Battery health status prediction method based on feedforward neural network
By using the feedforward neural network model to process the aging characteristic parameters and pressure difference vectors when the battery pack is connected to the load, the accurate prediction of the battery health status is achieved, and the problem of disconnecting the load during detection in the prior art is solved. , improves the reliability of battery life management.
Patent Information
- Application Number
- CN202211137882.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-19
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-09-19
AI Technical Summary
The prior art requires disconnection of the load during the battery health status detection, and the accuracy of the estimation value is not high, making it difficult to accurately estimate the health status of the battery.
Using a feedforward neural network method, when the battery pack is connected to the load, aging characteristic parameters such as open circuit voltage, ambient temperature and valley voltage are obtained, characteristic parameter vectors and pressure difference vectors are constructed, and input them to the trained feedforward nerve. In the network model, the health status estimate is output to achieve accurate prediction of the battery's health status.
With the continuous battery load turned on, the accuracy of battery health status detection is improved, the battery health status can be predicted in a timely and accurate manner, and the reliability of battery life management is improved.
Smart Images

Figure CN115372852B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of storage batteries, and in particular to a method for predicting the health status of storage batteries based on a feedforward neural network. Background Art
[0002] Acid batteries are batteries that use acidic aqueous solutions as electrolytes. They have the advantages of high operating voltage, wide operating temperature, good high and low rate discharge performance, abundant raw material sources, and low prices. The most typical of these batteries is lead-acid batteries. Lead-acid batteries are widely used in uninterruptible power supply (UPS) systems, and more than 65% of UPS system power failures are directly related to lead-acid batteries. Therefore, battery-related problems such as power variations, abnormal heating, improper charging, or excessive electrolyte levels in specific types of batteries may cause technical failures in the UPS.
[0003] In order to improve the life cycle of the battery and ensure the stable operation of the UPS system, it is very necessary to estimate the battery's state of health (SOH) in a timely and accurate manner. Due to the complex electrochemical reactions inside the battery and the complex and changeable external environment, it is very difficult to accurately estimate the battery's state of health.
[0004] The existing methods for estimating the health status of batteries mainly include: electrochemical impedance spectroscopy measurement method, probability statistics method, recursive and probabilistic neural network model method; among them, the electrochemical impedance spectroscopy measurement method has inaccurate results due to influencing factors, the probability statistics method is very time-consuming to obtain data from the results, and the recursive and probabilistic neural network model method needs to disconnect the load during detection. Summary of the invention
[0005] The purpose of the present invention is to provide a battery health state prediction method based on a feedforward neural network to solve the technical problems in the prior art that the load needs to be disconnected during battery health state detection and the estimated value of the battery health state is not accurate enough.
[0006] The purpose of the present invention can be achieved through the following technical solutions:
[0007] The battery health status prediction method based on feedforward neural network includes:
[0008] When the battery pack is connected to a load, aging characteristic parameters of each battery in the battery pack are obtained, and characteristic parameter vectors of each battery are constructed according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley point voltage, and a second valley point voltage, wherein the second valley point voltage is greater than the first valley point voltage;
[0009] obtaining a first voltage difference according to the open circuit voltage and the first valley voltage, obtaining a second voltage difference according to the first valley voltage and the second valley voltage, and obtaining a voltage difference vector of each of the batteries according to the first voltage difference and the second voltage difference;
[0010] The characteristic parameter vector and the pressure difference vector are input as input data into a trained feedforward neural network model, and the health status estimation value of each battery is output; the feedforward neural network model is a neural network model trained by using a BP algorithm;
[0011] The health state of each of the batteries is predicted according to the health state estimation value of each of the batteries.
[0012] Optionally, obtaining a first voltage difference according to the open circuit voltage and the first valley voltage includes:
[0013] The first voltage difference is a difference between the open circuit voltage and the first valley voltage.
[0014] Optionally, obtaining a second voltage difference according to the first valley voltage and the second valley voltage includes: the second voltage difference is a difference between the first valley voltage and the second valley voltage.
[0015] Optionally, the characteristic parameter vector of each battery includes at least:
[0016] the open circuit voltage, the ambient temperature, the first valley voltage, and the second valley voltage.
[0017] Optionally, predicting the health state of each of the batteries according to the health state estimation value of each of the batteries includes:
[0018] When the estimated health state value of the battery is greater than a preset health state threshold, the battery is in a healthy state;
[0019] Otherwise, the battery is in an unhealthy state.
[0020] Optionally, the feedforward neural network model is:
[0021] Multi-layer feedforward neural network model, including input layer, multiple hidden layers and output layer;
[0022] The input layer receives the input characteristic parameter vector, the multiple hidden layers process the characteristic parameter vector, the output result of the previous hidden layer is used as the input vector of the next hidden layer, and the output layer outputs the estimated value of the health status of each battery.
[0023] Optionally, the training process of the feedforward neural network model includes:
[0024] Inputting the characteristic parameter matrix of the training samples in the training set into the feedforward neural network model and propagating forward through the network to calculate the network net input and network output of each neuron in the hidden layer and the output layer;
[0025] Propagate the error backward through the network and calculate the error of each neuron in the hidden layer and the output layer;
[0026] Update the weights and biases of neurons according to preset rules;
[0027] The above process is repeated until the error on the validation set is less than the preset error threshold.
[0028] Optionally, the activation function of the feedforward neural network model is:
[0029]
[0030] Among them, S is the sigmoid function and x is the input feature parameter vector.
[0031] The present invention also provides a battery health status prediction system based on a feedforward neural network, comprising:
[0032] A characteristic parameter vector construction module is used to obtain aging characteristic parameters of each battery in the battery pack when the battery pack is connected to a load, and to construct a characteristic parameter vector of each battery according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley point voltage and a second valley point voltage, wherein the second valley point voltage is greater than the first valley point voltage;
[0033] a voltage difference vector construction module, configured to obtain a first voltage difference according to the open circuit voltage and the first valley voltage, obtain a second voltage difference according to the first valley voltage and the second valley voltage, and obtain a voltage difference vector of each battery according to the first voltage difference and the second voltage difference;
[0034] A health state estimation value acquisition module, used to input the characteristic parameter vector and the pressure difference vector as input data into a trained feedforward neural network model, and output the health state estimation value of each battery; the feedforward neural network model is trained using a BP algorithm;
[0035] The health status prediction module is used to predict the health status of each of the batteries according to the health status estimation value of each of the batteries.
[0036] Optionally, the health state prediction module predicts the health state of each battery according to the health state estimation value of each battery, including:
[0037] When the estimated health state value of the battery is greater than a preset health state threshold, the battery is in a healthy state;
[0038] Otherwise, the battery is in an unhealthy state.
[0039] The present invention provides a method and system for predicting the health status of batteries based on a feedforward neural network, wherein the method comprises: when a battery pack is connected to a load, obtaining aging characteristic parameters of each battery in the battery pack, and constructing a characteristic parameter vector of each battery according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley voltage and a second valley voltage, wherein the second valley voltage is greater than the first valley voltage; obtaining a first voltage difference according to the open circuit voltage and the first valley voltage, obtaining a second voltage difference according to the first valley voltage and the second valley voltage, and obtaining a pressure difference vector of each battery according to the first voltage difference and the second voltage difference; inputting the characteristic parameter vector and the pressure difference vector as input data into a trained feedforward neural network model, and outputting a health status estimation value of each battery; the feedforward neural network model is a neural network model trained using a BP algorithm; and predicting the health status of each battery according to the health status estimation value of each battery.
[0040] In view of this, the beneficial effects brought by the present invention are:
[0041] When the battery pack is connected to a load, the aging characteristic parameters such as open circuit voltage, ambient temperature, and valley voltage are obtained. The change in open circuit voltage can reflect the change in the performance of the positive and negative electrodes of the battery, impedance, and battery energy storage capacity. The ambient temperature will affect the charging and discharging, service life, and internal resistance of the battery. The valley voltage can reflect the low voltage of the battery. The corresponding characteristic parameter vector is constructed according to the aging characteristic parameters and input into the trained feedforward neural network model. A more accurate health status estimate can be obtained. The health status of the battery can be predicted in a timely and accurate manner based on the health status estimate. Compared with the traditional cycle counting method, the present invention can improve the accuracy of battery health status detection without disconnecting the battery load. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic diagram of the process of the present invention;
[0043] Figure 2 It is a schematic diagram of the structure of the feedforward neural network of the present invention;
[0044] Figure 3 It is a schematic diagram of the three-step process of FFNN of the present invention;
[0045] Figure 4Schematic diagram of estimating SOH by FFNN of the present invention;
[0046] Figure 5 Schematic diagram of a feedforward neural network model of the present invention;
[0047] Figure 6 It is a schematic diagram of the flow of the neural network feedforward algorithm of the present invention;
[0048] Figure 7 A schematic diagram of a voltage waveform of a voltage test according to a specific example of the present invention;
[0049] Figure 8 It is a schematic diagram of OCV characteristics under a discharge test of a specific example of the present invention;
[0050] Fig. 9 It is a schematic diagram of typical influence of impedance during aging in a specific embodiment of the present invention;
[0051] Fig.10 It is a schematic diagram of experimental results of estimating SOH according to a specific example of the present invention;
[0052] Fig.11 It is a performance verification diagram of a specific example of the present invention. DETAILED DESCRIPTION
[0053] The embodiment of the present invention provides a method for predicting the health status of a battery based on a feedforward neural network to solve the technical problems in the prior art that the load needs to be disconnected during the battery health status detection and the estimated value of the battery health status is not accurate enough.
[0054] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.
[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0056] Lead-acid batteries are widely used in uninterruptible power supply (UPS) systems. Batteries are an important component in UPS systems. In order to improve the life cycle of batteries, active testing and maintenance are required. However, problems that occur in batteries are unpredictable. Abnormal temperature, frequent charging and discharging, and rotation play a vital role in accelerating battery failure. Therefore, active testing of batteries is inevitable. To avoid failures, a comprehensive battery management system that estimates the battery's state of health (SOH) is necessary.
[0057] The existing implementation schemes that are most similar to the present invention include: a measurement method based on electrochemical impedance spectroscopy, a measurement method based on a probabilistic statistical method combined with existing technology and mathematical implementation technology, and a measurement method based on a recursive neural network and a probabilistic neural network model. However, there are the following problems:
[0058] (1) Based on electrochemical impedance spectroscopy measurement method
[0059] This method uses the impedance of lead-acid batteries to monitor SOH by analyzing the relationship between the resistance and capacity of lead-acid batteries. Although the results obtained have good robustness and low computational complexity, influencing factors run through the model and produce inaccurate results.
[0060] (2) Measurement method based on probability statistics combined with existing technology and mathematical realization technology
[0061] This method uses probabilistic statistical methods to estimate the battery health state, and combines existing techniques and mathematical implementation techniques in these methods to make the system more reliable, but obtaining data from the results is very time-consuming.
[0062] (3) Measurement methods based on recurrent neural network and probabilistic neural network models
[0063] This method uses a recurrent neural network model and a probabilistic neural network model of electrochemical impedance spectroscopy for estimation. However, its results depend on the input data, it is time-consuming to obtain the results, and the load needs to be disconnected during SOH detection.
[0064] See also Figure 1 The present invention provides an embodiment of a method for evaluating the penetration rate of new energy sources in an AC / DC parallel power grid, comprising:
[0065] S100: When the battery pack is connected to a load, obtain aging characteristic parameters of each battery in the battery pack, and construct a characteristic parameter vector of each battery according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley voltage and a second valley voltage, wherein the second valley voltage is greater than the first valley voltage;
[0066] S200: obtaining a first voltage difference according to the open circuit voltage and the first valley voltage, obtaining a second voltage difference according to the first valley voltage and the second valley voltage, and obtaining a voltage difference vector of each battery according to the first voltage difference and the second voltage difference;
[0067] S300: inputting the characteristic parameter vector and the pressure difference vector as input data into a trained feedforward neural network model, and outputting an estimated value of the health state of each battery; the feedforward neural network model is a neural network model trained using a BP algorithm;
[0068] S400: Predicting the health status of each of the batteries according to the health status estimation value of each of the batteries.
[0069] See also Figure 2 , Figure 2 is a FFNN architecture with input and output, where is the interconnection weight of neuron “i” in layer “m”, and the input of m comes from neuron “j” in layer “(m-1)”. The FFNN architecture is denoted as S0, S1, ······, S m . Among them, p and n are input vectors, specifically, p is the characteristic parameter vector, n is the pressure difference vector, S m is the number of neurons in the mth layer, where m = 1, 2, 3, ······, M; S 0 =p; is the bias signal of neuron i in the mth layer; f m is the transfer function of all neurons in the mth layer, is the output of neuron i in the mth layer.
[0070] In the FFNN structure, the signal flows only in the forward direction from input to output, and each neuron model in the hidden layer may have a nonlinear function. The most commonly used nonlinear functions in the hidden layer are tan-sigmoid and log-sigmoid. The output layer of this architecture can contain a pure linear function to obtain better results. The feedforward neural network model includes: an input layer, several hidden layers, and an output layer. According to the number of hidden layers, the feedforward neural network model is divided into two types: a single-layer feedforward architecture (containing 1 hidden layer) and a multi-layer feedforward architecture (containing multiple hidden layers).
[0071] A multi-layer feedforward neural network is a fully connected network consisting of an input layer, several hidden layers, and an output layer. There are no loops or loops. Each layer consists of multiple neurons. The neurons and input vectors are fully connected, but there is no connection between neurons, and there is no cross-layer connection. The output of each layer will be used as the input vector of the next layer. For example, the output vector of hidden layer1 is the input vector of hidden layer2. The input layer does not process the data, but only represents the initial input vector of the entire neural network, that is, the feature vector of the sample. The input of the neural network is a vector, and the final output is also a vector. The dimension of the output vector depends on the number of neurons in the output layer.
[0072] It should be noted that, in this embodiment, the output layer only outputs the estimated SOH value of the battery, and the dimension of the output vector is only 1. Therefore, there is only one neuron in the output layer.
[0073] Neural network methods are a viable approach to calculating battery SOH and require a large amount of input data to train the inputs. Data moves from input X to activation function F, output Y, through the components being evaluated. There are no input relationships where the results of the model are transmitted back to itself. When FFNN is extended to provide feedback links, this type of neural network is called a recurrent neural network. FFNN is calculated as follows: the input and weight are multiplied, plus the bias signal, activation signal, and output signal.
[0074] In this embodiment, a data-driven feedforward neural network model (FFNN model) is used to estimate the health status of the battery. When the battery pack is connected to a load, the aging characteristic parameters of each battery in the battery pack are obtained. The aging characteristic parameters include at least: open circuit voltage OCV, ambient temperature T, first valley point voltage VP1 and second valley point voltage VP2, and the second valley point voltage is greater than the first valley point voltage.
[0075] Then, the first voltage difference Vd1 is obtained according to the open circuit voltage OCV and the first valley voltage VP1. Specifically, the first voltage difference is the difference between the open circuit voltage and the first valley voltage, that is, Vd1=OCV-VP1. The second voltage difference Vd2 is obtained according to the first valley voltage VP1 and the second valley voltage VP2. Specifically, the second voltage difference Vd2 is the difference between the first valley voltage VP1 and the second valley voltage VP2, that is, Vd2=VP2-VP1. It can be understood that the first voltage difference Vd1 is the voltage difference between the open circuit voltage OCV and the first valley voltage VP1, and the second voltage difference Vd2 is the voltage difference between the second valley voltage VP2 and the first valley voltage VP1. According to Vd1 and Vd2, the trend of voltage reduction can be understood.
[0076] In this embodiment, the aging characteristic parameters of the battery are evaluated in the input layer, which are open circuit voltage (OCV), temperature (T), first valley voltage (VP1) and second valley voltage 2 (VP2). Each aging characteristic parameter corresponds to a neuron in the input layer. Therefore, it can be determined that the input layer in this embodiment includes four neurons.
[0077] In this embodiment, since the change of the open circuit voltage OCV can reflect the change of the positive and negative electrode performance, impedance, and battery energy storage capacity of the battery; the ambient temperature T will affect the charging and discharging, service life and internal resistance of the battery; the first valley voltage can reflect the lowest voltage of the battery, and the second valley voltage can reflect the second lowest voltage of the battery, these aging characteristic parameters can accurately reflect the health status of the lead-acid battery. According to the aging characteristic parameters, the corresponding characteristic parameter vector is constructed, and the characteristic parameter vector is input into the trained feedforward neural network model, a more accurate health status estimate can be obtained, which can improve the accuracy of the health status detection of the lead-acid battery.
[0078] A preferred implementation manner, this embodiment adopts a multi-layer feedforward neural network model, including an input layer, multiple hidden layers and an output layer; wherein the input layer receives the characteristic parameter vector of each battery as input, the multiple hidden layers process the characteristic parameter vector, the output result of the previous hidden layer is used as the input vector of the next hidden layer, and the output layer outputs the estimated value of the health status of each battery.
[0079] The three-step process of FFNN is as follows Figure 3 As shown in the figure, a multi-layer feedforward neural network can be regarded as a nonlinear composite function. Its learning process is to forward propagate the input information X layer by layer along the direction of the network structure until the output layer, and then update the weights and biases through the back propagation algorithm (BP). Then the whole process is repeated until the error on the validation set is less than the preset error threshold.
[0080] See also Figure 4 , Figure 4 This is a schematic diagram of the process of estimating SOH using FFNN. By measuring the open circuit voltage (OCV), temperature (T), first valley voltage (VP1) and second valley voltage (VP2) of each battery in the battery pack, the characteristic parameter vector of each battery is obtained; the first voltage difference Vd1 (open circuit voltage minus first valley voltage) and the second voltage difference Vd2 (second valley voltage minus first valley voltage) are calculated based on the measured data to obtain the pressure difference vector of each battery; the characteristic parameter vector and the pressure difference vector are combined and input into the feedforward neural network model to obtain the estimated value of the battery's health state, i.e., the SOH value. In a preferred embodiment, the battery pack contains 10 batteries.
[0081] The number of neurons in the hidden layer is selected from the series of trajectory runs to obtain the ideal organization structure with the minimum error. The complete data set generated by the input parameters is divided into three subsets for the training process. The neural network uses the training subset to obtain its parameters. The validation subset is used for training supervision and to detect the occurrence of overfitting. The performance of the neural network is measured using the test subset, which is also considered as an independent data subset.
[0082] The neural network method is a feasible method for calculating the estimated value of battery SOH. It requires a large amount of input data to train the input. The advantage of the neural network is that the chemical reaction is only considered when the error estimate is less than the maximum error value determined in advance.
[0083] In this embodiment, the training process of the feedforward neural network model includes:
[0084] Inputting the characteristic parameter matrix of the training samples in the training set into the feedforward neural network model and propagating forward through the network to calculate the network net input and network output of each neuron in the hidden layer and the output layer;
[0085] Propagate the error backward through the network and calculate the error of each neuron in the hidden layer and the output layer;
[0086] Update the weights and biases of neurons according to preset rules;
[0087] The above process is repeated until the error on the validation set is less than the preset error threshold.
[0088] In this embodiment, a multi-layer feedforward structure network is used, and the parameters considered are the weights and bias terms of the neurons. The total number of multi-layer feedforward structure parameters is obtained by formula (1):
[0089]
[0090] Among them, "PMLFF" stands for "perfectly matched layer Feed-Forward", "Weights" is "weight", and "Biases" is "bias signal".
[0091] In this embodiment, the design of the neural network model is mainly based on the type of learning algorithm, and the learning algorithm is used to optimize the neural network parameters to minimize the output mean square error function. Different types of learning algorithms are used to train the neural network parameters to obtain better performance indicators. Learning algorithms are mainly divided into first-order learning algorithms and second-order learning algorithms.
[0092] The steep descent method is a first-order approximation algorithm developed from the most commonly used first-order approximation algorithms such as the error back propagation algorithm (BP) and the variable learning rate algorithm (VLR). The BP algorithm is used to re-optimize the weights and bias signals during the training process. The following is a brief description of the neural learning algorithm in this embodiment:
[0093] The first step is to propagate the input information forward through the network, and the direction of the arrow is the direction of data operation. The input information in this embodiment is the open circuit voltage (OCV), the ambient temperature (T), the first valley voltage (VP1) and the second valley voltage (VP2), the first voltage difference (Vd1) and the second voltage difference (Vd2). Specifically, the open circuit voltage OCV, the ambient temperature T, the first valley voltage VP1 and the second valley voltage VP2 are input to the input layer of the feedforward neural network model, the first voltage difference Vd1 and the second voltage difference Vd2 are input to the hidden layer of the feedforward neural network model, and the input information is input as a vector into the feedforward neural network model, specifically, the characteristic parameter vector and the voltage difference vector are input into the feedforward neural network model.
[0094] It should be noted that neurons are the most important component of a neural network, and each layer of a feedforward neural network model is composed of several neurons. The input layer only represents the initial input vector of the entire neural network, and the neurons in the input layer do not process the data.
[0095] a 0 =p (2)
[0096] a m+1 =f m+1 (W m+1 a m +b m+1 )for m=0, 1,..., M-1 (3)
[0097] a=a M (4)
[0098] p is the input vector, a0 is the input vector accepted by the input layer neuron, M is the number of hidden layers, a m is the output vector of the mth hidden layer, a m+1 is the output vector of the m+1th hidden layer, W m+1 is the weight vector from the mth hidden layer to the m+1th hidden layer, b m+1 is the bias vector from the mth hidden layer to the m+1th hidden layer, f m+1 is the activation function.
[0099] In this embodiment, p is the characteristic parameter vector of the battery, p = (open circuit voltage OCV, ambient temperature T, first valley voltage VP1, second valley voltage VP2), n is the voltage difference vector, n = (first voltage difference Vd1, second voltage difference Vd2). The second step is to propagate the error (sensitivity) backward through the network. Finally, the weights and biases are updated using the approximate steepest descent rule.
[0100] like Figure 5 The FFNN shown is a multi-layer feedback neural network model in which any error can be transmitted back and corrected, and the corresponding signal is forward propagated. The feedforward-forward neural network uses the BP algorithm to deal with key nonlinear problems. At the same time, the back-propagation feedforward neural network can give a suitable model based on the sampled data. In view of the problem of calculating SOH by traditional methods, the FFNN method is used to re-optimize the training data. The FFNN solution is a better solution to obtain the best solution.
[0101] The heuristic FFNN method provided by the present invention can monitor SOH without disconnecting the battery load. When more sampling data is used for training to obtain a trained feedforward neural network model, a more accurate SOH estimation value can be obtained. According to the SOH estimation value, the health status of the lead-acid battery can be predicted in a timely and accurate manner.
[0102] The output of the corresponding neural network is shown in formula (5):
[0103] H i =S(W ij I i +B i ) (5)
[0104] Among them, H i represents the i-th output of the hidden layer, W ij is the weight matrix connecting the j-th output of the input layer and the i-th input of the hidden layer, I i is the i-th input of the hidden layer, B i Indicates the i-th deviation signal input. Deviation signal input B i They are connected to the hidden layer respectively to compare the input values. S represents the sigmoid function;
[0105]
[0106] Output,Y=S(W ij H i +B O ) (7)
[0107] Among them, x is the input characteristic parameter vector, Y represents the battery health state estimation value, i.e., the SOH estimation value, which is the final output of the feedforward neural network model, and B0 is the bias connected to the output.
[0108] In step S400, predicting the health state of each battery according to the health state estimation value of each battery includes: when the health state estimation value of the battery is greater than a preset health state threshold, the battery is in a healthy state; otherwise, the battery is in an unhealthy state. In a preferred embodiment, the preset health state threshold may be 30%.
[0109] It should be noted that SOH is the state of the battery from the beginning to the end of its life, and it characterizes the ability of the current battery to store electrical energy relative to a new battery. The preset health status threshold can be determined based on the actual situation of the battery's user. For example, when the battery is used in a car, the health status threshold can be set to 80%, indicating that it is necessary to consider replacing the battery. A health status of 0% means that the battery is completely scrapped.
[0110] The battery health status prediction method based on feedforward neural network provided in the embodiment of the present invention obtains aging characteristic parameters such as open circuit voltage, ambient temperature, and valley voltage when the battery group is connected to a load. The change in open circuit voltage can reflect the change in the positive and negative electrode performance, impedance, and battery energy storage capacity of the battery. The ambient temperature will affect the charging and discharging, service life and internal resistance of the battery. The valley voltage can reflect the low voltage condition of the battery. A corresponding characteristic parameter vector is constructed according to the aging characteristic parameters and input into the trained feedforward neural network model, so as to obtain a more accurate health status estimation value. The health status of the battery can be predicted timely and accurately according to the health status estimation value. The present invention can improve the accuracy of battery health status detection without disconnecting the battery load.
[0111] The present invention also provides an embodiment of a battery health status prediction system based on a feedforward neural network, comprising:
[0112] A characteristic parameter vector construction module is used to obtain aging characteristic parameters of each battery in the battery pack when the battery pack is connected to a load, and to construct a characteristic parameter vector of each battery according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley point voltage and a second valley point voltage, wherein the second valley point voltage is greater than the first valley point voltage;
[0113] A health state estimation value acquisition module, used to input the characteristic parameter vector as input data into a trained feedforward neural network model, and output the health state estimation value of each battery; the feedforward neural network model is trained using a BP algorithm;
[0114] The health status prediction module is used to predict the health status of each of the batteries according to the health status estimation value of each of the batteries.
[0115] Specifically, the health state prediction module predicts the health state of each battery according to the health state estimation value of each battery, including:
[0116] When the estimated health state value of the battery is greater than a preset health state threshold, the battery is in a healthy state; otherwise, the battery is in an unhealthy state.
[0117] Specific examples of the present invention are as follows:
[0118] The SOH was measured using a 12V, 42Ah Valve Regulated Lead Acid (VRLA) battery. The battery was connected to a variable resistor load, with the load varying from 2A to a maximum load of 12A. The selected battery had 300 life cycles according to the manufacturer. The starting voltage of the battery was calculated using a Field Programmable Gate Array (FPGA) controller and the output was obtained using a Digital Storage Oscilloscope (DSO). Whenever the battery reaches a safe SOC, it needs to be recharged. The safe remaining capacity (State of Charge, SOC) of the battery is considered to be 20% of the total SOC. The time to reach the safe SOC depends on the load. All tests were performed in the temperature range of 25-26 degrees Celsius.
[0119] It should be noted that SOC (safe remaining capacity, also called state of charge) is a parameter used in the cycle counting method. When the number of cycles increases, the performance of the battery decreases, and the size of the charging cycle provides a health status of the battery being consumed.
[0120] See also Figure 7 The voltage waveform of the battery test shown in the figure, the experiment considered 10 lead-acid batteries with different aging degrees, and began to collect data such as open circuit voltage (OCV), temperature (T), and the first valley point voltage (VP1) and the second valley point voltage (VP2). The experimental data was collected through the experimental bench and used to create a lookup table for neural network training. As the battery aging process progresses, the first valley point voltage VP1 and the second valley point voltage VP2 have a tendency to decrease.
[0121] See also Figure 8 , Figure 8 The OCV and SOC characteristics of the old battery after 25 cycles are given, showing the change of discharge voltage with the battery charge state. In the SOH estimation method, the size of the battery open circuit voltage data OCV plays an important role. OCV can be obtained from an equivalent circuit model. However, when discharging, the discharge voltage of the new battery is between 14V and 10.5V, and the discharge voltage of the old battery is between 12V and 10.5V. 25 cycles are calculated using the cycle counting method.
[0122] See also Fig. 9 and Fig.10 , Fig. 9 Impedance etc. with respect to the number of cycles are given, explaining the increase in impedance values from 0 (new battery) to 300 cycles (dead battery, according to the manufacturer's data). The impedance value of the battery increases due to anode corrosion, therefore, increasing the number of cycles will affect the changes in impedance, discharge voltage, SOC, SOH value and battery energy storage capacity. The experimental results show that with increasing number of cycles, the SOH value and battery capacity gradually decrease. From Fig.10 As can be seen in the graph, after 200 cycles (66.66 of the total health value), the battery degrades faster.
[0123] The number of neurons in the hidden layer is selected from the series of trajectory runs to obtain the ideal organization structure with the minimum error. The complete data set generated by the input parameters is divided into three subsets for the training process. The neural network uses the training subset to obtain its parameters. The validation subset is used for training supervision and to detect the occurrence of overfitting. The performance of the neural network is measured using the test subset, which is also considered as an independent data subset.
[0124] See also Fig.11 The performance verification diagram shown, Fig.11 The development of the training mean square error for the three subsets is shown. After a few trainings, the fine-tuning phase begins; the error decreases significantly for all data subsets. Training is stopped after 20 epochs when overfitting occurs.
[0125] Regression analysis was performed and the regression equation was
[0126] Target=X output+Y (8)
[0127] Where X is the slope of the regression fit and Y is the offset of the regression fit.
[0128] For the entire data set, the regression value is almost equal to 1. The FFNN method proposed in this embodiment is superior to the traditional cycle counting method, and its error rate is reduced from 9.88% to 4.92%. The maximum error value of the conventional cycle counting method can be calculated by referring to the manufacturer's data and the SOH value derived from formula (8).
[0129] The results show that the conventional method overestimates the battery SOH, while the FFNN method is able to provide more accurate SOH when trained with more sampled data. The sampled data is input to the FFNN through methods such as OCV, ambient temperature measurement, and valley point detection. The conventional method and heuristic method are solved using MATLAB to achieve accurate calculation of the battery SOH. The load is not disconnected from the battery which is an additional advantage of using the FFNN scheme. The future scope of this work is to develop a prototype and test it under different temperature and operating conditions to evaluate the effectiveness of the algorithm.
[0130] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0131] In the embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0132] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0133] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0134] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0135] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A battery health status prediction method based on a feedforward neural network, characterized in that: include: When the battery pack is connected to a load, aging characteristic parameters of each battery in the battery pack are obtained, and characteristic parameter vectors of each battery are constructed according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley point voltage, and a second valley point voltage, wherein the second valley point voltage is greater than the first valley point voltage; obtaining a first voltage difference according to the open circuit voltage and the first valley voltage, obtaining a second voltage difference according to the first valley voltage and the second valley voltage, and obtaining a voltage difference vector of each of the batteries according to the first voltage difference and the second voltage difference; The characteristic parameter vector and the pressure difference vector are input as input data into a trained feedforward neural network model, and the health status estimation value of each battery is output; the feedforward neural network model is a neural network model trained by using a BP algorithm; predicting the health status of each of the batteries according to the estimated health status of each of the batteries; The first valley voltage is the lowest voltage of the battery; The second valley voltage is the second lowest voltage of the battery; The obtaining a first voltage difference according to the open circuit voltage and the first valley voltage comprises: The first voltage difference is a difference between the open circuit voltage and the first valley voltage; The obtaining a second voltage difference according to the first valley voltage and the second valley voltage comprises: The second voltage difference is a difference between the first valley voltage and the second valley voltage.
2. The method for predicting battery health status based on a feedforward neural network according to claim 1, characterized in that: The characteristic parameter vector of each battery includes at least: the open circuit voltage, the ambient temperature, the first valley voltage, and the second valley voltage.
3. The method for predicting the health status of a battery based on a feedforward neural network according to claim 1 or 2, characterized in that: Predicting the health status of each of the batteries according to the health status estimation value of each of the batteries comprises: When the estimated health state value of the battery is greater than a preset health state threshold, the battery is in a healthy state; Otherwise, the battery is in an unhealthy state.
4. The method for predicting battery health status based on a feedforward neural network according to claim 1, characterized in that: The feedforward neural network model is: Multi-layer feedforward neural network model, including input layer, multiple hidden layers and output layer; The input layer receives the input characteristic parameter vector, the multiple hidden layers process the characteristic parameter vector, the output result of the previous hidden layer is used as the input vector of the next hidden layer, and the output layer outputs the estimated value of the health status of each battery.
5. The method for predicting battery health status based on a feedforward neural network according to claim 1, characterized in that: The training process of the feedforward neural network model includes: Inputting the characteristic parameter matrix of the training samples in the training set into the feedforward neural network model and propagating forward through the network to calculate the network net input and network output of each neuron in the hidden layer and the output layer; Propagate the error backward through the network and calculate the error of each neuron in the hidden layer and the output layer; Update the weights and biases of neurons according to preset rules; The above process is repeated until the error on the validation set is less than the preset error threshold.
6. The method for predicting battery health status based on a feedforward neural network according to claim 1, characterized in that: The activation function of the feedforward neural network model is: Among them, S is the sigmoid function and x is the input feature parameter vector.
7. The battery health status prediction system based on feedforward neural network is characterized by: include: A characteristic parameter vector construction module is used to obtain aging characteristic parameters of each battery in the battery pack when the battery pack is connected to a load, and to construct a characteristic parameter vector of each battery according to the aging characteristic parameters; wherein the aging characteristic parameters at least include: open circuit voltage, ambient temperature, a first valley point voltage and a second valley point voltage, wherein the second valley point voltage is greater than the first valley point voltage; a voltage difference vector construction module, configured to obtain a first voltage difference according to the open circuit voltage and the first valley voltage, obtain a second voltage difference according to the first valley voltage and the second valley voltage, and obtain a voltage difference vector of each battery according to the first voltage difference and the second voltage difference; A health state estimation value acquisition module, used to input the characteristic parameter vector and the pressure difference vector as input data into a trained feedforward neural network model, and output the health state estimation value of each battery; the feedforward neural network model is trained using a BP algorithm; A health status prediction module, used for predicting the health status of each of the batteries according to the estimated health status of each of the batteries; The first valley voltage is the lowest voltage of the battery; The second valley voltage is the second lowest voltage of the battery; The obtaining a first voltage difference according to the open circuit voltage and the first valley voltage comprises: The first voltage difference is a difference between the open circuit voltage and the first valley voltage; The obtaining a second voltage difference according to the first valley voltage and the second valley voltage comprises: The second voltage difference is a difference between the first valley voltage and the second valley voltage.
8. The battery health status prediction system based on feedforward neural network according to claim 7 is characterized in that: The health state prediction module predicts the health state of each battery according to the health state estimation value of each battery, including: When the estimated health state value of the battery is greater than a preset health state threshold, the battery is in a healthy state; Otherwise, the battery is in an unhealthy state.
Citation Information
Patent Citations
Battery charging state evaluation system and method based on combination of extended Kalman filtering and genetic BP neural network
CN110175672A
Electric forklift lithium ion battery health state prediction method based on LSTM-FFNN
CN111308375A
Lithium ion battery state-of-health estimation method based on discharge process
CN112684363A
Lithium ion battery health state diagnosis method based on deep learning model
CN113009349A
Energy storage battery state-of-charge and state-of-health joint estimation method and system
CN113064093A
