Plateau energy storage battery monitoring method and system based on physical data dual drive

Abstract

The invention relates to a plateau energy storage battery monitoring method and system based on physical data dual drive, and the method comprises the steps: firstly constructing a battery internal resistance physical model considering plateau environment correction, predicting the internal resistance in real time, and generating a failure prior probability; meanwhile, a support vector machine (SVM) and a random forest (RF) are used for constructing a data driving model, battery operation characteristics are extracted, and fault probability and uncertainty are output. On the basis, a Bayesian fusion method is used, the weights of a physical model and a data driving model are dynamically adjusted, the battery failure posterior probability is calculated, and the total uncertainty is quantified. And finally, setting a decision threshold, and when the posterior probability exceeds the threshold, triggering early warning and taking protection measures. The method effectively fuses the theoretical advantages of a physical model and the actual data mining capability of a data driving model, adapts to the plateau environment, and provides guarantee for safe and stable operation of the energy storage battery of the plateau photovoltaic power station.

Description

Technical Field

[0001] The present invention relates to the field of reliability monitoring of energy storage batteries in plateau areas, and in particular to a plateau energy storage battery monitoring method and system based on physical data dual drive. Background Art

[0002] With the growing global demand for clean energy, photovoltaic power stations are increasingly being deployed in plateau regions. Plateau regions boast abundant solar resources, providing unique conditions for photovoltaic power generation. However, the unique characteristics of the plateau environment pose significant challenges to the reliability and performance of energy storage batteries. Harsh environmental factors in plateau regions, such as low air pressure, thin air, large diurnal temperature swings, high UV intensity, and potentially low temperatures, severely impact the chemical reaction processes, thermal management, and mechanical stability of energy storage batteries. For example, low air pressure can cause internal gas expansion in batteries, compromising their sealing and internal structure; while large diurnal temperature swings cause thermal expansion and contraction of battery materials, accelerating battery aging. These combined environmental factors result in significantly higher failure rates for energy storage batteries in plateau regions than in plains, impacting the stability and power generation efficiency of the entire photovoltaic power station. Traditional battery monitoring methods, primarily based on single physical models or statistical methods, fail to fully and accurately reflect the true state of batteries in the complex plateau environment. Physical models often exhibit parameter deviations when describing the electrochemical reactions within batteries due to the unique characteristics of the plateau environment. While data-driven models can capture some patterns in actual operating data, they lack a deep understanding of the battery's underlying physical characteristics, potentially leading to inaccurate predictions when faced with extreme plateau operating conditions. Therefore, a reliability monitoring method for energy storage batteries that comprehensively considers physical mechanisms and actual operating data and is adapted to the plateau environment is urgently needed to ensure the safe and stable operation of photovoltaic power plants in plateaus. Summary of the Invention

[0003] The present invention aims to provide a plateau energy storage battery monitoring method based on dual-drive of physical data. The method realizes accurate and real-time monitoring of the reliability of plateau energy storage batteries by deeply integrating the data-driven method with the physical mechanism model and introducing the Bayesian fusion framework. The present invention fully considers the impact of plateau environmental factors on battery performance, uses physical models to provide prior knowledge of battery reliability status, and uses data-driven models to mine potential fault characteristics in actual operation data. The advantages of the two models are then dynamically integrated through the Bayesian method to quantify uncertainty and adaptively adjust weights. At the decision-making level, potential faults are warned in a timely manner based on the fused posterior probability, and corresponding protection measures are triggered, thereby significantly improving the operational safety and stability of plateau energy storage batteries, reducing the risk of failure, and ensuring efficient power generation and stable operation of plateau photovoltaic power stations.

[0004] The specific plan is as follows:

[0005] A plateau energy storage battery monitoring method based on dual-drive physical data includes the following steps:

[0006] S1. Build a battery internal resistance physical model that takes plateau environment correction into account, predict internal resistance in real time, and generate a priori probability of failure.

[0007] S2. Use the support vector machine (SVM) model and random forest (RF) model to build a data-driven model, extract battery operation characteristics, and output failure probability and uncertainty;

[0008] S3. Use the Bayesian fusion method to dynamically adjust the weights of the battery internal resistance physical model and the data-driven model, calculate the posterior probability of battery failure, and quantify the total uncertainty;

[0009] S4. Set a decision threshold. When the posterior probability exceeds the threshold, trigger an early warning and take protective measures.

[0010] Furthermore, step S1 specifically includes: processing the output of the physical model and Bayesian fusion

[0011] (1) Output results and acquisition methods of energy storage battery physical model

[0012] ① Thermal characteristic model of lithium-ion batteries based on internal resistance (plateau environment correction model)

[0013] Output battery internal resistance prediction value R b :

[0014] R b =R b0 +K·(T s -T s0 )+ΔR b,alt

[0015] in:

[0016] R b It is the predicted value of the battery internal resistance under the current operating conditions, reflecting the comprehensive resistance characteristics of the battery's internal electrodes and electrolyte, and the unit is ohm (Ω);

[0017] R b0 is the reference surface temperature T s0 The battery internal resistance reference value under standard pressure is obtained through experimental calibration and the unit is ohm (Ω);

[0018] K is the internal resistance temperature coefficient, which indicates the rate at which the battery's internal resistance changes with surface temperature. It is obtained through experimental fitting and is expressed in ohms per degree Celsius (Ω / °C).

[0019] T s The actual surface temperature of the battery is collected in real time by the temperature sensor in degrees Celsius (℃);

[0020] T s0 It is the reference surface temperature, usually the surface temperature of the battery under standard test conditions, in degrees Celsius (℃);

[0021] ΔR b,alt It is the plateau environment correction term, which is used to quantify the impact of plateau low pressure on the battery internal resistance. The calculation formula is:

[0022]

[0023] in:

[0024] β alt is the plateau environmental sensitivity coefficient, which reflects the sensitivity of the battery internal resistance to changes in air pressure. It is obtained through experimental fitting and is dimensionless.

[0025] P a,sea is the standard atmospheric pressure at sea level, usually 101.325 kilopascals (kPa);

[0026] P a,alt It is the actual air pressure on the plateau, collected in real time by the air pressure sensor, and its unit is kilopascal (kPa).

[0027] ②Digital twin parameter optimization

[0028] Output optimized physical model parameters (such as K * 、 ): Minimize the objective function J by particle swarm optimization (PSO) algorithm:

[0029]

[0030] in:

[0031] R b,measured (t i ) is at time t i Real-time collected battery internal resistance measurement value, in ohms (Ω);

[0032] R b,model (t i ) is at time t i The predicted value of the battery internal resistance calculated based on the physical model, in ohms (Ω);

[0033] N is the number of data points collected to evaluate the deviation between the model predictions and the actual measurements.

[0034] (2) Probabilistic processing of physical model output

[0035] Convert the physical model output into a probability distribution:

[0036] Battery internal resistance R b :Assume normal distribution where μ phy =R b , that is, the mean of the distribution is the predicted value of internal resistance calculated by the physical model; the variance Determined by model error analysis, taking into account the additional uncertainty brought by the plateau environment:

[0037]

[0038] in:

[0039] It is the basic model error variance, obtained through statistical analysis of historical data, reflecting the degree of deviation between the physical model prediction value and the actual value, and the unit is ohm square (Ω 2 );

[0040] γ alt is the plateau environmental uncertainty coefficient, which is used to quantify the effect of the plateau environment on the uncertainty of model predictions. It is determined through experiments or historical data analysis and is dimensionless.

[0041] H is the actual altitude, which is collected in real time by an altitude sensor or obtained through geographic information system (GIS) data, in meters (m);

[0042] H0 is the reference altitude, which is usually a known altitude used for model calibration, in meters (m).

[0043] (3) Construction of prior probability in Bayesian fusion

[0044] The physical model output is used as a prior distribution, considering the impact of the plateau environment on the failure threshold:

[0045]

[0046] in:

[0047] P(fail) is the prior probability of battery failure, which indicates the possibility of battery failure after considering the prediction results of the physical model and the correction of the plateau environment, and its value range is [0,1];

[0048] Φ(·) is the standard normal distribution cumulative distribution function, which is used to convert the standardized internal resistance deviation into a probability value and is dimensionless;

[0049] R b,crit,alt is the battery internal resistance failure threshold in plateau environments. When the battery internal resistance exceeds this value, it is considered that the battery may be in a failed state. The calculation formula is:

[0050]

[0051] in:

[0052] R b,crit It is the failure threshold of the battery's internal resistance under standard conditions, determined through experiments or data provided by the manufacturer, and is expressed in ohms (Ω).

[0053] δ alt It is the plateau environment failure threshold adjustment coefficient, which is used to quantify the impact of the plateau environment on the battery internal resistance failure threshold. It is determined through experiments or data analysis and is dimensionless.

[0054] Furthermore, step S2 specifically includes: processing the output results of the data-driven model and Bayesian fusion

[0055] (1) Output results of data-driven model

[0056] ①Support Vector Machine (SVM)

[0057] Output failure probability p svm : Convert the classification results into probabilities through Platt Scaling:

[0058]

[0059] in:

[0060] p svm The battery failure probability prediction value based on the SVM model has a value range of [0,1];

[0061] f(x) is the decision function output of the SVM model, and the calculation formula is:

[0062]

[0063] in:

[0064] α i is the Lagrange multiplier, determined by the SVM training process, dimensionless;

[0065] y i is the category label of the training sample (normal or faulty), with a value of +1 or -1;

[0066] K(x i ,x) is the kernel function, which is used to map the input features to a high-dimensional space and calculate the similarity between samples. Commonly used methods include radial basis function (RBF) and are dimensionless;

[0067] x i is the eigenvector of the i-th support vector;

[0068] x is the current input feature vector, which contains the plateau-corrected battery voltage, current, temperature and other parameters;

[0069] b is the bias term, which is determined by the SVM training process and is dimensionless;

[0070] The SVM model converts the classification results into failure probability p through Platt Scaling svm , the formula is A and B are the parameters of Platt Scaling, which are calibrated through cross-validation and are used to convert the decision function output of SVM into a dimensionless probability value.

[0071] ② Random Forest (RF)

[0072] Output failure probability P rf : Class probabilities based on majority voting:

[0073]

[0074] in:

[0075] P rf is the battery failure probability prediction value based on the RF model, with a value range of [0,1];

[0076] T is the number of decision trees in the random forest. A larger value is usually chosen to improve the stability and accuracy of the model. It is a positive integer.

[0077] I(·) is the indicator function, which takes the value of 1 when the prediction result of decision tree t is failure, and 0 otherwise;

[0078] f t (x) is the prediction result of the t-th decision tree, which determines whether the battery status corresponding to the current input feature vector x is faulty.

[0079] (2) Uncertainty Quantification of Probabilistic Output

[0080] SVM variance estimation: based on the confidence of support vector distance, considering the influence of plateau environment:

[0081]

[0082] in:

[0083] It is the variance of the SVM output probability after considering the plateau environment correction, which is used to quantify the uncertainty of the model prediction and the unit is dimensionless;

[0084] N sv is the number of support vectors, that is, the number of samples selected during training to construct the decision boundary, which is a positive integer.

[0085] RF variance estimation: Based on the variance of the Bootstrap sample, considering the influence of the plateau environment:

[0086]

[0087] in:

[0088] It is the variance of the RF output probability after considering the plateau environment correction, which is used to quantify the uncertainty of the model prediction and has dimensionless units;

[0089] It is the predicted value of the failure probability output by the t-th decision tree, and its value range is [0,1].

[0090] (3) Likelihood function construction in Bayesian fusion

[0091] The data-driven model output is used as a likelihood function, taking into account the plateau environment correction:

[0092]

[0093] in:

[0094] P(D|fail) is the likelihood probability of observing the current data D in the case of battery failure. It is used to quantify the relationship between the output of the data-driven model and the battery failure state. The value range is [0,1];

[0095] It is a normal distribution, indicating that the probability value output by the data-driven model follows a mean of p data,alt , the variance is Normal distribution;

[0096] p data,alt is the predicted value of the failure probability of the fused data-driven model, and the calculation formula is:

[0097]

[0098] in:

[0099] It is the weight of the SVM model after considering the plateau environment correction, reflecting the credibility of the SVM model prediction results;

[0100] It is the weight of the RF model after considering the plateau environment correction, reflecting the credibility of the RF model prediction results.

[0101] Furthermore, step S3 specifically includes: the mechanism of action of the Bayesian fusion model

[0102] (1) Calculation of posterior probability

[0103] Based on Bayes' theorem:

[0104]

[0105] Substituting the physical model prior P(fail) and data-driven likelihood P(D|fail) corrected for the plateau environment, we get:

[0106]

[0107] in:

[0108] P(fail|D) refers to the posterior probability of battery failure when data D is observed. Its value range is [0,1] and it is the final reliability assessment result after integrating the physical model and data-driven model.

[0109] P(D) is the marginal probability of data D, which is used for normalization to ensure that the sum of the posterior probabilities is 1;

[0110] ensemble mean

[0111] ensemble variance

[0112] (2) Dynamic weight adjustment

[0113] The weight coefficient is adaptively adjusted according to the environmental complexity Cenv, taking into account plateau environmental factors:

[0114]

[0115] in:

[0116] ω phy,alt (C env ) is the dynamic weight of the physical model after considering environmental complexity and plateau environment correction;

[0117] ω data,alt (C env ) is the dynamic weight of the data-driven model after considering environmental complexity and plateau environment correction;

[0118] is the initial weight value, which is usually determined based on prior knowledge or initial model performance evaluation and is dimensionless;

[0119] λ is the weight adjustment coefficient, which controls the influence of environmental complexity on the weight. It is determined through experiments or data analysis and is dimensionless.

[0120] C env Environmental complexity is calculated based on real-time data, taking into account various environmental factors such as temperature change rate, charge and discharge rate fluctuations, and air pressure changes. It is dimensionless.

[0121] ζ alt It is the plateau environment weight adjustment coefficient, which is used to quantify the impact of the plateau environment on the dynamic adjustment of weights. It is determined through experiments or data analysis and is dimensionless.

[0122] (3) Uncertainty Propagation

[0123] Total uncertainty after fusion:

[0124]

[0125] in:

[0126] U fuse,alt is the total uncertainty after fusion, which is used to quantify the confidence interval of the final posterior probability and is dimensionless;

[0127] is the uncertainty of the physical model, which is the width of the confidence interval of the physical model prediction value after considering the plateau environment correction;

[0128] is the uncertainty of the data-driven model, which is the width of the confidence interval of the predicted value of the data-driven model after considering the plateau environment correction;

[0129] σ phy,alt , σ data,alt are the standard deviations of the physical model and the data-driven model after correction for the plateau environment, respectively, and the units are dimensionless;

[0130] The critical values ​​of the t distribution for the corresponding physical model and data-driven model are used to construct confidence intervals with a given significance level α and degrees of freedom df. phy,alt 、df data,alt Related, dimensionless.

[0131] Furthermore, step S4 specifically includes:

[0132] Set a reasonable decision threshold α(C env ,H), the calculated posterior probability P(fail|D) and the decision threshold α(C env ,H) for comparison. If P(fail|D) exceeds the decision threshold, it is judged that the energy storage battery may have a failure risk, triggering a warning signal, and taking corresponding protection measures to ensure the safe operation of the energy storage battery.

[0133] A high-altitude energy storage battery monitoring system based on physical data dual drive, based on sensors, data acquisition modules, data processing modules, model calculation modules and decision modules, including:

[0134] ① Input layer: The data acquisition module collects environmental parameters (temperature, air pressure, altitude), electrical signals (battery voltage, current), and thermal parameters (battery surface temperature) monitored by sensors in real time and sends them to the data processing module.

[0135] ② Physical model layer: The battery internal resistance Rb after plateau environment correction is calculated through the model calculation module to generate the prior probability P(fail).

[0136] ③Data driving layer: extract the characteristics of plateau environment correction (voltage ripple rate γ ripple , charge and discharge current fluctuation rate γ current , temperature gradient ▽T, output failure probability p data,alt .

[0137] ④ Bayesian fusion layer: Dynamically weighted fusion of prior and likelihood through the model calculation module to calculate the posterior probability P(fail|D). ⑤ Decision layer: Decision module is used to make judgments. If P(fail|D)>α(C env ,H), triggering early warning or protection measures, such as limiting the charge and discharge power, starting the heating / cooling system, etc., where α(C env ,H) is the decision threshold considering the plateau environment.

[0138] The beneficial effects of the present invention are:

[0139] (1) Improving the battery internal resistance physical model based on the characteristics of the plateau environment can accurately reflect the changes in battery internal resistance in the plateau environment, reduce prediction deviation, improve failure prediction accuracy, and provide early warning to ensure safety.

[0140] (2) Use support vector machine (SVM) and random forest (RF) to improve the data-driven model, mine the fault characteristics in the actual operation data, comprehensively capture complex fault modes, improve the accuracy of fault probability prediction, and enhance the reliability of reliability assessment.

[0141] (3) The Bayesian fusion method is used to improve the model fusion method, integrate the advantages of physical models and data-driven models, dynamically adjust weights, quantify total uncertainty, accurately judge battery failure risks, and improve assessment accuracy and robustness.

[0142] (4) Based on the posterior probability, the decision threshold is set and early warning protection measures are taken to effectively avoid the occurrence of faults, ensure the safe operation of energy storage batteries, and reduce the impact on the stability of photovoltaic power stations.

[0143] (5) Construct a reliability monitoring system for energy storage batteries in plateau areas. Each module works together to achieve comprehensive and real-time reliability monitoring, improve monitoring efficiency and accuracy, and provide strong support for the operation and management of plateau photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0144] Figure 1 This is a system block diagram of the energy storage battery reliability monitoring method in plateau areas, showing the connection relationship and data flow between modules.

[0145] Figure 2 This is a schematic diagram of the energy storage battery physical model, which describes in detail the physical relationship between the battery's internal resistance and the influence of temperature and plateau environment.

[0146] Figure 3 This is a training flowchart for data-driven models, which illustrates how to use historical data to train support vector machines and random forest models.

[0147] Figure 4 This is the computational flow chart of the Bayesian fusion model, showing the fusion process of the physical model and data-driven model outputs and the calculation method of the posterior probability. DETAILED DESCRIPTION

[0148] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0149] As shown in the figure, the present invention provides a plateau energy storage battery monitoring method based on physical data dual drive, including the following steps:

[0150] 1. Monitoring system construction

[0151] Build an energy storage battery monitoring system that includes sensors, data acquisition modules, data processing modules, model calculation modules, and decision modules. These modules work together to achieve comprehensive monitoring of the reliability of energy storage batteries in high-altitude photovoltaic power plants.

[0152] 2. Application of physical models

[0153] (1) Model parameter acquisition

[0154] ① Experimental calibration: In a laboratory environment, the energy storage battery is subjected to charge and discharge cycle tests. The internal resistance of the battery is measured under different temperature (such as 0°C, 20°C, 40°C, etc.) and air pressure (simulating the air pressure at different altitudes in the plateau, such as 80kPa, 70kPa, 60kPa, etc.).

[0155] ②Data fitting: Correlation analysis is performed on the measured internal resistance data with temperature, air pressure and other conditions. The internal resistance Rb0, internal resistance temperature coefficient K, and plateau environmental sensitivity coefficient β of the energy storage battery under different temperature, air pressure and other conditions are obtained through data fitting. alt and other physical model parameters.

[0156] (2) Model calculation and correction

[0157] ① Real-time calculation: During the actual operation, the data acquisition module collects the battery surface temperature Ts, the actual plateau air pressure P in real time a,alt and other parameters, and transmit them to the data processing module.

[0158] ② Internal resistance prediction: The data processing module uses the physical model formula R b =R b0 +K·(T s -T s0 )+ΔR b,alt , substitute the collected data into the model and calculate the predicted value of battery internal resistance R b Among them, the plateau environment correction term ΔR b,alt According to the formula Calculated and used to correct the impact of plateau environment on battery internal resistance.

[0159] ③Function of the model: The physical model is based on the physical properties of the battery and can accurately describe the relationship between the changes in the battery's internal resistance and temperature and plateau environment, providing a theoretical basis and prior knowledge for subsequent reliability assessment.

[0160] 3. Application of data-driven models

[0161] (1) Model training

[0162] ① Data Collection: Collect a large amount of operational data from energy storage batteries in high-altitude photovoltaic power plants, including sample data under normal operation and fault conditions. This data covers electrical and thermal parameters such as battery voltage, current, temperature, and internal resistance, as well as corresponding environmental parameters (such as temperature, air pressure, and altitude).

[0163] ② Feature extraction: Feature extraction is performed on the collected data. The extracted features include voltage ripple rate, charge and discharge current fluctuation rate, temperature gradient, etc. These features can reflect the battery's operating status and potential fault information.

[0164] ③ Model training: The extracted feature data is divided into training and test sets. The model is trained using the Support Vector Machine (SVM) and Random Forest (RF) algorithms. During training, algorithm parameters (such as the kernel function parameters of the SVM and the number of decision trees in the RF) are adjusted to enable the model to accurately predict battery failure probability. The trained model is validated and evaluated using the test set to ensure its accuracy and reliability.

[0165] (2) Model output and uncertainty quantification

[0166] ① Fault probability prediction: During the actual monitoring process, the data processing module inputs the real-time collected and processed data into the trained SVM and RF models. The SVM model converts the classification results into the fault probability p through Platt Scaling. svm , the formula is in Parameters A and B are calibrated through cross-validation. The RF model outputs the failure probability P based on the class probability of the majority vote. rf , the formula is Where I(·) is the indicator function.

[0167] ②Uncertainty quantification: Calculate the variance of the SVM and RF output probabilities based on the uncertainty quantification method of the model and The SVM variance estimation is based on the confidence of the support vector distance, taking into account the influence of the plateau environment. The formula is: RF variance estimation is based on the variance of the Bootstrap sample, taking into account the influence of the plateau environment, and the formula is:

[0168] ③Model function: The data-driven model can mine the potential failure characteristics and patterns of batteries based on actual operation data, identify and predict complex failure modes that cannot be accurately described by physical models, and provide data support for reliability assessment.

[0169] 4. Bayesian Fusion and Decision Making

[0170] (1) Bayesian fusion calculation

[0171] ① Construction of prior probability and likelihood function: The internal resistance prediction value output by the physical model and its uncertainty are converted into the prior probability P(fail), considering the impact of the plateau environment on the failure threshold. The formula is: where Φ(·) is the cumulative function of the standard normal distribution, is the battery internal resistance failure threshold in plateau environment. The failure probability and its uncertainty output by the data-driven model are constructed as the likelihood function P(D|fail), and the plateau environment correction is considered. The formula is: in Weight

[0172] ② Calculation of posterior probability: Based on Bayes’ theorem, the prior probability P(fail) and the likelihood function P(D|fail) are combined to calculate the posterior probability P(fail|D). The formula is Substituting the physical model prior and data-driven likelihood after plateau environment correction, the expanded value is:

[0173]

[0174] The fusion mean variance

[0175] ③ Dynamic weight adjustment: The weight coefficient is dynamically adjusted according to the environmental complexity Cenv and the altitude H. The formula is:

[0176]

[0177] Among them alt Adjusts the weight of the plateau environment. When the environment complexity and altitude increase, the weight of the physical model is reduced to enhance the robustness of data-driven performance.

[0178] ④ Uncertainty propagation: Calculate the total uncertainty after fusion. The formula is:

[0179]

[0180] in

[0181] ⑤Model function: The Bayesian fusion model can integrate the advantages of physical models and data-driven models. By dynamically adjusting weights and quantifying uncertainties, it can achieve accurate assessment of the reliability status of energy storage batteries and improve the accuracy and reliability of monitoring results.

[0182] (2) Decision-making and early warning

[0183] ① Set the decision threshold: According to the historical operation data and reliability requirements of the energy storage battery of the plateau photovoltaic power station, set a reasonable decision threshold α (C env ,H). This threshold takes into account the impact of factors such as environmental complexity and altitude on battery reliability.

[0184] ②Decision judgment: Compare the calculated posterior probability P(fail|D) with the decision threshold α(C env ,H) for comparison. If P(fail|D) exceeds the decision threshold, it is determined that the energy storage battery may have a failure risk, triggering a warning signal and taking corresponding protection measures, such as limiting the charge and discharge power, starting the heating / cooling system, etc., to ensure the safe operation of the energy storage battery.

[0185] ③Model function: The decision-making module makes timely and accurate decisions based on the output results of the Bayesian fusion model, ensuring the reliable operation of energy storage batteries in plateau environments, reducing failure risks, and improving the safety and stability of the system.

[0186] The above is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several adjustments and improvements can be made without departing from the core concept of the present invention, and these adjustments and improvements should also be regarded as the scope of protection of the present invention.

Claims

1. A plateau energy storage battery monitoring method based on physical data dual drive, characterized in that: The following steps are involved: S1. Build a battery internal resistance physical model that takes plateau environment correction into account, predict internal resistance in real time, and generate a priori probability of failure. S2. Use the support vector machine (SVM) model and random forest (RF) model to build a data-driven model, extract battery operation characteristics, and output failure probability and uncertainty; S3. Use the Bayesian fusion method to dynamically adjust the weights of the battery internal resistance physical model and the data-driven model, calculate the posterior probability of battery failure, and quantify the total uncertainty; S4. Set a decision threshold. When the posterior probability exceeds the threshold, trigger an early warning and take protective measures.

2. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 1 is characterized in that: In step S1, the calculation formula for real-time prediction of internal resistance is: R b =R b0 +K·(T s -T s0 )+ΔR b,alt Among them, R b is the predicted value of the battery internal resistance under the current working conditions, R b0 is the reference surface temperature T s0 and the battery internal resistance reference value under standard pressure; K is the internal resistance temperature coefficient; T s is the actual surface temperature of the battery; T s0 is the reference surface temperature; ΔR b,alt is the plateau environment correction term, and its calculation formula is: Among them, β alt is the plateau environmental sensitivity coefficient; P a,sea is the standard pressure at sea level; P a,alt is the actual air pressure on the plateau; Output optimized physical model parameters: Minimize the objective function J through the particle swarm optimization algorithm: Among them, R b,measured (t i ) is at time t i Real-time collected battery internal resistance measurement value; R b,model (t i ) is at time t i The predicted value of the battery internal resistance calculated based on the physical model; N is the number of collected data points, which is used to evaluate the deviation between the model prediction value and the actual measurement value.

3. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 2 is characterized in that: In step S1, the method for calculating the prior probability of failure is: Convert the physical model output into a probability distribution: Battery internal resistance R b :Assume normal distribution where μ phy =R b , that is, the mean of the distribution is the predicted value of internal resistance calculated by the physical model; the variance Determined by model error analysis, taking into account the additional uncertainty brought by the plateau environment: in: is the basic model error variance; γ alt is the uncertainty coefficient of plateau environment; H is the actual altitude; H0 is the reference altitude; The physical model output is used as a prior distribution, considering the impact of the plateau environment on the failure threshold: Where P(fail) is the prior probability of battery failure, ranging from [0,1]; Φ(·) is the cumulative distribution function of the standard normal distribution; R b,crit,alt is the battery internal resistance failure threshold in plateau environments. When the battery internal resistance exceeds this value, it is considered that the battery may be in a failed state. The calculation formula is: Among them, R b,crit is the failure threshold of the battery internal resistance under standard conditions; δ alt is the failure threshold adjustment coefficient for plateau environment.

4. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 3 is characterized in that: In step S2, the failure probability output by the SVM model is: Among them, p svm The battery failure probability prediction value based on the SVM model has a value range of [0,1]. f(x) is the decision function output of the SVM model, and the calculation formula is: Among them, α i is the Lagrange multiplier; y i is the category label of the training sample, with a value of +1 or -1; K(x i ,x) is the kernel function; x i is the eigenvector of the i-th support vector; x is the eigenvector of the current input; and b is the bias term.

5. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 4 is characterized in that: In step S2, the failure probability output by the RF model is: Among them, P rf is the battery failure probability prediction value based on the RF model, with a value range of [0, 1]; T is the number of decision trees in the random forest, which is a positive integer; I(·) is the indicator function, which takes a value of 1 when the prediction result of decision tree t is failure, and 0 otherwise; f t (x) is the prediction result of the t-th decision tree, which determines whether the battery status corresponding to the current input feature vector x is faulty.

6. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 5 is characterized in that: In step S2, the uncertainty of the probability output is quantified, specifically: Calculate the SVM variance estimate based on the confidence level of the support vector distance, taking into account the influence of the plateau environment: in, is the variance of the SVM output probability after considering the plateau environment correction; N sv is the number of support vectors; Calculate the RF variance estimate based on the variance of the Bootstrap sample, taking into account the influence of the plateau environment: in, is the variance of RF output probability after considering the plateau environment correction; It is the predicted value of the failure probability output by the t-th decision tree, and its value range is [0,1].

7. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 6 is characterized in that: In step S3, the calculation method of the posterior probability of battery failure is: First, the internal resistance prediction value output by the physical model and its uncertainty are converted into a priori probability P(fail), which is: Where Φ(·) is the cumulative function of the standard normal distribution; Then, the failure probability and its uncertainty output by the data-driven model are constructed as a likelihood function P(D|fail), taking into account the plateau environment correction, and the formula is: Where P(D|fail) is the likelihood probability of observing the current data D in the case of battery failure, and its value range is [0,1]; It is a normal distribution, indicating that the probability value output by the data-driven model follows a mean of p data,alt , the variance is Normal distribution; p data,alt is the predicted value of the failure probability of the fused data-driven model, and the calculation formula is: in, is the weight of the SVM model after considering the plateau environment correction; is the weight of the RF model after considering the plateau environment correction; Then, based on Bayes’ theorem, the prior probability P(fail) and the likelihood function P(D|fail) are fused to calculate the posterior probability P(fail|D), which is: Substituting the physical model prior P(fail) and data-driven likelihood P(D|fail) corrected for the plateau environment, we get: Where: P(fail|D) refers to the posterior probability of battery failure when data D is observed, and its value range is [0,1]. P(D) is the marginal probability of data D, which is used for normalization to ensure that the sum of posterior probabilities is 1. ensemble mean ensemble variance 8. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 7 is characterized in that: In step S3, the calculation method for quantifying the total uncertainty is: The weight coefficient is adaptively adjusted according to the environmental complexity Cenv, taking into account plateau environmental factors: Among them, ω phy,alt (C env ) is the dynamic weight of the physical model after considering environmental complexity and plateau environment correction; ω data,alt (C env ) is the dynamic weight of the data-driven model after considering environmental complexity and plateau environment correction; is the initial weight value; λ is the weight adjustment coefficient; C env is the environmental complexity; alt is the plateau environment weight adjustment coefficient; The calculation formula of the total uncertainty after fusion is: Among them, U fuse,alt is the total uncertainty after fusion; is the uncertainty of the physical model; is the uncertainty of the data-driven model; σ phy,alt , σ data,alt are the standard deviations of the physical model and the data-driven model after correction for the plateau environment; are the critical values ​​of the t distribution corresponding to the physical model and the data-driven model, respectively, which are used to construct the confidence interval.

9. The plateau energy storage battery monitoring method based on physical data dual drive according to claim 8 is characterized in that: In step S4, a reasonable decision threshold α (C env ,H), the calculated posterior probability P(fail|D) and the decision threshold α(C env ,H) for comparison. If P(fail|D) exceeds the decision threshold, it is judged that the energy storage battery may have a failure risk, triggering a warning signal, and taking corresponding protection measures to ensure the safe operation of the energy storage battery.

10. A plateau energy storage battery monitoring system based on physical data dual drive, characterized in that: The method for implementing any one of claims 1 to 9 comprises: Input layer: real-time collection of environmental parameters, electrical signals, and thermal parameters; Physical model layer: Calculate the battery internal resistance R after plateau environment correction b , generate the prior probability P(fail); Data-driven layer: extracts features corrected for plateau environments, including voltage ripple rate γ ripple , charge and discharge current fluctuation rate γ current , temperature gradient Output failure probability p data,alt ; Bayesian fusion layer: Dynamically weighted fusion of priors and likelihoods to calculate the posterior probability P(fail|D); Decision layer: If P(fail|D)>α(C env ,H), triggering early warning or protection measures, where α(C env ,H) is the decision threshold considering the plateau environment.

Images