Energy storage battery full-time-domain thermal state evaluation and risk positioning method based on machine learning

Abstract

The invention discloses a machine learning-based energy storage battery full-time-domain thermal state evaluation and risk positioning method, and the method comprises the steps: carrying out the thermal state evaluation processing of an energy storage battery sample in a standing day through the temperature data of an energy storage battery sample module; and constructing and training a fusion evaluation model, carrying out risk positioning processing on an energy storage battery sample in a running day, and constructing a module temperature matrix. Inputting the data of the energy storage battery to be subjected to thermal state evaluation into a construction fusion evaluation model to obtain a thermal state evaluation result of the energy storage battery, constructing a module temperature matrix for the data of the energy storage battery to be subjected to risk positioning, and performing thermodynamic diagram rendering on the module temperature matrix to obtain a risk positioning result, and displaying a result and carrying out operation and maintenance linkage according to the result so as to predict the thermal state of the energy storage battery in the standing day and accurately position the risk point of the energy storage battery in the running day.

Description

Technical Field

[0001] This invention belongs to the field of full-time-domain management technology for energy storage batteries, specifically involving a machine learning-based method for full-time-domain thermal state assessment and risk location of energy storage batteries. Background Technology

[0002] Throughout the entire lifecycle of an energy storage battery system, the uniformity of temperature distribution directly affects the battery's performance stability and safety reliability. Abnormal temperatures (such as localized overheating or excessive temperature differences) are key causes of battery capacity decay, increased internal resistance, and even thermal runaway. Therefore, accurate assessment and risk identification of the thermal state of energy storage batteries are crucial.

[0003] Current technologies for monitoring and assessing the thermal state of energy storage batteries face the following core challenges: 1. Incomplete time domain coverage: Existing methods mostly target a single operating state (such as only the operating stage or the resting stage), lacking comprehensive coverage of the thermal state of energy storage batteries across the entire time domain, making it difficult to fully reflect the thermal safety status of the batteries; 2. Single feature dimension: Traditional assessment methods rely heavily on absolute temperature values ​​to judge anomalies, without fully considering key features such as temperature fluctuations and relative temperature differences, resulting in insufficient accuracy in anomaly judgment and a tendency for false alarms or missed alarms. 3. Insufficient model adaptability: A single machine learning algorithm is difficult to balance the generalization and accuracy of anomaly detection. It does not design differentiated weights for core features, resulting in limited reliability of evaluation results.

[0004] Therefore, there is an urgent need for a method for assessing the thermal state and locating risks of energy storage batteries that can cover the entire time domain, fuse multiple features, efficiently aggregate data, and accurately pinpoint risk points. Summary of the Invention

[0005] The purpose of this invention is to provide a machine learning-based method for full-time-domain thermal state assessment and risk location of energy storage batteries, which solves the technical problems of incomplete time-domain coverage, single feature dimension, and insufficient model adaptability in the existing technology for monitoring and assessing the thermal state of energy storage batteries.

[0006] The technical solution of this invention to solve its technical problem is as follows: A machine learning-based method for full-time thermal state assessment and risk localization of energy storage batteries includes the following steps: S1: Collect temperature monitoring data of energy storage battery samples during rest days and operation days, and obtain the module temperature data after screening and aggregation of the temperature monitoring data. S2: Perform thermal state assessment on stationary energy storage battery samples, construct and train a fusion assessment model to obtain an optimized fusion assessment model; and perform risk location processing on operational energy storage battery samples, construct a module temperature matrix, and render a heat map of the module temperature matrix. S3: Input the energy storage battery data to be thermally assessed to build a fusion assessment model, obtain the abnormal score of the energy storage battery and the thermal state assessment result, and construct the module temperature matrix with the energy storage battery data to be risk located, and render the module temperature matrix as a heat map to obtain the risk location result. S4: Display the thermal status assessment results and risk location results, and coordinate operation and maintenance based on the results.

[0007] Preferably, step S1 further includes extracting the date information of the temperature monitoring data. Step S1 specifically involves: collecting temperature monitoring data from k temperature channels corresponding to m clusters of energy storage battery samples on resting days and operating days; filtering the temperature monitoring data and extracting the date information of the temperature monitoring data through a data quality control algorithm; dividing the k temperature channels into n energy storage battery sample modules through aggregation processing, with each energy storage battery sample module containing P temperature channels; and calculating the average temperature of the P temperature channels in each energy storage battery sample module as the temperature data of that module.

[0008] By aggregating the k temperature channels, the energy storage battery sample modules are divided into n energy storage modules, forming a two-dimensional data structure of "cluster-module", which not only retains key temperature information but also improves data processing efficiency.

[0009] Preferably, step S2 involves performing thermal state assessment on the static energy storage battery samples, constructing and training a fusion assessment model to obtain an optimized fusion assessment model; this includes the following steps: D1: Calculate the daily average temperature, temperature standard deviation, and daily average temperature difference data for each energy storage battery sample module, establish a feature matrix, and perform standardization transformation on the feature matrix; normalize the daily average temperature difference, set an anomaly threshold based on the mean and standard deviation of the daily average temperature difference, and use the anomaly threshold to screen each energy storage battery sample module and generate anomaly or normal labels. D2: Construct a fusion evaluation model that includes an isolated forest and a support vector machine. Train the isolated forest model using the feature matrix and train the support vector machine model based on the label generation to obtain the optimized fusion evaluation model.

[0010] Preferably, the abnormal threshold is μ+1.5σ, where μ is the mean of the daily average temperature difference and σ is the standard deviation of the daily average temperature difference.

[0011] Preferably, the tag generation status is either: balanced tag distribution or unbalanced tag distribution; the criterion for judging the tag generation status is: whether the proportion of abnormal tags in the total samples of each energy storage battery sample module is less than 3% or greater than 97%; if yes, the tag distribution is unbalanced; if no, the tag distribution is balanced.

[0012] Preferably, the abnormal score of the energy storage battery in step S3 is: When the label distribution is balanced: ; When the label distribution is unbalanced: ; in, For abnormal scores, For outlier scores in support vector machine models, The normalized score for the daily average temperature difference. This represents the outlier score for the isolated forest model.

[0013] Preferably, the grading rules corresponding to the abnormal scores are as follows: 0~0.5 is normal, 0.5~0.7 is slightly inconsistent, 0.7~0.85 is moderately inconsistent, and 0.85~1.0 is severely inconsistent.

[0014] Preferably, the step of performing risk location processing on the operational energy storage battery samples, constructing a module temperature matrix, and rendering a heat map of the module temperature matrix includes the following steps: Q1: Construct a module temperature matrix for m clusters × n modules, and verify the integrity of the module temperature matrix; Q2: Generate a heat map corresponding to the module temperature matrix using drawing software, set a safe temperature threshold, mark the heat map according to the safe temperature threshold, and generate a module temperature heat map containing temperature labels.

[0015] Preferably, the specific parameters for the heat map generated by the drawing software are as follows: the color gradation uses blue to represent low temperature and red to represent high temperature, the color depth is positively correlated with the temperature value, the font is SimHei, the chart size is 10×4 inches, the resolution is 300dpi, the temperature value is marked in the heat map cells, an image file containing the date is generated, and the temperature labels are arranged horizontally.

[0016] Preferably, the thermal state assessment results include: a graded assessment CSV report, a consistency score heatmap, and a score distribution Ridge plot; the risk location results include: a module temperature heatmap.

[0017] The beneficial effects of this invention are as follows: By utilizing temperature data from energy storage battery sample modules, thermal state assessment is performed on energy storage battery samples during periods of inactivity, a fusion assessment model is constructed and trained to obtain an optimized fusion assessment model; and risk location processing is performed on energy storage battery samples during periods of operation, a module temperature matrix is ​​constructed, and a heatmap of the module temperature matrix is ​​rendered. Different processing schemes are applied to different energy storage battery states to ensure full-time domain coverage of energy storage batteries, achieving full-time domain monitoring on both inactive and operational days. Furthermore, the energy storage battery data to be assessed for thermal state is input into the fusion assessment model to obtain anomaly scores and thermal state assessment results. A module temperature matrix is ​​constructed from the energy storage battery data to be located for risk, and a heatmap of the module temperature matrix is ​​rendered to obtain risk location results. The results are displayed, and operational coordination is performed based on the results, achieving accurate prediction of the thermal state of energy storage batteries during periods of inactivity and precise location of risk points for energy storage batteries during periods of operation. This enables closed-loop management from "thermal status assessment" to "operation and maintenance execution," transforming traditional "post-fault repair" into "predictive maintenance," significantly reducing operation and maintenance costs and safety risks. Attached Figure Description

[0018] Figure 1 This is a flowchart of the machine learning-based full-time-domain thermal state assessment and risk location method for energy storage batteries according to the present invention; Figure 2 To perform thermal state assessment on static energy storage battery samples, a fusion assessment model was constructed and trained, resulting in the flowchart corresponding to the optimized fusion assessment model. Figure 3 To perform risk identification processing on the operational energy storage battery samples, a module temperature matrix was constructed, and a corresponding flowchart was rendered from the module temperature matrix using a heat map. Figure 4 This is a heatmap showing the consistency score in the thermal state assessment results of Example 1; Figure 5 This is the graded assessment CSV report from the thermal state assessment results of Example 1; Figure 6 This is a ridge plot showing the score distribution from the thermal state assessment results of Example 1. Figure 7 This is a thermal map of the operating module temperature from the thermal risk location results of Example 1. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] like Figure 1 As shown, this invention discloses a method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning, including the following steps: S1: Collect temperature monitoring data of energy storage battery samples during resting and operating days. After filtering and aggregating the temperature monitoring data, obtain the module's temperature data; extract the date information from the temperature monitoring data. Specifically: Collect temperature monitoring data from k temperature channels corresponding to m clusters of energy storage battery samples during resting and operating days. Filter the temperature monitoring data and extract the date information using a data quality control algorithm. Through aggregation, divide the k temperature channels into n energy storage battery sample modules. Each energy storage battery sample module contains P temperature channels. Calculate the average temperature of the P temperature channels in each energy storage battery sample module as the module's temperature data. Aggregation processing divides the k temperature channels into n energy storage battery sample modules, forming a two-dimensional "cluster-module" data structure, which retains key temperature information while improving data processing efficiency.

[0021] S2: Perform thermal state assessment on stationary energy storage battery samples, construct and train a fusion assessment model to obtain an optimized fusion assessment model; and perform risk location processing on operating energy storage battery samples, construct a module temperature matrix, and render a heat map of the module temperature matrix. like Figure 2 As shown, thermal state assessment is performed on static energy storage battery samples, and a fusion assessment model is constructed and trained to obtain an optimized fusion assessment model; the process includes the following steps: D1: Calculate the daily average temperature, temperature standard deviation, and daily average temperature difference data for each energy storage battery sample module, establish a feature matrix, and perform standardization transformation on the feature matrix; normalize the daily average temperature difference, set an anomaly threshold based on the mean and standard deviation of the daily average temperature difference, and use the anomaly threshold to screen each energy storage battery sample module and generate anomaly or normal labels. The standardization transformation specifically involves converting the daily average temperature, temperature standard deviation, and daily average temperature difference data into standardized data with a mean of 0 and a variance of 1, eliminating the influence of dimensions and placing the three features on the same weight level. The anomaly threshold is μ + 1.5σ, where μ is the mean of the daily average temperature difference and σ is the standard deviation of the daily average temperature difference. Tag generation is assessed as either balanced or unbalanced. The criteria for judging tag generation are: whether the proportion of abnormal tags in the total samples of each energy storage battery module is less than 3% or greater than 97%; if yes, the tag distribution is unbalanced; otherwise, the tag distribution is balanced. Daily average temperature ( mean ): Calculates the arithmetic mean of the temperature data of each energy storage battery sample module at all times on a single day. This reflects the overall temperature level of each energy storage battery sample module and is the most direct indicator for determining whether the temperature exceeds the standard; temperature standard deviation ( std): Calculates the standard deviation of the temperature data of each energy storage battery sample module on a single day, reflecting the degree of temperature fluctuation. The larger the standard deviation, the worse the temperature stability and the higher the potential risk of thermal anomalies. Daily average temperature difference (delta_avg): Calculates the difference between the daily average temperature of each energy storage battery sample module and the minimum daily average temperature of all m modules on that day. The formula is: delta_avg = daily average temperature of each energy storage battery sample module - minimum daily average temperature of all energy storage battery sample modules on that day. This feature can highlight the relative temperature difference between each energy storage battery sample module and is the core indicator for judging the uniformity of temperature distribution. Excessive relative temperature difference is often an early sign of thermal runaway.

[0022] D2: Construct a fusion evaluation model that includes Isolation Forest and Support Vector Machine (SVM). Train the Isolation Forest model using the feature matrix and train the Support Vector Machine model based on the label generation to obtain the optimized fusion evaluation model.

[0023] The Isolation Forest model is trained using all energy storage battery samples to construct multiple decision trees to separate anomalous samples, making it suitable for unsupervised or semi-supervised scenarios. In practical applications, an Isolation Forest model containing 500 decision trees can be constructed and trained using a standardized 3D feature matrix as input. The Isolation Forest model outputs an anomaly score (negative average path length) for each sample; a higher score indicates a greater anomaly probability. Subsequently, Min-Max normalization is used to map the anomaly score to the 0-1 range, facilitating fusion with other scores.

[0024] The Support Vector Machine (SVM) model is trained. If the generated labels include both abnormal labels (1) and normal labels (0), the SVM model is trained based on all the classified energy storage battery samples. The class weights are balanced to address the data imbalance problem where the number of normal samples may be much greater than that of abnormal samples. The model outputs the probability (0~1) that a sample is abnormal. If the generated labels contain only one class (all normal labels (0) or all abnormal labels (1)), it indicates that there is no obvious abnormal difference in the current data. The SVM training is skipped, and only the abnormal score of the Isolation Forest model is used.

[0025] like Figure 3 As shown, risk localization processing is performed on the operational energy storage battery sample, a module temperature matrix is ​​constructed, and a heat map rendering of the module temperature matrix is ​​performed, including the following steps: Q1: Construct a module temperature matrix for m clusters × n modules, and verify the integrity of the module temperature matrix; Q2: Generate a heatmap corresponding to the module temperature matrix using graphing software, set a safe temperature threshold, and mark the heatmap according to the safe temperature threshold to generate a module temperature heatmap with temperature labels. Specific parameters for the graphing software to draw the heatmap are: color gradation with blue representing low temperature and red representing high temperature, color depth positively correlated with temperature value, SimHei font, chart size 10×4 inches, resolution 300dpi, temperature values ​​labeled in heatmap cells, generation of image files including dates, and horizontal arrangement of temperature labels. Higher temperatures result in darker colors, quickly and intuitively presenting temperature distribution differences; simultaneously, labeling each cell with a specific temperature value ensures the accuracy of quantitative analysis. By comparing the color depth of the module temperature heatmap, the darkest cell is automatically identified, and its corresponding "cluster-module" unit is the hottest module; simultaneously, the temperature value of this cell is extracted and compared with the preset safe temperature threshold to determine if there is an overheating risk, outputting the location information (cluster number, module number) and temperature value of the hottest module, providing maintenance personnel with accurate maintenance navigation.

[0026] S3: Input the energy storage battery data to be thermally assessed to build a fusion assessment model, obtain the abnormal score of the energy storage battery and the thermal state assessment result, and construct the module temperature matrix with the energy storage battery data to be risk located, and render the module temperature matrix as a heat map to obtain the risk location result. The anomaly score for the energy storage battery is: When the label distribution is balanced: ; When the label distribution is unbalanced: ; in, For abnormal scores, For outlier scores in support vector machine models, The normalized score for the daily average temperature difference. The outlier scores for the isolated forest model are used. By fusing the normalized score of the daily average temperature difference with the outlier scores of the isolated forest model and the support vector machine (SVM) model, the role of core features is highlighted while taking into account the model's generalization ability, thus improving the accuracy of the evaluation.

[0027] The grading rules for abnormal scores are as follows: 0~0.5 is normal, 0.5~0.7 is slightly inconsistent, 0.7~0.85 is moderately inconsistent, and 0.85~1.0 is severely inconsistent.

[0028] Specifically: 0≤Score<0.5: Under normal conditions, the temperature level of the energy storage battery module is reasonable with small fluctuations, and the temperature difference with other energy storage battery modules is within the allowable range, with no risk of thermal anomalies; 0.5≤Score<0.7: Slight inconsistency, the temperature of the energy storage battery module is slightly abnormal or fluctuates slightly, and there is a slight temperature difference with other energy storage battery modules. The subsequent trend needs to be monitored, but no immediate repair is required. 0.7≤Score<0.85: Moderate inconsistency, the temperature of the energy storage battery module is abnormally high and fluctuates greatly, and the temperature difference with other energy storage battery modules is significant. There may be poor heat dissipation or minor defects in the internal structure, and targeted inspection is required. 0.85≤Score≤1.0: Severe inconsistency, the temperature of the energy storage battery module is seriously abnormal or fluctuates drastically, and there is a huge temperature difference with other energy storage battery modules, with a high risk of thermal runaway.

[0029] S4: Display the thermal status assessment results and risk location results, and coordinate operation and maintenance based on the results. Transform complex assessment and location results into intuitive and actionable information, and link them with operation and maintenance strategies to form a closed-loop management system.

[0030] The thermal status assessment results include: a graded assessment CSV report, a consistency score heatmap, and a score distribution ridge plot; the risk location results include: a module temperature heatmap. Specifically: Tiered assessment CSV report: sorted in descending order by date and anomaly score, containing fields such as "date, cluster, module, anomaly score, inconsistency level", supporting data statistical analysis and traceability; Consistency Score Heatmap: Outputs anomaly score heatmaps for m×n energy storage battery modules by date, using discrete color levels of green (normal), yellow (slight inconsistency), orange (moderate inconsistency), and red (severe inconsistency), with added manual legends to quickly identify the spatial distribution of abnormal modules; Scoring distribution ridge plot: Draws anomaly score distribution ridge plots for m clusters, adding threshold lines of 0.5, 0.7, and 0.85 levels to visually display the score distribution characteristics of each cluster and the proportion of anomalous modules, providing support for overall risk assessment.

[0031] Module Temperature Heatmap: A module temperature heatmap is generated for each m×n energy storage battery module. A discrete color scale of green (normal), yellow (slight inconsistency), orange (moderate inconsistency), and red (severe inconsistency) is used, with manual legends added to quickly identify the spatial distribution of abnormal modules. The heatmap color contrast automatically identifies the cell with the darkest color, and its corresponding "cluster-module" unit is the hottest module. At the same time, the temperature value of the cell is extracted and compared with the preset safe temperature threshold to determine whether there is an overheating risk. The location information (cluster number, module number) and temperature value of the hottest module are output. Example

[0032] Taking an energy storage battery with 6 clusters, 64 temperature channels, and 16 modules, with each module containing 4 temperature channels, as an example, the specific process for full-time thermal state assessment and risk identification of the energy storage battery is as follows: For energy storage battery samples that have been left undisturbed for several days, temperature monitoring data from 64 temperature channels was collected. After filtering and aggregation of the temperature monitoring data, temperature data for 16 modules were obtained. The daily average temperature, standard deviation, and daily average temperature difference of the 16 modules were calculated to establish a feature matrix, which was then standardized. The daily average temperature difference was normalized, and an anomaly threshold was set based on the mean and standard deviation of the daily average temperature difference. This threshold was used to filter each energy storage battery sample module and generate anomaly or normal labels. A fusion evaluation model incorporating Isolation Forest and Support Vector Machine (SVM) was constructed. The Isolation Forest model was trained using the feature matrix, and the SVM model was trained based on the label generation results, resulting in an optimized fusion evaluation model. The energy storage battery data to be evaluated for thermal status was then input into the fusion evaluation model to obtain the anomaly score and thermal status evaluation results for the energy storage battery. Figure 4 , Figure 5 , Figure 6 As shown, the thermal state assessment results are clearly displayed through the graded evaluation CSV report, consistency score heatmap, and score distribution Ridge plot.

[0033] For the operational energy storage battery sample, temperature monitoring data from 64 temperature channels was collected. After filtering and aggregation of the temperature monitoring data, temperature data for 16 modules were obtained. A two-dimensional temperature matrix of 6 clusters × 16 modules was constructed, and the integrity of the matrix dimensions (rows = 6, columns = 16) was simultaneously verified to avoid matrix anomalies caused by missing data. Figure 7 As shown, a heat map corresponding to the module temperature matrix is ​​generated by drawing software. A safe temperature threshold is set, and the heat map is marked according to the safe temperature threshold to generate a module temperature heat map containing temperature labels. Different colors represent different temperatures, so as to clearly display the risk positioning results of the energy storage battery.

[0034] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning, characterized in that, Includes the following steps: S1: Collect temperature monitoring data of energy storage battery samples during rest days and operation days, and obtain the module temperature data after screening and aggregation of the temperature monitoring data. S2: Perform thermal state assessment on stationary energy storage battery samples, construct and train a fusion assessment model to obtain an optimized fusion assessment model; and perform risk location processing on operational energy storage battery samples, construct a module temperature matrix, and render a heat map of the module temperature matrix. S3: Input the energy storage battery data to be thermally assessed to build a fusion assessment model, obtain the abnormal score of the energy storage battery and the thermal state assessment result, and construct the module temperature matrix with the energy storage battery data to be risk located, and render the module temperature matrix as a heat map to obtain the risk location result. S4: Display the thermal status assessment results and risk location results, and coordinate operation and maintenance based on the results.

2. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 1, characterized in that: Step S1 further includes extracting the date information of the temperature monitoring data. Specifically, step S1 involves: collecting temperature monitoring data from k temperature channels corresponding to m clusters of energy storage battery samples on resting days and operating days; filtering the temperature monitoring data and extracting the date information of the temperature monitoring data through a data quality control algorithm; dividing the k temperature channels into n energy storage battery sample modules through aggregation processing, with each energy storage battery sample module containing P temperature channels; and calculating the average temperature of the P temperature channels in each energy storage battery sample module as the temperature data of that module.

3. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 2, characterized in that, Step S2 involves performing thermal state assessment on the static energy storage battery samples, constructing and training a fusion assessment model, and obtaining an optimized fusion assessment model; this includes the following steps: D1: Calculate the daily average temperature, temperature standard deviation, and daily average temperature difference data for each energy storage battery sample module, establish a feature matrix, and perform standardization transformation on the feature matrix; normalize the daily average temperature difference, set an anomaly threshold based on the mean and standard deviation of the daily average temperature difference, and use the anomaly threshold to screen each energy storage battery sample module and generate anomaly or normal labels. D2: Construct a fusion evaluation model that includes an isolated forest and a support vector machine. Train the isolated forest model using the feature matrix and train the support vector machine model based on the label generation to obtain the optimized fusion evaluation model.

4. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 3, characterized in that: The abnormal threshold is μ+1.5σ, where μ is the mean of the daily average temperature difference and σ is the standard deviation of the daily average temperature difference.

5. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 4, characterized in that: The label generation status is defined as either balanced or unbalanced. The criteria for judging the label generation status are as follows: whether the proportion of abnormal labels in the total samples of each energy storage battery sample module is less than 3% or greater than 97%. If yes, the label distribution is unbalanced; otherwise, the label distribution is balanced.

6. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 5, characterized in that, The abnormal score of the energy storage battery in step S3 is: When the label distribution is balanced: ; When the label distribution is unbalanced: ; in, For abnormal scores, For outlier scores in support vector machine models, The normalized score for the daily average temperature difference. This represents the outlier score for the isolated forest model.

7. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 6, characterized in that: The grading rules corresponding to the abnormal scores are as follows: 0~0.5 is normal, 0.5~0.7 is slightly inconsistent, 0.7~0.85 is moderately inconsistent, and 0.85~1.0 is severely inconsistent.

8. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 2, characterized in that, The process of risk localization processing for operational energy storage battery samples, constructing a module temperature matrix, and rendering a heatmap of the module temperature matrix includes the following steps: Q1: Construct a module temperature matrix for m clusters × n modules, and verify the integrity of the module temperature matrix; Q2: Generate a heat map corresponding to the module temperature matrix using drawing software, set a safe temperature threshold, mark the heat map according to the safe temperature threshold, and generate a module temperature heat map containing temperature labels.

9. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 8, characterized in that, The specific parameters for drawing heat maps using the drawing software are as follows: the color gradation uses blue to represent low temperature and red to represent high temperature, the color depth is positively correlated with the temperature value, the font is SimHei, the chart size is 10×4 inches, the resolution is 300dpi, the temperature value is labeled in the heat map cells, an image file containing the date is generated, and the temperature labels are arranged horizontally.

10. The method for full-time-domain thermal state assessment and risk location of energy storage batteries based on machine learning according to claim 1, characterized in that, The thermal state assessment results include: a graded assessment CSV report, a consistency score heatmap, and a score distribution Ridge plot; the risk location results include: a module temperature heatmap.

Images