Method for estimating energy state of lithium ion battery of energy storage power station

Abstract

The invention discloses an energy storage power station lithium ion battery energy state estimation method, and relates to the technical field of lithium batteries, and the method comprises the steps: building a fractional order battery equivalent circuit model, and obtaining a state equation and a measurement equation; obtaining a relation curve between the open-circuit voltage Uocv and the SOC through a pulse discharge experiment; dynamic stress is used for testing working condition data, and model parameters are identified by adopting a particle swarm optimization algorithm; establishing an adaptive fractional order extended Kalman filtering observer based on the result, estimating the SOC value at the moment k, and obtaining a preliminary SOE value through function mapping; a back propagation neural network is introduced, current, voltage, a Kalman gain matrix and terminal voltage observation output errors are used as input, an SOE estimation error compensation value is output, and a final SOE estimation value is the sum of an AFOEKF estimation value and a BP neural network compensation value. According to the method, the problems of data saturation and noise accumulation errors in Kalman filtering are effectively improved, and the energy state estimation precision and the environmental adaptability are improved.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, specifically to a method for estimating the state of energy of lithium-ion batteries in energy storage power stations. Background Technology

[0002] Lithium-ion batteries, with their superior performance and high reliability, have been widely and deeply applied in various fields. However, in actual operation, the performance of lithium-ion batteries is constrained by a variety of complex factors, all of which significantly affect the battery's internal parameters. Therefore, the key to improving the stability of the battery management system (BMS) is to establish an intelligent equivalent model of battery performance and simultaneously research state estimation optimization strategies.

[0003] In practical applications, the battery's state of charge (SOC) and state of energy (SOE) are two core monitoring indicators, often collected and utilized in parallel. SOC only tracks the cumulative current over time and cannot reveal the dynamic changes in terminal voltage over time, thus it cannot directly map the battery's instantaneous power capability. In contrast, SOE uses energy as a benchmark and has an explicit functional relationship with available power, making it more suitable for accurate estimation of driving range and real-time optimization of energy management strategies.

[0004] In recent years, considering the unavoidable cumulative errors caused by the inherent biases of current and voltage sensors in traditional power integration methods, researchers have drawn inspiration from SOC estimation, categorizing SOE estimation into three aspects: power integration, model-driven, and data-driven. Integration methods are hampered by initial errors, with drift amplifying over time. While data-driven methods can approximate arbitrary nonlinearities, they rely on massive amounts of high-quality samples, resulting in high training costs and computational complexity. Model-driven methods attempt to strike a trade-off between accuracy and computational cost, but existing literature largely focuses on integer-order equivalent circuits, with limited research on the crucial mechanism of low-frequency hysteresis in solid-state diffusion in fractional-order 2-RC networks. Furthermore, most model-driven methods are based on Kalman filtering, failing to consider the saturation issues caused by excessive data and the cumulative errors from noise in actual calculations. A SOE estimation method that combines high accuracy with strong environmental adaptability has become a bottleneck that urgently needs to be overcome in the field of battery management. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by proposing a method for estimating the state of energy (SEE) of lithium-ion batteries in energy storage power stations. This method aims to effectively improve the problem of accumulated errors caused by saturation noise statistics when there is too much data during the Kalman filter estimation process, thereby enhancing the accuracy of SEE estimation.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for estimating the state of energy of lithium-ion batteries in an energy storage power station includes the following steps:

[0008] Step 1: Establish a fractional-order equivalent circuit model of the battery and calculate the state equation and measurement equation of the battery.

[0009] Step 2: Obtain the relationship curve between open circuit voltage Uocv and SOC using a lithium-ion battery pulse discharge experiment;

[0010] Step 3: Using the dynamic stress test data of lithium-ion batteries, the dynamic stress test (DST) data is identified by the particle swarm optimization algorithm (PSO) to obtain model parameters;

[0011] Step 4: Based on the discretized state equation and measurement equation of the battery obtained in Step 1, establish an adaptive fractional extended Kalman filter (AFOEKF) observer. Input the open-circuit voltage Uocv and SOC relationship curve obtained in Step 2 and the model parameters obtained in Step 3 into the adaptive fractional extended Kalman filter observer. The observer iteratively estimates the SOC value at time k. The preliminary SOE estimate is obtained through the functional mapping relationship between SOC and SOE.

[0012] Step 5: Construct a backpropagation neural network to integrate the current, voltage values, and the Kalman gain matrix obtained from the adaptive fractional extended Kalman filter observer algorithm at time k. The terminal voltage observation output error is used as the input to the BP neural network, and the output of the BP neural network is the difference between the true SOE and the preliminary SOE estimated by AFOEKF.

[0013] Step 6: Sum the difference between the initial SOE value estimated by AFOEKF and the output of the BP neural network to obtain the final SOE estimate of the lithium-ion battery.

[0014] As a further technical solution of the present invention: the state equation and measurement equation of the battery are respectively:

[0015]

[0016] The test equation is:

[0017]

[0018] in, Represented as The voltages of components R1-CPE1, R2-CPE2, and W at that moment; ; U ocv Indicates open-circuit voltage; U LR is the battery terminal voltage; R0 is the battery internal ohmic resistance; R1 is the battery electrochemical polarization resistance; R2 is the battery concentration polarization resistance; C1, C2, and W are the elemental parameters of the fractional-order model, s is a complex variable; α, β, and These are the orders of C1, C2, and W, respectively.

[0019] As a further technical solution of the present invention: the working process of the adaptive fractional extended Kalman filter observer in step 4 includes the following sub-steps:

[0020] I. State Initialization:

[0021]

[0022] II. Prediction Phase:

[0023] After initialization, calculate the predicted values ​​of the state variables. Predicted values ​​of error covariance :

[0024]

[0025] III. Measurement Update:

[0026] Calculate the Kalman gain:

[0027] Update state variables and error covariance:

[0028]

[0029] IV. , Adaptive update: definition The new information of the moment Residual information ;

[0030]

[0031]

[0032] In the formula Estimating the variance of the new information The length of the sliding data window;

[0033] Noise update:

[0034]

[0035] V. Function Mapping: Through Establish a functional mapping relationship between SOC and SOE to obtain preliminary SOE estimates, where a and b are fitting coefficients.

[0036] As a further technical solution of the present invention: the BP neural network in step 5 adopts a three-layer structure, including an input layer, a hidden layer, and an output layer, wherein:

[0037] The input layer consists of four nodes, corresponding to the current, voltage value, Kalman gain matrix Kk at time k, and terminal voltage observation output error, respectively.

[0038] The output layer consists of one node, which represents the difference between the actual SOE and the preliminary estimated SOE.

[0039] The BP neural network adjusts its parameters based on a gradient descent strategy, with a learning rate η∈(0,1), and the mean squared error is calculated using the mean squared error formula.

[0040]

[0041] Optimize network weights and thresholds.

[0042] This technology proposes a method for estimating the state of energy of lithium-ion batteries in energy storage power stations, which has the following advantages and benefits:

[0043] This invention proposes an Adaptive Fractional Extended Kalman Filter-Backpropagation Neural Network (AFOEKF-BP) algorithm for SOE estimation. This algorithm utilizes the self-learning and nonlinear approximation capabilities of the BP neural network to compensate for the estimation error of AFOEKF, thereby enabling deeper feature and pattern extraction in real-time processing and enhancing the filter's flexibility in complex environments. Attached Figure Description

[0044] Figure 1 This is a technical roadmap of the present invention;

[0045] Figure 2 This is a diagram of the equivalent circuit of a fractional-order lithium-ion battery.

[0046] Figure 3 It is a graph showing the relationship between the fitted OCV and SOC.

[0047] Figure 4 This is the flowchart of the AFOEKF-BP algorithm;

[0048] Figure 5 This is a function mapping diagram between SOC and SOE;

[0049] Figure 6 This is a graph showing the prediction results of SOE with an initial value of 0.8. Detailed Implementation

[0050] The present invention will be further described below with reference to the embodiments. It should be noted that these are merely examples and descriptions of the inventive concept. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in the claims, they should all be considered to fall within the protection scope of the present invention.

[0051] like Figure 1-6 As shown, the present invention proposes a method for estimating the state of energy of lithium-ion batteries in energy storage power stations, comprising the following steps:

[0052] Step 1, establish a fractional-order battery equivalent circuit model as follows: Figure 2 As shown, the state equation and measurement equation of the battery are calculated.

[0053] Furthermore, the state equation and measurement equation of the battery are as follows:

[0054]

[0055]

[0056] in, Represented as The voltages of components R1-CPE1, R2-CPE2, and W at that moment; ; U ocv Indicates open-circuit voltage; U L R is the battery terminal voltage; R0 is the battery internal ohmic resistance; R1 is the battery electrochemical polarization resistance; R2 is the battery concentration polarization resistance; C1, C2, and W are the elemental parameters of the fractional-order model, s is a complex variable; α, β, and These are the orders of C1, C2, and W, respectively.

[0057] Step 2: Using a lithium-ion battery pulse discharge experiment, obtain the relationship curve between open-circuit voltage Uocv and SOC, as shown in the figure. Figure 3 As shown;

[0058] Step 3: Using the dynamic stress test data of lithium-ion batteries, the model parameters are obtained by using the particle swarm optimization algorithm (PSO) to identify the dynamic stress test (DST) data.

[0059] Step 4: Estimate the battery SOE using the adaptive fractional extended Kalman filter-backpropagation neural network algorithm. Figure 4 As shown. The method includes the following steps:

[0060] (1) State initialization.

[0061]

[0062] (2) Prediction stage.

[0063] After initialization, calculate the predicted values ​​of the state variables. Predicted values ​​of error covariance :

[0064]

[0065] (3) Measurement update.

[0066] Calculate the Kalman gain:

[0067] Update state variables and error covariance:

[0068]

[0069] (4) Adaptive update.

[0070] , Adaptive update: Definition The new information of the moment Residual information .

[0071]

[0072]

[0073] In the formula Estimating the variance of the new information The length of the sliding data window.

[0074] Noise update:

[0075]

[0076] After completing the adaptive update of the noise, the updated noise is used as the current state variable through the AFOEKF algorithm. The SOC is updated to obtain the optimal estimate.

[0077] Next, establish the function mapping relationship between SOC and SOE: To obtain the best-fit function, and thus the observed SOE value. Figure 5 As shown.

[0078] Furthermore, to compensate for the estimation error of AFOEKF, a backpropagation (BP) network is introduced to extract features and patterns more deeply in real-time processing, thereby enhancing the flexibility of the filter in complex environments.

[0079] Backpropagation (BP) neural networks typically employ a three-layer structure; increasing the number of neurons in each layer can improve fitting accuracy. The SOE prediction model uses a three-layer structure: an input layer, hidden layers, and an output layer. The output layer... The threshold of each neuron is used This indicates that the hidden layer is... One neuron is used Indicates. The first input layer The first neuron and the hidden layer The connection weights between neurons are Hidden layer The nth neuron and the output layer The connection weights between neurons are The hidden layer is... The input received by each neuron is The output layer's first The input received by each neuron is .in, For the hidden layer The output of each neuron. For the training example Assume the output of the neural network is:

[0080]

[0081] Then the neural network in The mean square error is:

[0082]

[0083] The BP algorithm is based on the gradient descent strategy, adjusting the parameters in the direction of the negative gradient of the target. Error in Given a learning rate ,have:

[0084]

[0085] First it affects the first Input value of each output neuron This then affects its output value. And then affect ,have:

[0086]

[0087] according to The definitions are as follows:

[0088]

[0089] Given the properties of the Sigmoid function:

[0090]

[0091] According to equation (8), we have:

[0092]

[0093] The formula substitute Regeneration That is, to obtain the BP algorithm Update formula:

[0094]

[0095] Connection weights from the output layer to the hidden layer of a BP neural network The update estimate is:

[0096]

[0097] The hidden layer of a BP neural network Threshold of a neuron The update formula is:

[0098]

[0099] Learning rate It controls the update step size in each iteration of the algorithm.

[0100] Finally, the Kalman gain matrix is ​​obtained by using the current, voltage values, and the AFOEKF algorithm at time k. The observed SOE error is used as the input to the BP neural network, and the output is the difference between the observed SOE value and the SOE estimated by AFOEKF. The final SOE estimate is the sum of the AFOEKF estimate and the output of the BP neural network.

[0101] Example

[0102] The lithium-ion battery data in this article comes from open-source experimental data of battery packs from the Center for Advanced Life Cycle Engineering (CALCE) at the University of Maryland. The test platform included INR18650 nickel-cobalt-manganese (NCM) / graphite lithium-ion cells, a temperature test chamber, an Arbin BT2000 battery testing system, and a computer with Arbin software. The battery has a capacity of 2.0 Ah and a nominal voltage of 3.6 V. Its operating voltage range is set between 2.5 V and 4.2 V, and the maximum current it can withstand is 22 A. Furthermore, its operating temperature range is between 0°C and 50°C, ensuring that the battery can operate normally and maintain stable performance within this temperature range.

[0103] The Dynamic Stress Test (DST), designed by the US Advanced Battery Consortium (USABC), simulates dynamic discharge conditions and can be scaled down to the maximum required performance based on the test data. Therefore, test data of the battery at 25 °C under DST conditions was chosen for the model parameter identification process.

[0104] The BP network structure consists of an input layer, an output layer, and hidden layers. There are 10 hidden nodes, and the training iterations are 1000. The input layer has 4 nodes: current, voltage, and the Kalman gain matrix at time k. Terminal voltage observation output error The input values ​​for the BP neural network are linearly mapped to [−1, 1]; the output layer has 1 node: the difference between the true SOE and the SOE estimated by AFOEKF is linearly mapped to [0, 1]. The learning rate is 0.01; the training set consists of the first 4000 points, and the test set consists of the last 6000 points, with no temporal overlap between the training and test sets. Both input and output variables are normalized.

[0105] Figure 6This paper presents the SOE estimation results and error comparison charts for lithium batteries based on an adaptive fractional extended Kalman filter-backpropagation neural network algorithm, specifically an embodiment of this algorithm. In the experimental setup, the initial SOE value was precisely set to the actual SOE value of 0.8, and SOE estimation was performed under standard 25°C, DST conditions. The comparison charts and error comparisons of SOE estimation using the AFOEKF-BP algorithm clearly show that the SOE estimate obtained using the AFOEKF-BP algorithm has a higher degree of overlap with the actual SOE value, with the SOE estimation error ranging only from -0.005 to 0.004. The results demonstrate that this invention effectively suppresses the interference of historical data on the current estimate by introducing a forgetting factor mechanism, improving the filter's adaptability to the time-varying characteristics of the model. Simultaneously, by using a BP neural network to model and compensate for the FOEKF estimation error, the estimation accuracy of SOE is significantly improved, especially exhibiting good robustness and convergence under complex conditions such as high noise and model mismatch.

[0106] The above is an exemplary description of the invention. Obviously, the specific implementation of the invention is not limited to the above-described manner. Any non-substantial improvement made using the inventive concept and technical solution of the invention, or the direct application of the inventive concept and technical solution to other situations without modification, is within the protection scope of the invention.

Claims

1. A method for estimating the state of energy of a lithium-ion battery of an energy storage power plant, characterized in that, The method comprises the following steps: Step 1, establishing a fractional order battery equivalent circuit model, and calculating a state equation and a measurement equation of the battery; Step 2, obtaining an open-circuit voltage Uocv and SOC relationship curve by using a lithium ion battery pulse discharge experiment; Step 3, obtaining model parameters by identifying dynamic stress test working condition data of the lithium ion battery through a particle swarm optimization algorithm; Step 4, establishing an adaptive fractional order extended Kalman filter observer based on the state equation and the measurement equation of the battery obtained in step 1, inputting the open-circuit voltage Uocv and SOC relationship curve obtained in step 2 and the model parameters obtained in step 3 into the adaptive fractional order extended Kalman filter observer, and estimating an SOC value at a k time point through the observer, so as to obtain a preliminary SOE estimation value through a function mapping relationship between the SOC and the SOE; Step 5, constructing a back propagation neural network, taking the current, voltage value, and the Kalman gain matrix obtained by the adaptive fractional order extended Kalman filter observer algorithm at time k as inputs, and taking the difference between the true SOE and the preliminary estimated SOE by the AFOEKF as output. and the terminal voltage observation output error as the input of the BP neural network, and taking the difference between the true SOE and the preliminary estimated SOE by the AFOEKF as output. Step 6, summing a difference between the preliminary SOE estimation value of the AFOEKF and a BP neural network output to obtain a final SOE estimation value of the lithium ion battery.

2. The method of claim 1, wherein, The state equation and the measurement equation of the battery are respectively: The test equation is: wherein, The voltage of R1-CPE1, R2-CPE2, W element is represented as The voltage of R1-CPE1, R2-CPE2, W element is represented as ; , U ocv The open circuit voltage is represented as U L The battery terminal voltage is represented as U; R0 is the battery ohmic resistance; R1 is the battery electrochemical polarization resistance; R2 is the battery concentration polarization resistance; C1, C2, W are element parameters of the fractional order model, s is a complex variable; α, β and are the orders of C1, C2 and W, respectively.

3. The method of claim 1, wherein, The working process of the adaptive fractional order extended Kalman filter observer in step 4 comprises the following sub-steps: I, state initialization: II, prediction stage: After initialization, the predicted values of the state variables and the predicted values of the error covariances are computed III, measurement update: Compute Kalman gain: Update state variables and error covariance: IV、 , Adaptive update of the definition of the innovation at the moment , residual information ;​ In the formula is the innovation estimation variance is the sliding data window length; Noise update: V. Function mapping: by The function mapping relationship between the SOC and the SOE is established to obtain a preliminary SOE estimation value, wherein a and b are fitting coefficients.

4. The method of claim 1, wherein, The BP neural network in step 5 adopts a three-layer structure, comprising an input layer, a hidden layer and an output layer, wherein: The input layer has four nodes, respectively corresponding to a current value, a voltage value, a Kalman gain matrix Kk at the k time point and an end voltage observation output error; The output layer has one node, corresponding to a difference between a real SOE and a preliminary SOE estimation value; The BP neural network adjusts parameters based on a gradient descent strategy, a learning rate η ∈ (0, 1), and optimizes network weights and thresholds through a mean square error formula .

Images