A method for predicting the remaining capacity of lead-acid batteries in DC systems

Abstract

The invention discloses a method for predicting the remaining capacity of a lead-acid battery in a direct current system, and specifically relates to the technical field of lead-acid batteries. The method collects and preprocesses historical data of batteries, establishes and analyzes linear and nonlinear aging models, identifies nonlinear characteristics in the aging process, and after identifying the nonlinear characteristics, evaluates and classifies the accuracy of the nonlinear model based on the model fitting goodness and prediction performance, selects a high-accuracy model for aging prediction, and compares and analyzes the model with actual data. According to the prediction deviation and accuracy evaluation results, the model parameters are dynamically adjusted using fuzzy logic to continuously improve the prediction accuracy. The method can not only effectively cope with the complex nonlinear changes in the battery aging process, but also ensure the accuracy of capacity prediction and the stable operation of the power system.

Description

Technical Field

[0001] The invention relates to the technical field of lead-acid batteries, and in particular to a method for predicting the remaining capacity of a lead-acid battery in a direct current system. Background Art

[0002] The prediction of the remaining capacity of lead-acid batteries in DC systems refers to estimating the amount of electrical energy that the battery can still provide under current operating conditions by measuring and analyzing battery parameters such as voltage, current, and temperature, using mathematical models or algorithms. This prediction helps assess the health status and remaining service life of the battery, ensure the stable operation of the power system, and avoid unexpected downtime or power outages. However, as the use time increases, the internal chemical composition and structure of the lead-acid battery will change. This aging process is nonlinear and difficult to model accurately. At the same time, if the model fails to fully consider these nonlinear factors, the capacity prediction may deviate seriously from the actual situation, which may cause system operation interruptions or power shortages. Summary of the invention

[0003] In order to achieve the above object, the present invention provides the following technical solution: a method for predicting the remaining capacity of a lead-acid battery in a DC system, comprising the following steps:

[0004] S1: Collect historical data of lead-acid batteries at different time points and under different working conditions, pre-process the historical data, and organize them according to time series;

[0005] S2: Establish a linear aging model and a nonlinear aging model, and fit the model to the historical data of lead-acid batteries to obtain the fitting parameters of each model. Use residual analysis to identify whether there are nonlinear characteristics in the aging process;

[0006] S3: When nonlinear characteristics exist in the aging process, the accuracy of the nonlinear model prediction is evaluated according to the fluctuation of the model fit goodness and the stability of the prediction performance, and it is divided into high-accuracy nonlinear model and low-accuracy nonlinear model according to the evaluation results;

[0007] S4: Predict the aging of lead-acid batteries based on a high-accuracy nonlinear model, and compare the predicted results with the actual operating data to determine the deviation between the predicted results and the actual data;

[0008] S5: According to the deviation between the prediction results and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic to improve the accuracy of capacity prediction.

[0009] Preferably, in S2, a linear aging model and a nonlinear aging model are established, and the historical data of the lead-acid battery are fitted to obtain the fitting parameters of each model, and residual analysis is used to identify whether there are nonlinear characteristics in the aging process, specifically:

[0010] Construct a linear aging model. If the capacity decay of the lead-acid battery is linearly related to time or number of cycles, the expression of the linear aging model is: ; Where C(t) is the battery capacity at time t, is the initial capacity, is the decay rate; a nonlinear aging model is constructed to capture complex aging behaviors, and the model expression is: ;in, is the nonlinear attenuation rate; determine the input variables of the model, including the number of charge and discharge cycles, temperature, and discharge depth; determine the output variables of the model, including the remaining capacity C(t) of the battery and the battery health status indicator; divide the historical data into a training set and a validation set;

[0011] The least squares method is used to fit the model and the best fitting parameters of the linear model are obtained by minimizing the sum of squares of the errors between the predicted values ​​and the actual values. and , the expression is: ; According to the fitted linear model, calculate the predicted value of battery capacity corresponding to each time point t , according to the actual value C(t) and the model prediction value The difference between the two is used to calculate the residual at each time point , the expression is: ;

[0012] The residual calculation results of all time points are organized into a residual sequence, and a residual graph is drawn with time t as the horizontal axis and residual e(t) as the vertical axis. If the residual shows a significant upward or downward trend rather than a random distribution, it indicates that there are nonlinear characteristics in the battery aging process.

[0013] Preferably, in S3, a local fitting fluctuation index is generated according to the fluctuation of the model fitting goodness, and the method for obtaining the local fitting fluctuation index is:

[0014] Obtain the model goodness of fit value within the time period s in real time, and construct the corresponding time series recorded as R(m), where m represents time. Set the initial sliding window size W0. At the mth moment, use the current sliding window size W(m) to calculate the mean of the goodness of fit value in the window. , the expression is: ; and the standard deviation of the goodness-of-fit values , ; In the formula, is the goodness of fit value of the i-th model in the time series; the standard deviation of the goodness of fit in the window is compared with the standard deviation reference threshold of the goodness of fit under the preset standard conditions, and the size of the sliding window is adjusted. The expression is: ; where k is the adjustment coefficient, is the benchmark value of the standard deviation, according to the adjusted sliding window W( ), calculate the local fitting fluctuation index LF, the expression is: .

[0015] Preferably, in S3, a prediction adjustment response index is generated according to the stability of the model prediction performance, and the method for obtaining the prediction adjustment response index is:

[0016] In the absence of disturbance, use the model to predict the future time point Q, record the baseline prediction value, and select the input variable at a certain moment. Apply disturbance, record the input variable value after disturbance, and generate a new input sequence; after introducing disturbance, the model will generate a new prediction value based on the new input sequence, which is recorded as the initial prediction value after disturbance , calculate the error between the initial prediction value after the disturbance and the baseline prediction value , the expression is: ; The model is divided into several time steps The predicted value generated after adjusting the disturbance after +ΔQ is marked as , calculate the error between the adjusted predicted value and the actual value , which represents the prediction accuracy after model adjustment, is expressed as: ; The computational model is disturbed The time required for the prediction error to reach a stable range , that is, the time required for the error to stabilize when the change in the prediction error is less than the threshold, and the prediction adjustment response index is calculated, and the expression is: ; In the formula, is the predicted adjustment response index. Preferably, the local fitting volatility index and the predicted adjustment response index are converted into a first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the accuracy value label predicted by each group of first eigenvectors to predict the nonlinear model as the prediction target, and minimizes the sum of the prediction errors of the accuracy value labels predicted by all nonlinear models as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence, and the model training is stopped. The accuracy value of the nonlinear model prediction is determined according to the model output result, wherein the machine learning model is a polynomial regression model. Preferably, the obtained accuracy value of the nonlinear model prediction is compared with the preset reference threshold of the accuracy value of the nonlinear model prediction. If the accuracy value of the nonlinear model prediction is greater than or equal to the preset reference threshold of the accuracy value of the nonlinear model prediction, it means that the accuracy of the nonlinear model prediction is high, and a high-accuracy prediction signal is generated at this time, and it is classified as a high-accuracy nonlinear model; if the accuracy value of the nonlinear model prediction is less than the preset reference threshold of the accuracy value of the nonlinear model prediction, it means that the accuracy of the nonlinear model prediction is low, and a low-accuracy prediction signal is generated at this time, and it is classified as a low-accuracy nonlinear model.

[0017] Preferably, in S4, the deviation between the predicted value and the actual value at each time point is calculated, and the calculated deviation is analyzed to generate a deviation trend index. The deviation between the predicted result and the actual data is determined according to the deviation trend index. The method for obtaining the deviation trend index is:

[0018] Calculate the deviation between the predicted value and the actual value at each time point and construct the corresponding time series , where v = 1, 2, ..., N, N is the number of time points; for the deviation time series Perform linear regression and fit the linear model. The expression is: ; Where M is the intercept, R is the regression slope, which represents the linear rate of change of the deviation over time, ϵ(v) is the regression residual, and the deviation trend index is calculated as follows: ; In the formula, is the standard deviation of the deviation series, used to normalize the slope, is the deviation trend index.

[0019] Preferably, in S5, according to the deviation between the prediction result and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic, specifically:

[0020] The accuracy value and deviation trend index of the nonlinear model prediction are used as the input items of fuzzy logic, and the amplitude of the model parameter adjustment is used as the output item; the deviation trend index, the prediction accuracy of the nonlinear model and the amplitude of the model parameter adjustment are fuzzified; a fuzzy rule set is constructed, and the input deviation and accuracy data are converted into corresponding fuzzy sets through fuzzy membership functions; the fuzzy value of the output variable parameter adjustment amplitude is obtained by fuzzy reasoning using the defined fuzzy rule set and the input fuzzy variables; the result of fuzzy reasoning is converted into a specific parameter adjustment amount, and a clear adjustment amount is calculated using a defuzzification method; and the parameters of the nonlinear model are adjusted according to the defuzzification result.

[0021] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0022] 1. The present invention combines linear and nonlinear models to accurately model and predict the aging process of lead-acid batteries, solving the problem that traditional methods are difficult to capture nonlinear aging characteristics. Through residual analysis and model evaluation, the accuracy of the model can be identified and classified, ensuring that a high-precision model is used for battery life prediction, thereby improving the reliability and effectiveness of the prediction results.

[0023] 2. The present invention can adaptively optimize the model performance and further improve the accuracy of capacity prediction by introducing fuzzy logic to dynamically adjust the model parameters. It not only reduces the systematic error, but also improves the robustness of the model under different working conditions, thereby better ensuring the stable operation of the power system and avoiding unexpected shutdowns or power shortages caused by inaccurate predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0025] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] For examples, see Figure 1 As shown, the method for predicting the remaining capacity of a lead-acid battery in a DC system described in this embodiment includes the following steps:

[0028] S1: Collect historical data of lead-acid batteries at different time points and under different working conditions, pre-process the historical data, and organize them according to time series;

[0029] S2: Establish a linear aging model and a nonlinear aging model, and fit the model to the historical data of lead-acid batteries to obtain the fitting parameters of each model. Use residual analysis to identify whether there are nonlinear characteristics in the aging process;

[0030] S3: When nonlinear characteristics exist in the aging process, the accuracy of the nonlinear model prediction is evaluated according to the fluctuation of the model fit goodness and the stability of the prediction performance, and it is divided into high-accuracy nonlinear model and low-accuracy nonlinear model according to the evaluation results;

[0031] S4: Predict the aging of lead-acid batteries based on a high-accuracy nonlinear model, and compare the predicted results with the actual operating data to determine the deviation between the predicted results and the actual data;

[0032] S5: According to the deviation between the prediction results and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic to improve the accuracy of capacity prediction.

[0033] In S1, historical data of lead-acid batteries at different time points and under different working conditions are collected, and the historical data are preprocessed and sorted according to the time series, specifically:

[0034] Determine the key parameters of the lead-acid battery that need to be collected, including voltage, current, temperature, state of charge (SOC), state of health (SOH), number of discharge / charge cycles, ambient temperature, etc. These parameters can reflect the working condition and aging of the battery. Set the frequency of data collection according to the working characteristics of the battery. For parameters that change faster (such as voltage, current), a higher sampling frequency can be used, while for parameters that change more slowly (such as SOC, SOH), a lower sampling frequency can be used. Collect historical data of the battery under different operating conditions (such as different discharge currents, temperature conditions, number of charge and discharge cycles, etc.). These operating conditions include normal working conditions, operation at extreme temperatures, different depths of discharge, etc., to cover the actual usage scenarios of the battery.

[0035] The collected raw data is denoised to eliminate noise caused by sensor errors, environmental interference, etc. Common methods include moving average filtering, wavelet transform, etc. In order to reduce the impact of data fluctuations, the data can be smoothed. Common smoothing methods include exponential weighted moving average method, Lorentz curve smoothing, etc., so as to more accurately reflect the true state of the battery. During the data collection process, data may be missing. Linear interpolation, spline interpolation, or Lagrange interpolation can be used to interpolate missing data to ensure data integrity. Identify outliers in the data (such as sudden changes in voltage or current), and decide whether to remove outliers or replace them with alternative values ​​(such as the average value of neighboring points) based on actual conditions.

[0036] Assign an accurate timestamp to each set of data to ensure that the timestamps of different parameter data are consistent. If there are different data collection frequencies, they can be aligned by interpolation or downsampling. Arrange the preprocessed data in chronological order to form a continuous time series. This step ensures that all parameter data can be analyzed synchronously and lays the foundation for subsequent time series modeling and analysis. Divide the time series data into different time periods or operating conditions according to actual needs. It can be segmented according to fixed time intervals, specific events (such as the end of the charge and discharge cycle), or battery state changes (such as SOC changes exceeding a certain threshold) to analyze battery performance at different stages. Store the sorted time series data in a database to ensure data traceability and security. It can be managed using a structured database (such as SQL) or a time series database (such as InfluxDB).

[0037] S2: Establish a linear aging model and a nonlinear aging model, and fit the model to the historical data of lead-acid batteries to obtain the fitting parameters of each model. Use residual analysis to identify whether there are nonlinear characteristics in the aging process.

[0038] Construct a linear aging model. If the capacity decay of the lead-acid battery is linearly related to time or number of cycles, the expression of the linear aging model is: ; Where C(t) is the battery capacity at time t, is the initial capacity, is the decay rate; a nonlinear aging model is constructed to capture complex aging behaviors, and the model expression is: ;in, is the nonlinear decay rate;

[0039] Determine the input variables of the model, including the number of charge and discharge cycles, temperature, and depth of discharge (DOD). Determine the output variables of the model, including the remaining capacity C(t) of the battery and the battery state of health (SOH). Divide the historical data into a training set and a validation set. The training set is used for model fitting, and the validation set is used to evaluate the generalization ability of the model.

[0040] The least squares method is used to fit the model and the best fitting parameters of the linear model are obtained by minimizing the sum of squares of the errors between the predicted values ​​and the actual values. and , the expression is: ; According to the fitted linear model, calculate the predicted value of battery capacity corresponding to each time point t , according to the actual value C(t) and the model prediction value The difference between the two is used to calculate the residual at each time point , the expression is: ; Arrange the residual calculation results of all time points into a residual sequence, draw a residual graph, with time t as the horizontal axis and residual e(t) as the vertical axis, and observe whether the residual in the residual graph has a trend of increasing or decreasing over time. If the residual shows a significant upward or downward trend instead of a random distribution, it indicates that there are nonlinear characteristics in the battery aging process.

[0041] S3: When nonlinear characteristics exist in the aging process, the accuracy of the nonlinear model prediction is evaluated according to the fluctuation of the model fit goodness and the stability of the prediction performance. According to the evaluation results, it is divided into high-accuracy nonlinear model and low-accuracy nonlinear model.

[0042] By analyzing fluctuations in the model's goodness of fit, you can identify conditions or time periods under which the model performs poorly. This helps determine whether the model is appropriate for a specific nonlinear aging process and avoids significant errors in certain critical conditions. Fluctuations in goodness of fit may reveal that the model is overfitting or underfitting in certain areas. By analyzing these fluctuations, you can decide whether to simplify or increase the model complexity to more accurately capture nonlinear characteristics.

[0043] The local fitting fluctuation index is generated according to the fluctuation of the model goodness of fit. The method for obtaining the local fitting fluctuation index is:

[0044] Obtain the model goodness of fit value in the time period s in real time, and construct the corresponding time series recorded as R(m), where m represents time. Set the initial sliding window size W0. This window size should be small enough to capture local changes, but not too small to avoid introducing too much noise. Generally, 5 to 20 data points are selected as the initial window size. W0 is the width of the initial window, which is usually selected based on experience or data characteristics.

[0045] At the mth moment, use the current sliding window size W(m) to calculate the mean of the goodness-of-fit values ​​within the window , the expression is: ; and the standard deviation of the goodness-of-fit values , ; In the formula, is the goodness of fit value of the i-th model in the time series; the standard deviation of the goodness of fit in the window is compared with the standard deviation reference threshold of the goodness of fit under the preset standard conditions, and the size of the sliding window is adjusted. The expression is: ; Where k is the adjustment coefficient, which controls the sensitivity of window size changes; is the reference value of the standard deviation, which is usually taken as the standard deviation or empirical value within the initial window. ), calculate the local fitting fluctuation index LF, the expression is: ; The larger the local fit fluctuation index, the lower the prediction accuracy of the nonlinear model in this local area. This is because the local fit fluctuation index reflects the volatility of the goodness of fit within the local time window. When the local fit fluctuation index is large, it means that the model is not stable in fitting the actual data during this period and may not accurately capture the nonlinear characteristics in the data. This instability usually means that the model performs poorly when dealing with complex changes or mutations, resulting in increased prediction errors. Therefore, a high local fit fluctuation index indicates that the model needs to be further optimized or adjusted to improve the prediction accuracy in this area.

[0046] The forecast adjustment response index is generated according to the stability of the model prediction performance. The method for obtaining the forecast adjustment response index is:

[0047] In the absence of disturbances, use the model to predict Q at future time points and record the baseline prediction value. Select input variables (such as temperature, load current, etc.) and at a certain moment Apply mutation or disturbance, record the input variable value after disturbance, and generate a new input sequence; after introducing disturbance, the model will generate a new prediction value based on the new input sequence, which is recorded as the initial prediction value after disturbance , calculate the error between the initial prediction value after the disturbance and the baseline prediction value , which represents the initial response of the model to the mutation, is expressed as: ; The model is divided into several time steps The predicted value generated after adjusting the disturbance after +ΔQ is marked as , calculate the error between the adjusted predicted value and the actual value , which represents the prediction accuracy after model adjustment, is expressed as: ; The computational model is disturbed The time required for the prediction error to stabilize within a small range , that is, the time required for the error to stabilize when the change in the prediction error is less than the threshold, and the prediction adjustment response index is calculated, and the expression is: ; In the formula, It is the prediction adjustment response index. The larger the PAR value, the faster the model adjusts when responding to sudden changes or nonlinear changes, and the higher the accuracy of the adjusted prediction.

[0048] The larger the forecast adjustment response index, the stronger the forecast adjustment ability of the nonlinear model in response to sudden changes in external conditions or nonlinear changes. Specifically, a larger PAR value indicates that the model is able to quickly adjust its forecasts after being disturbed and significantly reduce the forecast error. This means that the model is not only able to respond quickly to changes, but also provides more accurate forecasts after adjustments. Therefore, the larger the PAR, the higher the forecast accuracy and adaptability of the model when dealing with complex and dynamic environments.

[0049] The local fitting volatility index and the predicted adjusted response index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the accuracy value label predicted by the nonlinear model for each group of first eigenvectors as the prediction target, and takes minimizing the sum of prediction errors of the accuracy value labels predicted by all nonlinear models as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The accuracy value of the nonlinear model prediction is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0050] The method for obtaining the accuracy value of the nonlinear model prediction is: from the first eigenvector training data of the trained machine learning model, obtain the corresponding function expression: ; In the formula, is the output function of the model, is the local fitted volatility index, To predict the adjusted response index, is the accuracy value of the nonlinear model prediction.

[0051] The acquired accuracy value of the nonlinear model prediction is compared with a preset reference threshold value of the accuracy value of the nonlinear model prediction. If the accuracy value of the nonlinear model prediction is greater than or equal to the preset reference threshold value of the accuracy value of the nonlinear model prediction, it indicates that the accuracy of the nonlinear model prediction is high. In this case, a high-accuracy prediction signal is generated and the model is classified as a high-accuracy nonlinear model. If the accuracy value of the nonlinear model prediction is less than the preset reference threshold value of the nonlinear model prediction accuracy, it indicates that the accuracy of the nonlinear model prediction is low. In this case, a low-accuracy prediction signal is generated and the model is classified as a low-accuracy nonlinear model.

[0052] S4: Predict the aging of lead-acid batteries based on a high-accuracy nonlinear model, and compare the predicted results with the actual operating data to determine the deviation between the predicted results and the actual data.

[0053] Collect the key operating parameters of the current lead-acid battery, such as voltage, current, temperature, number of charge and discharge cycles, SOC (state of charge), etc., and input these data into the trained nonlinear model. Use the nonlinear model to predict the aging state of the battery in the future and obtain the predicted time series of the battery's remaining capacity, internal resistance change, or other related health indicators. Through sensors or monitoring systems, continuously collect the actual operating data of the lead-acid battery and record the actual aging state of the battery in the same time period, including remaining capacity, internal resistance, SOC, etc. Organize the actual data collected to ensure that the data is complete and the time series is aligned with the model prediction data.

[0054] Calculate the deviation between the predicted value and the actual value at each time point, analyze the calculated deviation and generate a deviation trend index. According to the deviation trend index, judge the deviation between the predicted result and the actual data. The method for obtaining the deviation trend index is as follows:

[0055] Calculate the deviation between the predicted value and the actual value at each time point and construct the corresponding time series , where v = 1, 2, ..., N, N is the number of time points; for the deviation time series Perform linear regression and fit the linear model. The expression is: ; Where M is the intercept, R is the regression slope, which represents the linear rate of change of the deviation over time, ϵ(v) is the regression residual, and the deviation trend index is calculated as follows: ; In the formula, is the standard deviation of the deviation sequence, which is used to normalize the slope and eliminate the impact of the deviation amplitude on the trend. is the deviation trend index.

[0056] BTI>0: Positive values ​​indicate that the bias increases over time, which may indicate that the model's prediction accuracy gradually deteriorates over time. BTI<0: Negative values ​​indicate that the bias decreases over time, which may indicate that the model's prediction accuracy gradually improves over time. BTI≈0: Close to zero indicates that the bias does not change significantly over time and the prediction error is relatively stable.

[0057] Through BTI, the prediction performance of the model in different time periods can be diagnosed, especially the changing trend of the prediction accuracy of the model in long-term use can be identified. According to the value of BTI, it can be decided whether the model parameters or structure need to be adjusted to correct the systematic error, especially when the BTI value is large, indicating that the model needs further optimization.

[0058] S5: According to the deviation between the prediction results and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic to improve the accuracy of capacity prediction.

[0059] The accuracy value and deviation trend index predicted by the nonlinear model are used as the input of fuzzy logic, and the amplitude of model parameter adjustment is used as the output;

[0060] The deviation trend index is fuzzified and defined as three fuzzy sets of "low deviation", "medium deviation" and "high deviation";

[0061] The prediction accuracy index of the nonlinear model is fuzzified and defined as three fuzzy sets: "low accuracy", "medium accuracy" and "high accuracy".

[0062] The output variable represents the magnitude of the model parameter adjustment, which is defined as three fuzzy sets: “no adjustment”, “small adjustment” and “large adjustment”.

[0063] Construct a fuzzy rule set, including: Rule 1: If the bias is "high bias" and the accuracy is "low accuracy", then the parameter adjustment is "large adjustment". Rule 2: If the bias is "medium bias" and the accuracy is "medium accuracy", then the parameter adjustment is "small adjustment". Rule 3: If the bias is "low bias" and the accuracy is "high accuracy", then the parameter adjustment is "no adjustment".

[0064] The input bias and accuracy data are converted into corresponding fuzzy sets through fuzzy membership functions. For example, the value of bias is converted into a fuzzy degree from "low bias" to "high bias".

[0065] By using the defined fuzzy rule set and combining the input fuzzy variables, the fuzzy value of the output variable parameter adjustment range is obtained through fuzzy reasoning (such as Mamdani fuzzy reasoning method).

[0066] The result of fuzzy reasoning is converted into a specific parameter adjustment amount, and a clear adjustment amount is calculated using a defuzzification method (such as the centroid method).

[0067] According to the defuzzification results, adjust the parameters of the nonlinear model. For example, if the defuzzification result is "big adjustment", make major changes to the key parameters of the model (such as learning rate, regularization coefficient, model complexity, etc.); if the result is "small adjustment" or "no adjustment", make corresponding fine-tuning of the parameters or keep them unchanged. After adjusting the parameters, retrain the nonlinear model to adapt to the new parameter settings, thereby improving the accuracy of capacity forecasting.

[0068] In this application, a more refined and adaptive model optimization can be achieved by dynamically adjusting the nonlinear model parameters through fuzzy logic based on the deviation trend index and the nonlinear model prediction accuracy. Specifically, this method can flexibly and automatically adjust key parameters according to the actual prediction error and model performance to ensure that the model can maintain high prediction accuracy under different conditions. Ultimately, this not only improves the accuracy of capacity forecasts, but also enhances the robustness of the model, reduces the impact of systematic errors on the prediction results, and thus maintains the stable performance of the model during long-term operation.

[0069] In this embodiment, linear and nonlinear aging models are established by collecting and preprocessing historical data, and nonlinear features in the aging process are identified through residual analysis. When nonlinear features exist, the prediction accuracy of the nonlinear model is further evaluated and divided into high-accuracy and low-accuracy models. The high-accuracy model is used to predict battery aging, and the prediction results are compared and analyzed with the actual data. Finally, fuzzy logic is used to dynamically adjust the model parameters according to the deviation and accuracy, thereby continuously improving the accuracy of capacity prediction and the adaptability of the model.

[0070] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0071] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0072] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for predicting the remaining capacity of a lead-acid battery in a DC system, characterized in that: The steps include: S1: Collect historical data of lead-acid batteries at different time points and under different working conditions, pre-process the historical data, and organize them according to time series; S2: Establish a linear aging model and a nonlinear aging model, and fit the model to the historical data of lead-acid batteries to obtain the fitting parameters of each model. Use residual analysis to identify whether there are nonlinear characteristics in the aging process; Step S2 specifically includes: constructing a linear aging model. If the capacity decay of the lead-acid battery is linearly related to time or the number of cycles, the expression of the linear aging model is: ; Where C(t) is the battery capacity at time t, is the initial capacity, is the decay rate; a nonlinear aging model is constructed to capture complex aging behaviors, and the model expression is: ;in, is the nonlinear decay rate; Determine the input variables of the model, including the number of charge and discharge cycles, temperature, and discharge depth; determine the output variables of the model, including the remaining capacity C(t) of the battery and the battery health status indicator; divide the historical data into a training set and a validation set, use the least squares method to fit the model, and obtain the best fitting parameters of the linear model by minimizing the sum of squares of the error between the predicted value and the actual value and , the expression is: ; According to the fitted linear model, calculate the predicted value of battery capacity corresponding to each time point t , according to the actual value C(t) and the model prediction value The difference between them is used to calculate the residual e(t) at each time point. The expression is: ; The residual calculation results of all time points are sorted into a residual sequence, and a residual graph is drawn with time t as the horizontal axis and residual e(t) as the vertical axis. If the residual shows a significant upward or downward trend instead of a random distribution, it indicates that there are nonlinear characteristics in the battery aging process; S3: When nonlinear characteristics exist in the aging process, the accuracy of the nonlinear model prediction is evaluated according to the fluctuation of the model fit goodness and the stability of the prediction performance, and it is divided into high-accuracy nonlinear model and low-accuracy nonlinear model according to the evaluation results; The step S3 specifically includes: generating a local fitting fluctuation index according to the fluctuation of the model fitting goodness, and the method for obtaining the local fitting fluctuation index is: Obtain the model goodness of fit value within the time period s in real time, and construct the corresponding time series recorded as R(m), where m represents time. Set the initial sliding window size W0. At the mth moment, use the current sliding window size W(m) to calculate the mean of the goodness of fit value in the window. , the expression is: ; and the standard deviation of the goodness-of-fit values , ; In the formula, R(i) is the goodness of fit value of the i-th model in the time series; the standard deviation of the goodness of fit in the window is compared with the standard deviation reference threshold of the goodness of fit under the pre-set standard conditions, and the size of the sliding window is adjusted. The expression is: ; where k is the adjustment coefficient, is the reference value of the standard deviation. According to the adjusted sliding window W(m), the local fitting fluctuation index LF is calculated. The expression is: ; The forecast adjustment response index is generated according to the stability of the model prediction performance. The method for obtaining the forecast adjustment response index is: In the absence of disturbance, use the model to predict the future time point Q, record the baseline prediction value, and select the input variable at a certain moment. Apply disturbance, record the input variable value after disturbance, and generate a new input sequence; after introducing disturbance, the model will generate a new prediction value based on the new input sequence, which is recorded as the initial prediction value after disturbance , calculate the error between the initial prediction value after the disturbance and the baseline prediction value , the expression is: ; The model is divided into several time steps + The predicted value generated after adjusting the disturbance is marked as , calculate the error between the adjusted predicted value and the actual value, Adjustment Error, which represents the prediction accuracy of the model after adjustment. The expression is: Adjustment Error= - ; The computational model is disturbed The time required for the prediction error to reach a stable range , that is, the time required for the error to stabilize when the change in the prediction error is less than the threshold, and the prediction adjustment response index is calculated, and the expression is: ;Where PAR is the predicted adjusted response index; S4: Predict the aging of lead-acid batteries based on a high-accuracy nonlinear model, and compare the predicted results with the actual operating data to determine the deviation between the predicted results and the actual data; S5: According to the deviation between the prediction results and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic to improve the accuracy of capacity prediction.

2. The method for predicting the remaining capacity of a lead-acid battery in a DC system according to claim 1, characterized in that: The local fitting volatility index and the predicted adjusted response index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model uses the accuracy value label predicted by the nonlinear model for each group of first eigenvectors as the prediction target, and takes minimizing the sum of prediction errors of the accuracy value labels predicted by all nonlinear models as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The accuracy value of the nonlinear model prediction is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

3. The method for predicting the remaining capacity of a lead-acid battery in a DC system according to claim 2, characterized in that: The acquired accuracy value of the nonlinear model prediction is compared with a preset reference threshold value of the accuracy value of the nonlinear model prediction. If the accuracy value of the nonlinear model prediction is greater than or equal to the preset reference threshold value of the accuracy value of the nonlinear model prediction, it indicates that the accuracy of the nonlinear model prediction is high. In this case, a high-accuracy prediction signal is generated and the model is classified as a high-accuracy nonlinear model. If the accuracy value of the nonlinear model prediction is less than the preset reference threshold value of the nonlinear model prediction accuracy, it indicates that the accuracy of the nonlinear model prediction is low. In this case, a low-accuracy prediction signal is generated and the model is classified as a low-accuracy nonlinear model.

4. The method for predicting the remaining capacity of a lead-acid battery in a DC system according to claim 1, characterized in that: In S4, the deviation between the predicted value and the actual value at each time point is calculated, and the calculated deviation is analyzed to generate a deviation trend index. The deviation between the predicted result and the actual data is determined according to the deviation trend index. The method for obtaining the deviation trend index is: Calculate the deviation between the predicted value and the actual value at each time point and construct the corresponding time series , where v = 1, 2, ..., N, N is the number of time points; for the deviation time series Perform linear regression and fit the linear model. The expression is: =M+Rv+ϵ(v); where M is the intercept, R is the regression slope, which represents the linear rate of change of the deviation over time, and ϵ(v) is the regression residual. The deviation trend index is calculated as follows: ; In the formula, is the standard deviation of the deviation series, used to normalize the slope, and BTI is the deviation trend index.

5. The method for predicting the remaining capacity of a lead-acid battery in a DC system according to claim 1, characterized in that: In S5, according to the deviation between the prediction result and the actual data and the accuracy of the nonlinear model prediction, the parameters of the nonlinear model are dynamically adjusted using fuzzy logic, specifically: The accuracy value and deviation trend index of the nonlinear model prediction are used as the input items of fuzzy logic, and the amplitude of the model parameter adjustment is used as the output item; the deviation trend index, the prediction accuracy of the nonlinear model and the amplitude of the model parameter adjustment are fuzzified; Construct a fuzzy rule set, transform the input deviation and accuracy data into corresponding fuzzy sets through fuzzy membership functions; use the defined fuzzy rule set, combined with the input fuzzy variables, and through fuzzy reasoning, obtain the fuzzy value of the output variable parameter adjustment range; The result of fuzzy reasoning is converted into a specific parameter adjustment amount, and a clear adjustment amount is calculated using a defuzzification method; according to the defuzzification result, the parameters of the nonlinear model are adjusted.

Images