment, the echo state neural network model based on deep learning is suitable for controlling a nonlinear system, a specific mathematical model is not needed, and situations such as severe change of a system working point can be dealt with; the optimized controller can better resist the change of external load, so that the error between the obtained inlet temperature and the target temperature value is smaller, the controller can be effectively applied to the thermal management of the fuel cell of the high-power hybrid electric vehicle, and the controller has more advantages in accuracy and stability.
As shown in fig. 7, a schematic structural diagram of an embodiment of a fuel cell thermal management device based on a neural network according to the present invention is shown. In this embodiment, the neural network-based fuel cell thermal management device 1 includes at least:
a data set acquisition unit 10 for running the test fuel cell stack on a test bench, acquiring an initial data set in time series, and forming a training data set and a test data set;
a regression model obtaining unit 11, configured to establish an echo state neural network regression model;
the training processing unit 12 is configured to train the established echo state neural network regression model according to the training data set, and obtain an echo state neural network prediction model;
the test processing unit 13 is configured to test the trained echo state neural network prediction model under a test data set to test the model prediction accuracy, so as to obtain a trained neural network prediction model;
the prediction processing unit 14 is configured to collect data of a fuel cell stack in an operating state in real time, set a target temperature value of an inlet and an outlet of the fuel cell stack, and obtain a prediction signal by using the trained neural network prediction model, where the prediction signal includes: and controlling the running states of the cooling water pump and the radiator according to the prediction signals, and controlling the temperature of the fuel cell stack at a reasonable level.
For more details, reference is made to the foregoing descriptions of fig. 1 to 6, and details are not repeated here.
Accordingly, as still another aspect of the present invention, there is also provided a neural network-based fuel cell thermal management system, the structure of which can be shown with reference to fig. 2, the system comprising at least: the fuel cell stack, the water tank, the water pump and the radiator are circularly connected;
a temperature sensor disposed at the inlet and outlet of the fuel cell stack;
and a controller connected with the fuel cell stack, the water pump, the temperature sensor and the radiator, wherein an operation unit is arranged in the controller and is used for executing the fuel cell thermal management method based on the neural network as described in the previous figures 1 to 6.
For more details, reference is made to the foregoing descriptions of fig. 1 to 6, and details are not repeated here.
The embodiment of the invention has the following beneficial effects:
the invention provides a fuel cell thermal management method, a device and a system based on a neural network, which aim to reduce the error between the inlet and outlet temperature and the target temperature value of a fuel cell stack by improving the thermal management precision of the fuel cell and apply an echo state neural network to the field of thermal management of the fuel cell. And training the echo state neural network through a data set obtained by testing the fuel cell stack, controlling cooling water flow and cooling air quantity signals by a trained prediction model, and finally completing the optimization process of the thermal management of the fuel cell system. The accuracy and efficiency of the temperature control of the fuel cell stack are improved;
in this embodiment, since the hidden layer of the echo state neural network is a dynamic reserve pool structure, the echo state attribute is provided, so that not only is the stability of network prediction enhanced, but also the network output weight is obtained by using a linear algorithm, the training process is simplified, and meanwhile, the problem that the convergence speed of the traditional neural network is low and the local minimum is easy to fall into is solved.
Meanwhile, the regression model obtained through training has obvious improvement effect on stability and accuracy of temperature control compared with a method based on a control theory, has better temperature adjustment capability compared with a traditional neural network, can better resist disturbance of external load, and has smaller deviation from a set value.
It will be apparent to those skilled in the art that embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the modules specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the foregoing description is only illustrative of the preferred embodiments of the present invention, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements or changes may be made without departing from the spirit and principles of the present invention.

Claims (10)

1. A fuel cell thermal management method based on a neural network, comprising at least the steps of:
step S10, operating a fuel cell stack for testing on a test bench, obtaining an initial data set according to a time sequence, and forming a training data set and a testing data set;
step S11, establishing an echo state neural network regression model;
step S12, training the established echo state neural network regression model according to the training data set to obtain an echo state neural network prediction model;
step S13, testing the trained echo state neural network prediction model under a test data set to test model prediction accuracy and obtain a trained neural network prediction model;
and S14, acquiring data of the fuel cell stack in an operating state in real time, obtaining a prediction signal by using the trained neural network prediction model, and controlling the operating states of the cooling water pump and the radiator according to the prediction signal to control the temperature of the fuel cell stack.
2. The method of claim 1, wherein the step S10 further comprises:
operating a fuel cell stack for testing on a test bench, and obtaining fuel cell target temperature, reaction gas humidity, reaction gas pressure, reaction gas flow rate, fuel cell current, cooling water flow and cooling air volume data at each time sequence through a plurality of sensors to form an initial data set;
preprocessing the initial data set, wherein the preprocessing comprises normalization and normalization;
the preprocessed initial data set is divided into a training data set and a test data set.
3. The method according to claim 2, wherein the step S11 includes at least:
establishing an echo state neural network regression model, which is provided with an input layer, a reserve tank and an output layer; wherein the input of the input layer is time series data, and the output layer outputs another time series data related to the input;
initializing the echo state neural network regression model, comprising:
determining the number of neuron nodes of the reserve pool;
randomly generating an internal connection matrix, wherein the internal connection matrix represents the connection state between each neuron in the reserve pool and comprises the connection direction and the weight value;
the weight size of the input connection matrix of the input layer and the spectrum radius of the internal connection matrix are determined.
At the end of the initialization process, the weight size of the input connection and the spectral radius of the internal connection matrix need to be set.
4. The method of claim 3, wherein said step S12 further comprises:
preheating the initialized echo state neural network regression model to form an internal state with time sequence characteristics;
selecting data of a training data set, taking a target temperature of a fuel cell, a temperature of a reaction gas, a humidity of the reaction gas, a pressure of the reaction gas, a flow rate of the reaction gas and a current of the fuel cell as input, and taking a cooling water flow and a cooling air quantity as output according to a time sequence, and training the echo state neural network regression model;
when the training result of the echo state neural network regression model is that fitting is not performed, and high accuracy is presented, namely, the condition that the root mean square error between the predicted value and the expected value is close to 0 and the fitting goodness is close to 1 is satisfied, training is finished, and the echo state neural network prediction model is obtained.
5. The method of claim 4, wherein training the echo state neural network regression model specifically comprises:
in the sampling phase, input samples need to be input into a reservoir of ESNs to produce the corresponding internal states; and after the sampling phase is completed, a set of reservoir states corresponding to each input sample can be obtained.
In the weight calculation stage, a linear regression method is used to calculate an output weight matrix of ESNs to map the internal states to the desired outputs.
6. The method of claim 5, further comprising, at the weight calculation stage:
determining an association between the interior of the reservoir and input at a previous time:
x(t+1)＝f[W res ·x(t)+W IR ·u(t)]
wherein W is LR The input connection weight matrix is used for representing the mapping relation input to the reserve pool; w (W) RO The output connection weight matrix is used for representing the mapping relation from the reserve pool to the output; f is an activation function, W res The internal connection weight matrix is used for representing the connection relation of each neuron in the reserve pool;
based on the system state matrix and the sample data collected during the sampling phase, an output connection weight matrix W is calculated RO The method comprises the steps of carrying out a first treatment on the surface of the Comprising the following steps:
the following objective function is obtained:
Figure FDA0004244460960000031
wherein, the state variable x (t) and the predicted output y (t) are in linear relation, y (t) is the predicted output,
Figure FDA0004244460960000033
is the desired output;
when the minimum mean square error of the system is satisfied, calculating to obtain an output connection weight matrix W RO ：
Figure FDA0004244460960000032
Wherein W is RO X (t) is the network output.
7. The method of claim 4, wherein in the step S13, the test data is the same as training data in dimension and is input in time series;
and further comprises: and optimizing parameters of the fuel cell stack, and optimizing the pool scale, the pool spectrum radius, the pool sparseness degree and the input unit scale parameters to meet the required fitting precision.
8. The method of any one of claims 1 to 7, wherein the fuel cell stack is a fuel cell stack employing proton exchange membrane fuel cells.
9. A fuel cell thermal management device based on a neural network, comprising at least:
a data set acquisition unit for operating the fuel cell stack for test on the test bench, acquiring an initial data set according to time sequence, and forming a training data set and a test data set;
the regression model acquisition unit is used for establishing an echo state neural network regression model;
the training processing unit is used for training the established echo state neural network regression model according to the training data set to obtain an echo state neural network prediction model;
the test processing unit is used for testing the trained echo state neural network prediction model under a test data set so as to test the model prediction precision and obtain a trained neural network prediction model;
and the prediction processing unit is used for acquiring data of the fuel cell stack in the running state in real time, obtaining a prediction signal by using the trained neural network prediction model, and controlling the running states of the cooling water pump and the radiator according to the prediction signal so as to control the temperature of the fuel cell stack.
10. A neural network-based fuel cell thermal management system, comprising: the fuel cell stack, the water tank, the water pump and the radiator are circularly connected;
a temperature sensor disposed at the inlet and outlet of the fuel cell stack;
a controller connected to the fuel cell stack, the water pump, the temperature sensor, and the radiator, the controller having an operation unit provided therein for performing the neural network-based fuel cell thermal management method according to any one of claims 1 to 8.
CN202310576482.4A 2023-05-22 2023-05-22 Fuel cell thermal management method, device and system based on neural network Pending CN116435557A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310576482.4A CN116435557A (en) 2023-05-22 2023-05-22 Fuel cell thermal management method, device and system based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310576482.4A CN116435557A (en) 2023-05-22 2023-05-22 Fuel cell thermal management method, device and system based on neural network

Publications (1)

Publication Number Publication Date
CN116435557A true CN116435557A (en) 2023-07-14

Family

ID=87087436

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310576482.4A Pending CN116435557A (en) 2023-05-22 2023-05-22 Fuel cell thermal management method, device and system based on neural network

Country Status (1)

Country Link
CN (1) CN116435557A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116995276A (en) * 2023-09-27 2023-11-03 爱德曼氢能源装备有限公司 Cooling method and system for fuel cell power generation system
CN117217031A (en) * 2023-11-09 2023-12-12 新研氢能源科技有限公司 Intelligent production method and system for fuel cell stack

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116995276A (en) * 2023-09-27 2023-11-03 爱德曼氢能源装备有限公司 Cooling method and system for fuel cell power generation system
CN116995276B (en) * 2023-09-27 2023-12-29 爱德曼氢能源装备有限公司 Cooling method and system for fuel cell power generation system
CN117217031A (en) * 2023-11-09 2023-12-12 新研氢能源科技有限公司 Intelligent production method and system for fuel cell stack
CN117217031B (en) * 2023-11-09 2024-02-20 新研氢能源科技有限公司 Intelligent production method and system for fuel cell stack

Similar Documents

Publication Publication Date Title
CN116435557A (en) Fuel cell thermal management method, device and system based on neural network
CN108134114B (en) Proton exchange membrane fuel cell temperature control method
Zheng et al. A double-fuzzy diagnostic methodology dedicated to online fault diagnosis of proton exchange membrane fuel cell stacks
CN108897286B (en) Fault detection method based on distributed nonlinear dynamic relation model
Zheng et al. Performance prediction of fuel cells using long short‐term memory recurrent neural network
CN114447378B (en) Parameter optimization method for proton exchange membrane fuel cell
Rezazadeh et al. Using adaptive neuro-fuzzy inference system (ANFIS) for proton exchange membrane fuel cell (PEMFC) performance modeling
Hua et al. Lifespan prediction for proton exchange membrane fuel cells based on wavelet transform and echo state network
Shen et al. Nonlinear modeling and adaptive fuzzy control of MCFC stack
Miao et al. Metamodel based design optimization approach in promoting the performance of proton exchange membrane fuel cells
WO2023216150A1 (en) Thermal management method for fuel cell
Deng et al. Degradation prediction of PEMFCs using stacked echo state network based on genetic algorithm optimization
CN117031950A (en) Modeling method and device for deep peak-shaving thermal power generating unit control system
Zhang et al. Degradation prediction model of PEMFC based on multi-reservoir echo state network with mini reservoir
Fu et al. An extreme gradient boosting-based thermal management strategy for proton exchange membrane fuel cell stacks
Yuan et al. A novel feature susceptibility approach for a PEMFC control system based on an improved XGBoost-Boruta algorithm
Wang et al. Long short-term memory deep learning model for predicting the dynamic performance of automotive PEMFC system
CN116865343B (en) Model-free self-adaptive control method, device and medium for distributed photovoltaic power distribution network
Wang et al. Modeling of a 5-cell direct methanol fuel cell using adaptive-network-based fuzzy inference systems
Zou et al. Working zone for a least-squares support vector machine for modeling polymer electrolyte fuel cell voltage
CN111952646A (en) Decoupling control method and system for fuel cell air system
Yuan et al. Fault Diagnosis of Fuel Cells by a Hybrid Deep Learning Network Fusing Characteristic Impedance
CN114137829B (en) Proton exchange membrane fuel cell subspace identification method based on ALMBO optimization algorithm
CN111179110B (en) Virtual power plant variable order aggregation equivalent robust dynamic model modeling method and device
CN116306270A (en) Method for predicting fuel cell performance and optimizing key input parameters

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination