Battery application management and control method

Abstract

The invention discloses a battery application management and control method, and the method comprises the steps: dynamically constructing a multi-attribute-feature time sequence prediction model cluster based on a plurality of battery attribute features, and carrying out the online selection of the most valuable attribute features for safety risk prediction, so as to meet the safety risk prediction demands of batteries of different manufacturers and different models under different working conditions, and improve the safety performance of the batteries. And security risk sample data are expanded by using a label propagation algorithm, and non-label data are deeply mined and fully utilized, so that security risk prediction can be performed on the lithium battery more accurately and stably.

Description

Technical Field

[0001] The present application belongs to the technical field of battery management systems, and specifically relates to a battery application management and control method. Background Art

[0002] At present, traditional base station batteries are only used for emergency backup power during power outages, and have not become a production tool for base station asset management, resulting in a large waste of silent energy storage equipment assets for operators. At present, the power sector divides the 24 hours of a day into several time periods, such as peak, peak, normal, and valley, based on the load changes of the power grid, and sets different electricity prices for each time period to encourage users to reasonably arrange electricity use time, stagger electricity use, improve equipment utilization efficiency and save energy.

[0003] Although intelligent peak-shifting technology can improve battery utilization and save electricity costs, the primary function of base station batteries is backup power. Therefore, the battery still needs to meet the backup power requirements after peak shifting, and the peak-shifting power strategy needs to be accurately calculated to ensure the reliability of backup power and maximize benefits. The battery backup time is affected by multiple factors such as load power, battery SOC, and battery age. There are huge differences in load power, battery capacity, and health status at different sites. Therefore, it is crucial to judge battery risks early and predict them in advance, and guide staff to deal with corresponding risks in a timely manner to ensure the safe operation of existing base stations that enable intelligent peak-shifting technology.

[0004] At present, batteries are facing the problem that mechanism research lags behind engineering implementation. Mechanism research aims to deeply understand the working principles and failure mechanisms inside lithium batteries, but in practice it is often difficult to keep up with the rapid development of engineering technology. This means that we cannot simply rely on theoretical models to predict and explain all types of lithium battery safety risks. Therefore, a data-driven approach is needed to gradually capture effective features and discard the interference of invalid features.

[0005] There are two key challenges in solving the safety risks of lithium batteries based on a data-driven approach.

[0006] First: It is difficult to characterize the battery safety risk characteristics.

[0007] Traditional battery safety risk prediction schemes rely only on cross-sectional data at a single point in time, and lack observation and analysis of leading indicator characteristics of lithium battery safety risks. Before the battery presents obvious safety risks, the leading indicator characteristics of lithium battery safety risks can predict some specific indicators or patterns of its potential problems. These characteristics require observing samples over a period of time to extract the implicit patterns. For example, by monitoring the battery's charge and discharge cycles, temperature changes, voltage fluctuations and other timing characteristics, by collecting and analyzing these time-varying data, some patterns or trends related to battery safety risks can be identified. This method of deeply digging into the hidden information behind the data is crucial to improving the safety and reliability of lithium batteries.

[0008] In addition, the leading indicators of safety risks of lithium batteries from different manufacturers, models and batches are not exactly the same. This is because different manufacturers have differences in battery design, material selection, manufacturing process, etc., which leads to differences in performance and reliability of their products, which also greatly increases the difficulty of determining effective characteristics to characterize battery safety risks.

[0009] Second: There is little labeled data, and massive amounts of unlabeled data / samples cannot be effectively utilized In the current lithium battery safety risk analysis, there is relatively little annotated data available, especially the limited number of samples with safety risks. This poses a challenge to building an accurate and reliable prediction model. In actual engineering deployment, due to resource limitations or technical difficulties, it is difficult to accurately annotate the safety risks of each sample. This leads to the scarcity of samples with safety risks, further exacerbating the small sample problem.

[0010] In addition, there are a large number of plausible samples in many engineering deployments, that is, samples that exhibit certain risk characteristics but do not fully conform to the typical security risk model. These samples may be abnormal situations caused by environmental factors, usage conditions or other unknown factors. However, due to the lack of clear labeling standards or expert judgment, these samples are often not effectively used. They cannot be regarded as normal samples and are excluded from analysis.

[0011] In summary, the difficulty in effectively characterizing battery safety risk characteristics and the low utilization rate of unlabeled samples are important challenges in lithium battery safety risk analysis technology and need to be urgently addressed. Summary of the invention

[0012] The purpose of this application is to solve the problems of the prior art and to provide a battery application management and control method to solve the problems in the battery risk prediction technical solution in intelligent peak shifting, such as the difficulty in determining the characteristics that characterize battery safety risks, the small sample size, and the low utilization rate of massive unlabeled samples.

[0013] In order to solve the technical problem, the technical solution of the present application is: a battery application management and control method, comprising the following steps: Step 1: Obtain battery operation business data as a multi-attribute feature set of the battery; Step 2: Preprocessing of battery operation business data, including but not limited to deleting erroneous values, duplicate values, replacing outliers, and filling missing values; Step 3: Build a time series prediction model cluster with multiple attribute features, and select a set of attribute features that effectively characterize battery safety risks based on the prediction accuracy of each model in the model cluster; Step 3-1: Build a time series prediction model cluster with multiple attribute features; Establish a time series prediction model of multi-attribute features. The model input includes a sampling sequence of multiple attribute features within a specific time of the battery. The model output is a battery safety risk prediction. The time series prediction model of multi-attribute features is trained and evaluated using the preprocessed safety risk samples of the multi-attribute feature set or the updated label data. The preprocessed multi-attribute feature set is traversed and combined, and a time series prediction model is trained for each attribute feature combination. The models constructed by all attribute feature combinations are combined together to form a multi-attribute feature time series prediction model cluster; Step 3-2: Online selection of a feature set of attributes that effectively characterize battery safety risks based on the prediction accuracy of each model in the time series prediction model cluster with multi-attribute features; Step 4: Label data expansion: Based on the effective feature set representing the battery safety risk attribute obtained through online selection, the unlabeled samples in the preprocessed multi-attribute feature set are labeled through the differential measurement algorithm and label propagation algorithm to obtain updated label data; Step 5: Retrain the multi-attribute feature time series prediction model cluster based on the updated label data, and after multiple rounds of iterations, until the proportion of unlabeled samples is lower than the threshold, output the battery safety risk prediction model; Step 6: Battery safety risk prediction: When a new battery operating condition sample arrives, it is input into the trained battery safety risk prediction model to obtain its corresponding safety risk prediction result. For base stations with high-risk batteries, the intelligent peak-shifting strategy will no longer be issued to ensure the safe operation of the base station.

[0014] Preferably, the battery operation service data in step 1 is divided into three types, namely battery operation data, battery status data and battery alarm data. As a multi-attribute feature set of the battery, the battery operation data includes the voltage, temperature and humidity data during the battery charging and discharging process, the battery status data includes the capacity, remaining capacity, maximum available capacity, internal resistance and life, and the battery alarm data includes voltage abnormality and temperature abnormality.

[0015] Preferably, the preprocessing of the battery operation service data in step 2 is specifically as follows: Delete the wrong values ​​and duplicate values; For outliers, based on the box plot, the third quartile is used to replace high outliers in the data, and the first quartile is used to replace low outliers in the data; For missing values, linear interpolation is used to fill in the missing values.

[0016] Preferably, the step 3 is specifically: Step 3-1: Build a time series prediction model cluster with multiple attribute features; The long short-term memory network LSTM is selected as the time series prediction model of multi-attribute features. When establishing the time series prediction model of multi-attribute features, the input and output of the model are first defined. The input includes a sampling sequence of multiple attribute features within a specific time of the battery. The output of the model is the battery safety risk prediction. The samples of the safety risk of the preprocessed multi-attribute feature set or the updated label data are divided into two parts at a ratio of 8:2, namely, a training set and a validation set. The time series prediction model of multi-attribute features is trained. The training set is used to train the model so that it can learn the temporal relationship between attribute features and the leading indicator features of safety risks. The validation set is used to evaluate the performance of the model. The preprocessed multi-attribute feature set is traversed and combined. Each attribute feature combination trains an LSTM time series prediction model. The LSTM time series prediction models constructed by all attribute feature combinations are combined together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. Step 3-2: Online selection of a feature set that effectively characterizes battery safety risk attributes; Attribute features are selected according to the model prediction accuracy of the time series prediction model with multiple attribute features, and the contribution value of each attribute feature is calculated and sorted by size. The top-k attribute features that are strongly correlated with safety risk prediction are selected to obtain a set of attribute features that effectively characterize battery safety risks, and the k attribute features are normalized.

[0017] Preferably, the calculation formula for the contribution value of each attribute feature is: in: Indicates the prediction result; Represents the expected value of the prediction result; v(S) represents the shap value of the attribute feature subset S; Represents the shap value of the new subset after adding attribute feature i.

[0018] Preferably, the calculation formula of the normalized weight w[k] is as follows: W[k]= in: weight[k] represents the weight value of the kth attribute feature; Represents the sum of all k attribute feature weight values.

[0019] Preferably, the step 4 is specifically: Step 4-1: Sample difference measurement; In battery safety risk prediction, DTW is used to compare whether the behavior patterns of two samples on a specific attribute feature are similar: D(a,b)=Σ(w[k]*DTW(a[k],b[k])) in: D(a,b) represents the difference measure between sample a and sample b; w[k] represents the normalized weight of the kth attribute feature; DTW(a[k],b[k]) represents the dynamic time warping distance between the attribute features of sample a and sample b at the kth time that effectively characterizes the battery safety risk; Step 4-2: Label Propagation Initialization: Assign an initial label to each unlabeled sample, usually randomly; Iterative update: According to the difference measure D(a, b) between samples, the weight between each unlabeled sample and the surrounding labeled samples is calculated. The larger the weight, the more similar the two samples are, and the greater the impact of label transfer. Label propagation: Update the labels of unlabeled samples according to the weights. This process is achieved through the weighted average aggregation method. Convergence judgment: Repeat the iterative update steps until the convergence condition is met, the number of iterations reaches a predetermined value or the label change is less than a threshold; Label the unlabeled samples in the preprocessed multi-attribute feature set to obtain updated label data.

[0020] Preferably, step 6 is specifically as follows: when a new battery operating condition sample is collected and transmitted to the system, it is input into the trained battery safety risk prediction model to obtain a safety risk prediction result, and real-time safety risk prediction and analysis are performed on the new battery operating condition sample. For base stations with high-risk batteries, intelligent peak-shifting strategies will no longer be issued to ensure the safe operation of the base stations.

[0021] Preferably, the battery application management method is executed by a battery management and control platform, which includes a basic function module, a battery health management module, a battery backup power analysis module, a battery life prediction module, an intelligent peak-shifting charging and discharging module, and a peak-shifting benefit calculation module; The basic function module can obtain and view the battery operation service data of the base station, including: configuration, battery operation data, battery status data and battery alarm data, and remotely control the battery; The battery health management module can evaluate the health status of each battery group based on the battery operation business data to obtain safety risk prediction results; support battery monitoring management of the entire network / subnet / site, remind through notifications and alarms, and provide relevant reports; The battery backup power analysis module can analyze the backup power capacity of a site or power supply system based on battery operation service data, load current, and related alarms; supports statistical analysis of the entire network / subnet / site, and provides backup power capacity analysis results; The battery life prediction module can predict the remaining life of the battery based on the battery installation date, ambient temperature, battery type, and battery operation condition data; The intelligent off-peak charging and discharging module can remotely set the off-peak power consumption parameters of each battery group, enable and disable the off-peak power consumption function, query, set, modify and delete the off-peak power consumption mechanism, and realize low-cost self-charging and high-cost self-discharging; The off-peak benefit calculation module can view the detailed information of off-peak power consumption of each battery group in real time, count and analyze the off-peak power consumption data and give the results, and evaluate the benefits of off-peak power consumption.

[0022] Compared with the prior art, the advantages of this application are: (1) This application discloses a battery application management and control method. Based on multiple battery attribute characteristics, a time series prediction model cluster of multiple attribute characteristics is dynamically constructed, and the attribute characteristics that are most valuable for safety risk prediction are selected online to adapt to the safety risk prediction needs of batteries of different manufacturers and models under different working conditions. The label propagation algorithm is used to expand the safety risk sample data, deeply mine and make full use of unlabeled data, so as to predict the safety risks of lithium batteries more accurately and stably. (2) This application builds a time series prediction model cluster based on multiple battery attribute features, selects multi-attribute leading indicator features that effectively characterize battery safety risks online, enhances model robustness, and solves the problem of difficulty in determining effective attribute features; (3) This application uses online selection to obtain multi-attribute advance indicator features and label propagation algorithms that effectively characterize battery safety risks, expands risk sample data, iteratively trains the battery safety risk prediction model, improves the model's generalization ability and robustness, and solves the problem of a small number of battery safety risk samples; (4) This application uses a data-driven approach to extract high-weighted original attribute features and eliminate invalid attribute features from battery multi-attribute feature sample data, which can enhance model stability; (5) When constructing a time series prediction model cluster of multi-attribute features, the present application traverses and combines the preprocessed multi-attribute feature set, trains an LSTM time series prediction model for each attribute combination, and combines the LSTM models constructed by all attribute combinations together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. This processing method can better utilize the complementary information between the various attribute features and improve the overall prediction ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 This is a flowchart of a battery application management method for this application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0026] This application proposes a battery application management and control method, which includes battery business data acquisition, battery business data preprocessing, online selection of a feature set that effectively represents battery safety risk attributes, label data expansion, and output of a battery safety risk prediction model, and then battery safety risk prediction is performed. For base stations with high-risk batteries, intelligent peak-shifting strategies will no longer be issued to ensure the safe operation of the base stations.

[0027] Based on multiple battery attribute characteristics, this application dynamically constructs a multi-attribute feature time series prediction model cluster and online selects the attribute characteristics that are most valuable for safety risk prediction to adapt to the safety risk prediction needs of batteries from different manufacturers and models under different working conditions. It also uses a label propagation algorithm to expand safety risk sample data, deeply mine and fully utilize unlabeled data, and can predict the safety risks of lithium batteries more accurately and stably.

[0028] The present application discloses a battery application management method, comprising the following steps: Step 1: Obtain battery operation business data as a multi-attribute feature set of the battery; Step 2: Preprocessing of battery operation business data, including but not limited to deleting erroneous values, duplicate values, replacing outliers, and filling missing values; Step 3: Build a time series prediction model cluster with multiple attribute features, and select a set of attribute features that effectively characterize battery safety risks based on the prediction accuracy of each model in the model cluster; Step 3-1: Build a time series prediction model cluster with multiple attribute features; Establish a time series prediction model of multi-attribute features. The model input includes a sampling sequence of multiple attribute features within a specific time of the battery. The model output is a battery safety risk prediction. The time series prediction model of multi-attribute features is trained and evaluated using the preprocessed safety risk samples of the multi-attribute feature set or the updated label data. The preprocessed multi-attribute feature set is traversed and combined, and a time series prediction model is trained for each attribute feature combination. The models constructed by all attribute feature combinations are combined together to form a multi-attribute feature time series prediction model cluster; Step 3-2: Online selection of a feature set of attributes that effectively characterize battery safety risks based on the prediction accuracy of each model in the time series prediction model cluster with multi-attribute features; Step 4: Label data expansion: Based on the effective feature set representing the battery safety risk attribute obtained through online selection, the unlabeled samples in the preprocessed multi-attribute feature set are labeled through the differential measurement algorithm and label propagation algorithm to obtain updated label data; Step 5: Retrain the multi-attribute feature time series prediction model cluster based on the updated label data, and after multiple rounds of iterations, until the proportion of unlabeled samples is lower than the threshold, output the battery safety risk prediction model; Step 6: Battery safety risk prediction; When the battery management and control platform collects battery operating condition data, it inputs it into the trained battery safety risk prediction model to obtain the corresponding safety risk prediction result. For base stations with high-risk batteries, the intelligent peak-shifting strategy will no longer be issued to ensure the safe operation of the base stations.

[0029] Preferably, the battery operation service data in step 1 is divided into three types, namely battery operation data, battery status data and battery alarm data. As a multi-attribute feature set of the battery, the battery operation data includes the voltage, temperature and humidity data during the battery charging and discharging process, the battery status data includes the capacity, remaining capacity, maximum available capacity, internal resistance and life, and the battery alarm data includes voltage abnormality and temperature abnormality.

[0030] Preferably, the preprocessing of the battery operation service data in step 2 is specifically as follows: Delete the wrong values ​​and duplicate values; For outliers, based on the box plot, the third quartile is used to replace high outliers in the data, and the first quartile is used to replace low outliers in the data; For missing values, linear interpolation is used to fill in the missing values.

[0031] Preferably, the step 3 is specifically: Step 3-1: Build a time series prediction model cluster with multiple attribute features; The long short-term memory network LSTM is selected as the time series prediction model of multi-attribute features. When establishing the time series prediction model of multi-attribute features, the input and output of the model are first defined. The input includes a sampling sequence of multiple attribute features within a specific time of the battery. The output of the model is the battery safety risk prediction. The samples of the safety risk of the preprocessed multi-attribute feature set or the updated label data are divided into two parts at a ratio of 8:2, namely, a training set and a validation set. The time series prediction model of multi-attribute features is trained. The training set is used to train the model so that it can learn the temporal relationship between attribute features and the leading indicator features of safety risks. The validation set is used to evaluate the performance of the model. The preprocessed multi-attribute feature set is traversed and combined. Each attribute feature combination trains an LSTM time series prediction model. The LSTM time series prediction models constructed by all attribute feature combinations are combined together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. Step 3-2: Online selection of a feature set that effectively characterizes battery safety risk attributes; Attribute features are selected according to the model prediction accuracy of the time series prediction model with multiple attribute features, and the contribution value of each attribute feature is calculated and sorted by size. The top-k attribute features that are strongly correlated with safety risk prediction are selected to obtain a set of attribute features that effectively characterize battery safety risks, and the k attribute features are normalized.

[0032] Preferably, the calculation formula for the contribution value of each attribute feature is: in: Indicates the prediction result; Represents the expected value of the prediction result; v(S) represents the shap value of the attribute feature subset S; Represents the shap value of the new subset after adding attribute feature i.

[0033] Preferably, the calculation formula of the normalized weight w[k] is as follows: W[k]= in: weight[k] represents the weight value of the kth attribute feature; Represents the sum of all k attribute feature weight values.

[0034] Preferably, the step 4 is specifically: Step 4-1: Sample difference measurement; In battery safety risk prediction, DTW is used to compare whether the behavior patterns of two samples on a specific attribute feature are similar: D(a,b)=Σ(w[k]*DTW(a[k],b[k])) in: D(a,b) represents the difference measure between sample a and sample b; w[k] represents the normalized weight of the kth attribute feature; DTW(a[k],b[k]) represents the dynamic time warping distance between the attribute features of sample a and sample b at the kth time that effectively characterizes the battery safety risk; Step 4-2: Label Propagation Initialization: Assign an initial label to each unlabeled sample, usually randomly; Iterative update: According to the difference measure D(a, b) between samples, the weight between each unlabeled sample and the surrounding labeled samples is calculated. The larger the weight, the more similar the two samples are, and the greater the impact of label transfer. Label propagation: Update the labels of unlabeled samples according to the weights. This process is achieved through the weighted average aggregation method. Convergence judgment: Repeat the iterative update steps until the convergence condition is met, the number of iterations reaches a predetermined value or the label change is less than a threshold; Label the unlabeled samples in the preprocessed multi-attribute feature set to obtain updated label data.

[0035] Preferably, step 6 is specifically as follows: when a new battery operating condition sample is collected and transmitted to the system, it is input into the trained battery safety risk prediction model to obtain a safety risk prediction result, and real-time safety risk prediction and analysis are performed on the new battery operating condition sample. For base stations with high-risk batteries, intelligent peak-shifting strategies will no longer be issued to ensure the safe operation of the base stations.

[0036] Preferably, the battery application management method is executed by a battery management and control platform, which includes a basic function module, a battery health management module, a battery backup power analysis module, a battery life prediction module, an intelligent peak-shifting charging and discharging module, and a peak-shifting benefit calculation module; The basic function module can obtain and view the battery operation service data of the base station, including: configuration, battery operation data, battery status data and battery alarm data, and remotely control the battery; The battery health management module can evaluate the health status of each battery group based on the battery operation business data to obtain safety risk prediction results; support battery monitoring management of the entire network / subnet / site, remind through notifications and alarms, and provide relevant reports; The battery backup power analysis module can analyze the backup power capacity of a site or power supply system based on battery operation service data, load current, and related alarms; supports statistical analysis of the entire network / subnet / site, and provides backup power capacity analysis results; The battery life prediction module can predict the remaining life of the battery based on the battery installation date, ambient temperature, battery type, and battery operation condition data; The intelligent off-peak charging and discharging module can remotely set the off-peak power consumption parameters of each battery group, enable and disable the off-peak power consumption function, query, set, modify and delete the off-peak power consumption mechanism, and realize low-cost self-charging and high-cost self-discharging; The off-peak benefit calculation module can view the detailed information of off-peak power consumption of each battery group in real time, count and analyze the off-peak power consumption data and give the results, and evaluate the benefits of off-peak power consumption.

[0037] Brief introduction of the battery management and control platform for this application: The battery management and control platform mainly includes the following functional modules: 1) Basic functional modules The system can obtain and view the battery information of the base station, including: configuration, battery operation data, battery status data and battery alarm data, and remotely control the battery.

[0038] 2) Battery health management module The system can evaluate the health status of each battery group based on the battery operation business data (voltage, current, SoH, SoC) to obtain safety risk prediction results; support battery monitoring management for the entire network / subnet / site, remind through notifications and alarms, and provide relevant reports.

[0039] 3) Battery backup analysis module The system can analyze the backup power capacity of a site or power supply system based on battery operation service data (voltage, current, SoH, SoC), load current, and related alarms; it supports statistical analysis of the entire network / subnet / site, and provides backup power capacity analysis results.

[0040] 4) Battery life prediction module The system can predict the remaining life of the battery based on the battery installation date, ambient temperature, battery type, and battery operation business data (e.g. voltage, current, SoH, SoC, number of cycles).

[0041] 5) Intelligent peak-shifting charging and discharging module The system can remotely set the peak-shifting power consumption parameters for each group of batteries, enable and disable the peak-shifting power consumption function, query, set, modify and delete the peak-shifting power consumption mechanism, and realize low-cost self-charging and high-cost self-discharging.

[0042] 6) Peak-shifting revenue calculation module The system can view detailed information on off-peak power consumption of each battery group in real time, count and analyze the off-peak power consumption data and give the results, and evaluate the benefits of off-peak power consumption.

[0043] Example 1 This embodiment discloses a battery application management and control method, which specifically includes the following steps: Step 1: Obtain battery operation service data Battery operation service data is divided into three types: (1) Battery operation data, such as voltage, temperature, and humidity data during battery charging and discharging; (2) Battery status data, such as capacity, remaining capacity, maximum available capacity, internal resistance, life span, etc.; (3) Battery warning data: such as abnormal voltage, abnormal temperature, etc.; Each type of data obtained as above is used as a battery multi-attribute feature set for modeling.

[0044] Step 2: Preprocessing of battery operation service data The quality of data determines the effectiveness of the prediction model and is an important preparatory work before model training. Through data preprocessing, we can avoid some erroneous values, duplicate values, missing values, outliers, etc. from having a negative impact on the construction of the model.

[0045] In this embodiment, different types of data are processed as follows: Wrong values ​​and duplicate values: delete them; Missing values: Linear interpolation was used to fill missing values; Outliers: Based on the box plot, high outliers in the data are replaced by the third quartile and low outliers in the data are replaced by the first quartile.

[0046] Step 3: (3-1) Establish and train a time series prediction model cluster with multiple attribute features The multi-attribute feature time series prediction model can adopt a variety of architectures. This embodiment chooses the long short-term memory network (Long Short-Term Memory, LSTM). LSTM is a variant of the recurrent neural network and is particularly suitable for processing sequence data with long-term dependencies. In the battery safety risk prediction, LSTM can capture the change pattern of the multi-attribute characteristics of the battery over time, thereby achieving early warning of safety risks.

[0047] When establishing a multi-attribute feature time series prediction model, the input and output of the model are first defined. The input includes a sampling sequence of multiple key attribute features of the target lithium battery over a period of time, such as voltage, current, temperature, etc. The output is a prediction of the battery safety risk status, such as high, medium, and low safety risk. The samples of safety risks of the preprocessed multi-attribute feature set are divided into two parts at an 8:2 ratio, namely the training set and the validation set. The training set is used to train the model so that it can learn the temporal relationship between attribute features and the leading indicator features of safety risks, and the validation set is used to evaluate the performance of the model.

[0048] The preprocessed multi-attribute feature sets are traversed and combined, and an LSTM time series prediction model is trained for each attribute combination. The LSTM models constructed by all attribute combinations are combined together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. This processing method can better utilize the complementary information between each attribute feature and improve the overall prediction ability of the model.

[0049] (3-2) Online selection of a feature set that effectively characterizes battery safety risk attributes In the LSTM neural network, each connection between neurons has a weight value, which represents the strength of the association between them. These weight values ​​are continuously updated and optimized through the back-propagation algorithm during the training process. Larger weight values ​​usually mean that the corresponding attribute features have a greater impact on the model's predictive ability. From the battery multi-attribute feature sample data, using a data-driven approach to extract high-weight original attribute features and eliminate invalid attribute features can enhance model stability.

[0050] Attribute features are selected based on the model prediction accuracy of the time series prediction model of multiple attribute features. By calculating the contribution value of each original attribute feature and sorting them by size, the top-k attribute features that are strongly related to safety risk prediction can be selected. These k attribute features are considered to be the most critical part of the model and play an important role in the accuracy of safety risk prediction. This embodiment uses the SHAP algorithm to analyze the key attribute features that affect battery safety risks. For each attribute feature, its contribution value to the prediction result is: Among them, f(x) represents the prediction result, E[f(x)] represents the expected value of the prediction result, and v(S) represents the shap value of the attribute feature subset S; Represents the shap value of the new subset after adding attribute feature i.

[0051] Normalization can make the contribution value of each attribute feature on a unified scale, which is convenient for subsequent analysis and comparison. At the same time, the normalized weight w[k] can also reflect the relative importance of each attribute feature to security risk prediction, which helps to further optimize the model and extract key features. Specifically, the weight value of each attribute feature can be divided by the sum of the weight values ​​of all k attribute features to obtain the normalized weight w[k].

[0052] The calculation formula of the normalized weight w[k] is as follows: W[k]= Among them, weight[k] represents the weight value of the kth attribute feature; Represents the sum of all k attribute feature weight values.

[0053] Step 4: Label Data Augmentation (1) Sample Difference Measurement Sample difference measurement refers to designing a method to quantify the degree of difference between two lithium battery safety risk samples. It requires comprehensive consideration of multiple attribute features and weighting them according to their importance. DTW is a method for measuring the similarity between two sequences, and is particularly suitable for processing time series data with different lengths or change rates. In lithium battery safety risk prediction, DTW can be used to compare whether the behavior patterns of two samples on a specific attribute feature are similar.

[0054] D(a,b)=Σ(w[k]*DTW(a[k],b[k])) Where D(a,b) represents the difference measure between sample a and sample b; w[k] represents the normalized weight of the k-th attribute feature; DTW(a[k],b[k]) represents the dynamic time warping (DTW) distance between sample a and sample b at the k-th attribute feature that effectively characterizes the battery safety risk.

[0055] (2) Label propagation The previously proposed sample difference measurement method D(a,b) is used to label the unlabeled training samples. Label propagation is a semi-supervised learning technique that achieves data enhancement and model performance improvement by transferring information of known labels to unlabeled samples. Specifically, for an unlabeled training sample a, we can infer its possible label and enhance the data sample by calculating the difference measurement D(a,b) between it and other labeled samples b, where the feature attributes used in the difference measurement are the online selected feature set obtained in step 3 that effectively characterizes the battery safety risk attributes.

[0056] The specific process is as follows: Initialization: Assign an initial label to each unlabeled sample, usually randomly or based on some heuristic rules.

[0057] Iterative update: Based on the difference measure D(a,b) between samples, the weight between each unlabeled sample and the surrounding labeled samples is calculated. The larger the weight, the more similar the two samples are, and the greater the impact of label transfer.

[0058] Label propagation: Update the labels of unlabeled samples according to the weights. This process is implemented by the weighted average aggregation method.

[0059] Convergence judgment: Repeat the iterative update steps until the convergence condition is met (such as the number of iterations reaches a predetermined value or the label change is less than a certain threshold).

[0060] Through the above methods, unlabeled training samples are labeled to achieve labeled data expansion. These labeled samples can be added to the training set for further training and optimization of the model. Data augmentation can increase the amount and diversity of training data and improve the generalization ability and robustness of the model.

[0061] Step 5: Retrain the time series prediction model cluster after multiple rounds of iterations until the proportion of unlabeled samples is lower than the threshold, and output the battery safety risk prediction model.

[0062] Step 6: Battery safety risk prediction When the battery management and control platform collects battery operating data, it needs to predict and analyze safety risks. The data collection process usually involves real-time monitoring and recording of multiple key attribute characteristics, such as battery voltage, current, temperature, etc.

[0063] Input the trained battery safety risk prediction model to obtain the corresponding safety risk prediction results, perform real-time safety risk prediction and analysis on new lithium battery operating condition samples, and no longer issue intelligent peak-shifting strategies for base stations with high-risk batteries to ensure the safe operation of base stations.

[0064] The battery safety risk prediction model outputs the prediction results as the probabilities of high risk, medium risk, and low risk. The highest probability that is greater than the preset value is the battery risk prediction level. For example, if the high risk probability is 60%, the medium risk probability is 30%, and the low risk probability is 10%, then the final output result is high risk.

[0065] Example 2 From September to November 2024, batteries and data acquisition equipment will be deployed in the existing computer room of Shaanxi Branch of China Tower Co., Ltd., and management platform software will be deployed as follows: Pilot station 1: West entrance of Biyuan Road, Qindu, Xianyang On-site working conditions: site load is about 92.9A, working voltage: 54.4V; Number of power supplies: one set of Vertiv multi-tenant power supplies, six 50A rectifiers; ZTE smart lithium battery: equipped with 3 sets of ZTE smart lithium battery ZXDC48FB100B3; Other battery packs: 5 sets of Sacred Sun ordinary lithium batteries.

[0066] Pilot station 2: Xianyang Weicheng 202 Research Institute On-site working conditions: site load is about 80A, working voltage: 55.1V; Number of power supplies: 1 set of Huawei multi-tenant power supply, 6 50A rectifiers; ZTE smart lithium battery: equipped with 4 sets of ZTE smart lithium battery ZXDC48 FB100B3; Other battery packs: 1 set of Pland 200AH ordinary lithium battery.

[0067] The utility description is shown in the following table: According to the test results of the pilot station, the following data were obtained: Lithium battery charge and discharge efficiency = discharge amount / charge amount ≈ 92%; The site load at the west entrance of Qindu Biyuan Road is relatively high, with an average load of about 91A. The three groups of smart lithium batteries stop discharging when the discharge depth reaches 50%, and the actual single discharge is less than 2h. Weicheng 202 Research Institute has a low site load, with an average load of about 76A. The four groups of smart lithium batteries are discharged for 2 hours, and the actual discharge depth does not reach 50%.

[0068] The economic benefit indicators are calculated as follows: Solution 1: The site already has ZTE smart lithium batteries, and no additional batteries are needed (calculated based on 2 smart lithium batteries per site) The site intelligent lithium battery is used for peak shifting, with a discharge depth of 50%, a charge and discharge efficiency of 92%, and one charge and one discharge; Annual off-peak income of a single battery = 48V*100Ah / 1000*50%*(1.1 yuan / kW·h-(0.38 yuan / kW·h / 92%))*365=602 yuan; Single-station investment: BCUA purchase and installation costs about 1,250 yuan, network management platform construction and operation and maintenance costs are shared at 200 yuan / station / year, the initial investment of a single station is 1,250 yuan, and the operation and maintenance costs during the contract period are 1,000 yuan; Calculated based on a 5-year contract period, the static payback period = 1,250 yuan ÷ (602*2*5-(1,250+1,000)+1,250)*10 years ≈ 2.49 years.

[0069] Solution 2: There is no ZTE smart lithium battery at the site, and additional batteries are added (calculated based on adding 2 smart lithium batteries to a single site) The site has added a new circulating intelligent lithium battery for peak shifting, with a cycle number of no less than 6,000 times, configured for 2-hour discharge, a discharge depth of 90%, a charge and discharge efficiency of 95%, and two charges and two discharges; Annual off-peak income of a single battery = 48V*100Ah / 1000*90%*(1.1 yuan / kW·h-(0.38 yuan / kW·h / 92%)+1.1 yuan / kW·h-(0.7 yuan / kW·h / 92%)*365=1618 yuan; Single-station investment: 5,500 yuan / unit for newly purchased circulating smart lithium batteries, 500 yuan / unit for newly added power modules, about 720 yuan / unit for battery installation, about 400 yuan / unit for power module installation, about 1,250 yuan / station for BCUA purchase and installation, 200 yuan / station / year for network management platform construction and operation and maintenance costs, 400 yuan / station / year for newly added batteries and rectifier modules, and 2 new circulating smart lithium batteries and 2 power modules are added to a single station. The initial investment cost of a single station is about 15,490 yuan, and the operation and maintenance costs during the contract period are about 6,000 yuan; Calculated based on a 10-year project cycle, static payback period = 15,490 yuan ÷ (1,618*2*10-(15,490+6,000)+15,490)*10 years ≈ 5.88 years.

[0070] The present application discloses a battery application management and control method. Based on multiple battery attribute characteristics, a time series prediction model cluster of multiple attribute characteristics is dynamically constructed, and the attribute characteristics that are most valuable for safety risk prediction are selected online to adapt to the safety risk prediction needs of batteries of different manufacturers and models under different working conditions. The label propagation algorithm is used to expand the safety risk sample data, deeply mine and make full use of unlabeled data, so as to predict the safety risks of lithium batteries more accurately and stably.

[0071] This application builds a time series prediction model cluster based on multiple battery attribute features, selects multi-attribute leading indicator features that effectively characterize battery safety risks online, enhances model robustness, and solves the problem of difficulty in determining effective attribute features.

[0072] This application is based on the multi-attribute advance indicator features and label propagation algorithms that effectively characterize battery safety risks obtained through online selection, expands risk sample data, iteratively trains the battery safety risk prediction model, improves the generalization ability and robustness of the model, and solves the problem of small number of battery safety risk samples; This application uses a data-driven approach to extract high-weight original attribute features and eliminate invalid attribute features from battery multi-attribute feature sample data, which can enhance model stability.

[0073] When constructing a time series prediction model cluster of multi-attribute features, the present application traverses and combines the preprocessed multi-attribute feature set, trains an LSTM time series prediction model for each attribute combination, and combines the LSTM models constructed by all attribute combinations together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. This processing method can better utilize the complementary information between each attribute feature and improve the overall prediction ability of the model.

[0074] The above contents are only specific implementation methods 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 included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A battery application management and control method, characterized in that: The following steps are involved: Step 1: Obtain battery operation business data as a multi-attribute feature set of the battery; Step 2: Preprocessing of battery operation business data, including but not limited to deleting erroneous values, duplicate values, replacing outliers, and filling missing values; Step 3: Build a time series prediction model cluster with multiple attribute features, and select a set of attribute features that effectively characterize battery safety risks based on the prediction accuracy of each model in the model cluster; Step 3-1: Build a time series prediction model cluster with multiple attribute features; Establish a time series prediction model of multi-attribute features. The model input includes a sampling sequence of multiple attribute features within a specific time of the battery. The model output is a battery safety risk prediction. The time series prediction model of multi-attribute features is trained and evaluated using the preprocessed safety risk samples of the multi-attribute feature set or the updated label data. The preprocessed multi-attribute feature set is traversed and combined, and a time series prediction model is trained for each attribute feature combination. The models constructed by all attribute feature combinations are combined together to form a multi-attribute feature time series prediction model cluster; Step 3-2: Online selection of a feature set of attributes that effectively characterize battery safety risks based on the prediction accuracy of each model in the time series prediction model cluster with multi-attribute features; Step 4: Label data expansion: Based on the effective feature set representing the battery safety risk attribute obtained through online selection, the unlabeled samples in the preprocessed multi-attribute feature set are labeled through the differential measurement algorithm and label propagation algorithm to obtain updated label data; Step 5: Retrain the multi-attribute feature time series prediction model cluster based on the updated label data, and after multiple rounds of iterations, until the proportion of unlabeled samples is lower than the threshold, output the battery safety risk prediction model; Step 6: Battery safety risk prediction; When the battery management and control platform collects battery operating condition data, it inputs it into the trained battery safety risk prediction model to obtain the corresponding safety risk prediction results. For base stations with high-risk batteries, intelligent peak-shifting strategies will no longer be issued to ensure the safe operation of the base stations.

2. A battery application management and control method according to claim 1, characterized in that: In step 1, the battery operation service data is divided into three types, namely, battery operation data, battery status data and battery alarm data. As a multi-attribute feature set of the battery, the battery operation data includes the voltage, temperature and humidity data during the battery charging and discharging process, the battery status data includes the capacity, remaining capacity, maximum available capacity, internal resistance and life, and the battery alarm data includes voltage abnormality and temperature abnormality.

3. A battery application management and control method according to claim 1, characterized in that: The specific preprocessing of the battery operation service data in step 2 is as follows: Delete the wrong values ​​and duplicate values; For outliers, based on the box plot, the third quartile is used to replace high outliers in the data, and the first quartile is used to replace low outliers in the data; For missing values, linear interpolation is used to fill in the missing values.

4. A battery application management and control method according to claim 1, characterized in that: The step 3 is specifically as follows: Step 3-1: Build a time series prediction model cluster with multiple attribute features; The long short-term memory network LSTM is selected as the time series prediction model of multi-attribute features. When establishing the time series prediction model of multi-attribute features, the input and output of the model are first defined. The input includes a sampling sequence of multiple attribute features within a specific time of the battery. The output of the model is the battery safety risk prediction. The samples of the safety risk of the preprocessed multi-attribute feature set or the updated label data are divided into two parts at a ratio of 8:2, namely, a training set and a validation set. The time series prediction model of multi-attribute features is trained. The training set is used to train the model so that it can learn the temporal relationship between attribute features and the leading indicator features of safety risks. The validation set is used to evaluate the performance of the model. The preprocessed multi-attribute feature set is traversed and combined. Each attribute feature combination trains an LSTM time series prediction model. The LSTM time series prediction models constructed by all attribute feature combinations are combined together to form an LSTM model cluster. Each model in the cluster is responsible for processing different attribute features or attribute feature combinations. Step 3-2: Online selection of a feature set that effectively characterizes battery safety risk attributes; Attribute features are selected according to the model prediction accuracy of the time series prediction model with multiple attribute features, and the contribution value of each attribute feature is calculated and sorted by size. The top-k attribute features that are strongly correlated with safety risk prediction are selected to obtain a set of attribute features that effectively characterize battery safety risks, and the k attribute features are normalized.

5. A battery application management and control method according to claim 4, characterized in that: The calculation formula of the contribution value of each attribute feature is: in: Indicates the prediction result; Represents the expected value of the prediction result; v(S) represents the shap value of the attribute feature subset S; Represents the shap value of the new subset after adding attribute feature i.

6. A battery application management and control method according to claim 4, characterized in that: The calculation formula of the normalized weight w[k] is as follows: W[k]= in: weight[k] represents the weight value of the kth attribute feature; Represents the sum of all k attribute feature weight values.

7. A battery application management and control method according to claim 6, characterized in that: The step 4 is specifically as follows: Step 4-1: Sample difference measurement; In battery safety risk prediction, DTW is used to compare whether the behavior patterns of two samples on a specific attribute feature are similar: D(a,b)=Σ(w[k]*DTW(a[k],b[k])) in: D(a,b) represents the difference measure between sample a and sample b; w[k] represents the normalized weight of the kth attribute feature; DTW(a[k],b[k]) represents the dynamic time warping distance between the attribute features of sample a and sample b at the kth time that effectively characterizes the battery safety risk; Step 4-2: Label Propagation Initialization: Assign an initial label to each unlabeled sample, usually randomly; Iterative update: According to the difference measure D(a, b) between samples, the weight between each unlabeled sample and the surrounding labeled samples is calculated. The larger the weight, the more similar the two samples are, and the greater the impact of label transfer. Label propagation: Update the labels of unlabeled samples according to the weights. This process is achieved through the weighted average aggregation method. Convergence judgment: Repeat the iterative update steps until the convergence condition is met, the number of iterations reaches a predetermined value or the label change is less than a threshold; Label the unlabeled samples in the preprocessed multi-attribute feature set to obtain updated label data.

8. A battery application management and control method according to claim 1, characterized in that: The specific steps of step 6 are as follows: when a new battery operating condition sample is collected and transmitted to the system, it is input into the trained battery safety risk prediction model to obtain the safety risk prediction result, and the new battery operating condition sample is subjected to real-time safety risk prediction and analysis. For base stations with high-risk batteries, the intelligent peak shifting strategy will no longer be issued to ensure the safe operation of the base station.

9. A battery application management and control method according to claim 1, characterized in that: The battery application management method is executed by a battery management and control platform, which includes a basic function module, a battery health management module, a battery backup power analysis module, a battery life prediction module, an intelligent peak-shifting charging and discharging module, and a peak-shifting benefit calculation module; The basic function module can obtain and view the battery operation service data of the base station, including: configuration, battery operation data, battery status data and battery alarm data, and remotely control the battery; The battery health management module can evaluate the health status of each battery group based on the battery operation business data to obtain safety risk prediction results; support battery monitoring management of the entire network / subnet / site, remind through notifications and alarms, and provide relevant reports; The battery backup power analysis module can analyze the backup power capacity of a site or power supply system based on battery operation service data, load current, and related alarms; supports statistical analysis of the entire network / subnet / site, and provides backup power capacity analysis results; The battery life prediction module can predict the remaining life of the battery based on the battery installation date, ambient temperature, battery type, and battery operation condition data; The intelligent off-peak charging and discharging module can remotely set the off-peak power consumption parameters of each battery group, enable and disable the off-peak power consumption function, query, set, modify and delete the off-peak power consumption mechanism, and realize low-cost self-charging and high-cost self-discharging; The off-peak benefit calculation module can view the detailed information of off-peak power consumption of each battery group in real time, count and analyze the off-peak power consumption data and give the results, and evaluate the benefits of off-peak power consumption.

Images