Novel energy storage intelligent management method and system based on big data

Abstract

The embodiment of the invention provides a novel energy storage intelligent management method and system based on big data, and the method comprises the steps: collecting the operation data of an energy storage system in real time, and carrying out the preprocessing of the operation data; constructing a data set according to the preprocessed operation data, performing clustering analysis on the data in the data set, and screening out a training sample set according to a clustering result; based on the screened training sample set, an improved Adaboost algorithm is adopted to carry out iterative training so as to construct a strong classifier; and inputting energy storage system operation data collected in real time into the strong classifier, outputting a classification prediction result of the energy storage state, and performing intelligent scheduling management on the energy storage system according to the classification prediction result. According to the technical scheme provided by the invention, comprehensive monitoring, intelligent scheduling and optimal management of the energy storage system can be realized, so that the operation efficiency and the energy utilization rate of the energy storage system are improved, and optimization and sustainable development of an energy structure are promoted.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of energy storage system management, and particularly relates to a new energy storage intelligent management method and system based on big data. BACKGROUND

[0002] With the transformation of global energy structure, the proportion of new energy is increasing, which makes new energy storage technology become crucial. Lithium ion battery, flow battery and compressed air energy storage technology not only play a key role in grid peak shaving and frequency modulation, but also have a profound impact on the sustainable use of energy and environmental protection. The innovation and application of these technologies are gradually changing our energy consumption mode and providing the possibility for building a more efficient and clean energy system.

[0003] The current management of energy storage systems mainly relies on manual monitoring and traditional data analysis methods. These methods are not up to the task when dealing with large-scale data, and it is difficult to achieve real-time monitoring and accurate prediction of energy storage systems. In addition, most existing management systems lack intelligent functions and cannot automatically adjust the operating state of the energy storage system according to real-time data, thereby limiting the performance of the energy storage system. Therefore, the existing technology still has the following shortcomings:

[0004] (1) Planning lag: Due to the intermittency and uncertainty of new energy generation, the planning and construction of energy storage systems often fail to match the demand of new energy generation, resulting in low utilization of energy storage systems, and even in some cases, over-construction or under-construction. This not only wastes valuable resources, but also limits the potential of new energy storage technology in the transformation of energy structure;

[0005] (2) Data island: Equipment, projects, and grid data are scattered, and lack the ability to integrate and analyze multi-source data. In this case, data resources cannot be shared, leading to fragmentation of information. Due to the lack of multi-source data integration and analysis capabilities, organizations are unable to grasp and utilize these scattered data resources as a whole, which affects the accuracy and efficiency of decision-making;

[0006] (3) Precision: In traditional methods, there is a lack of effective mechanisms to deal with the accumulation of errors that may occur during the prediction process. Over time, these errors will gradually accumulate and amplify, eventually leading to a large deviation in the generation of control effects. This deviation can negatively affect the performance and accuracy of the entire system, thereby affecting the correctness of the decision and the effectiveness of the operation. SUMMARY

[0007] The new energy storage intelligent management method and system based on big data provided by the embodiments of the present application can comprehensively monitor, intelligently schedule and optimally manage the energy storage system, thereby improving the operation efficiency and energy utilization rate of the energy storage system and promoting the optimization of energy structure and sustainable development.

[0008] In a first aspect, the embodiments of the present application provide a new energy storage intelligent management method based on big data, comprising:

[0009] Real-time collection of operation data of the energy storage system, and preprocessing of the operation data to obtain preprocessed operation data;

[0010] Construction of a data set according to the preprocessed operation data, clustering analysis of data in the data set, and screening of a training sample set according to the clustering result;

[0011] Based on the screened training sample set, an improved Adaboost algorithm is used for iterative training to construct a strong classifier; wherein the improved Adaboost algorithm includes a weight threshold adjustment mechanism;

[0012] The real-time collected operation data of the energy storage system is input into the strong classifier, and a classification prediction result of the energy storage state is output, and the energy storage system is intelligently scheduled and managed according to the classification prediction result.

[0013] In an optional implementation, the operation data of the energy storage system includes voltage data, current data, temperature data and cumulative charge and discharge power data.

[0014] In an optional implementation, after real-time collection of operation data of the energy storage system and preprocessing of the operation data to obtain preprocessed operation data, the method further comprises:

[0015] Real-time battery state of charge estimation based on the preprocessed operation data to obtain a battery state of charge estimation value;

[0016] When the change rate of the battery state of charge estimation value exceeds a first preset threshold, an abnormal alarm is triggered.

[0017] In an optional implementation, the real-time battery state of charge estimation based on the preprocessed operation data to obtain a battery state of charge estimation value comprises:

[0018] The basic value of the state of charge is calculated by using the ampere-hour integral method;

[0019] The basic value is dynamically corrected by using an extended Kalman filter algorithm to obtain the state of charge estimation value.

[0020] In one alternative implementation, the battery state-of-charge estimation further includes a temperature compensation model for adjusting the battery internal resistance and nominal battery capacity based on temperature changes.

[0021] In one alternative implementation, cluster analysis is performed on the data in the dataset, including:

[0022] Initialize class centers by randomly selecting K vectors from the dataset as initial class centers;

[0023] Calculate the Euclidean distance of each vector to each initial class center, and assign all feature vectors to their corresponding classes according to the nearest neighbor principle;

[0024] For each class, calculate the sum of distances from all vectors within the class to other vectors within the class, and update the vector with the smallest sum of distances as the new class center;

[0025] Iteratively update the cluster centers until they no longer change, and obtain the clustering result.

[0026] In one alternative implementation, the Euclidean distance is calculated using the following formula:

[0027] ;

[0028] Where r represents the i-th vector D i With the kth class center C k The Euclidean distance between them, where t represents the dimension of the t-th time-series load feature vector, T represents the total number of dimensions of the time-series load feature vectors, and D it D represents the i-th vector. i The attribute value C on the t-th time-series load feature vector dimension kt Represents the k-th class center C k The attribute value in the t-th time-series load feature vector dimension.

[0029] In one optional implementation, a training sample set is selected based on the clustering results. Specifically, the sum of the distances from each vector in each class to other vectors in that class is calculated, and vectors whose sum of distances is greater than a second preset threshold are removed. The remaining vectors are then used to form the training sample set.

[0030] In one alternative implementation, based on the selected training sample set, an improved Adaboost algorithm is used for iterative training to build a strong classifier, specifically including:

[0031] Initialize the weights of each training sample;

[0032] Train a weak classifier and calculate its error rate;

[0033] Update the weights of the weak classifier and the sample weights based on the error rate;

[0034] A strong classifier is formed by combining multiple weak classifiers.

[0035] In one alternative implementation, when updating the weights of the weak classifier and the sample weights based on the error rate, the sample weights are reset to their initial values ​​when they exceed a weight threshold.

[0036] In one optional implementation, the weight threshold ζ ranges from 0.4 < ζ < 0.6.

[0037] In one alternative implementation, the strong classifier is obtained by weighted summation of all the weak classifiers.

[0038] In one alternative implementation, the operational data of the energy storage system is preprocessed based on the Spark big data platform.

[0039] Secondly, embodiments of this application provide a novel intelligent energy storage management system based on big data, comprising:

[0040] The data acquisition module is used to collect the operating data of the energy storage system in real time and preprocess the operating data to obtain preprocessed operating data.

[0041] The training sample set construction module is used to construct a dataset based on the preprocessed running data, perform cluster analysis on the data in the dataset, and select the training sample set based on the clustering results.

[0042] The classifier training module is used to iteratively train a strong classifier based on the selected training sample set using the improved Adaboost algorithm; the improved Adaboost algorithm includes a weight threshold adjustment mechanism.

[0043] The output module is used to input the real-time collected energy storage system operation data into the strong classifier, output the classification result of the energy storage status, and generate early warning information based on the classification result.

[0044] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method provided in embodiments of this application.

[0045] Fourthly, embodiments of this application provide a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed in a computer, causes the computer to perform the method provided in embodiments of this application.

[0046] The technical solution provided in this application has the following beneficial effects:

[0047] This invention proposes a classification and prediction method for energy storage data that combines the Spark big data platform with an improved Adaboost algorithm. The Spark platform enables distributed processing of large-scale data, improving data processing efficiency. Simultaneously, the improved Adaboost algorithm is used for iterative training of energy storage data, constructing a high-precision strong classifier that effectively avoids overfitting and improves the model's generalization ability. Furthermore, this invention employs a specific clustering method to select representative training samples. Through in-depth analysis of energy storage data, it can automatically identify and aggregate data points with high similarity, thereby quickly selecting representative training samples. This method not only improves the efficiency of data processing but also effectively reduces the impact of noisy data on model training, further enhancing the accuracy of classification prediction. This invention introduces an innovative weight threshold adjustment mechanism into the Adaboost algorithm. When the weight of a weak classifier increases beyond a preset threshold, the weight of that weak classifier is reset to its initial value, effectively limiting its influence in subsequent iterations. This weight threshold adjustment mechanism not only prevents a single weak classifier from occupying too large a proportion in the strong classifier, causing the model to become overly reliant on the performance of that weak classifier, but also promotes a balanced contribution among different weak classifiers, improving the model's robustness and generalization ability. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a novel intelligent energy storage management method based on big data, provided in an embodiment of this application.

[0049] Figure 2 This is a schematic diagram of a novel intelligent energy storage management system based on big data, provided in an embodiment of this application. Detailed Implementation

[0050] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] This application provides a novel intelligent energy storage management method based on big data. Figure 1 This is a flowchart illustrating a novel intelligent energy storage management method based on big data, provided in an embodiment of this application. The method can be executed by a novel intelligent energy storage management system based on big data. The system can be implemented by software and / or hardware and can be configured in electronic devices such as computers.

[0052] like Figure 1 As shown, the technical solution provided in this application includes the following steps:

[0053] S110. Real-time acquisition of operating data from the energy storage system, and preprocessing of the operating data to obtain preprocessed operating data.

[0054] In some embodiments, the operating data of the energy storage system includes voltage data, current data, temperature data, and cumulative charge / discharge capacity data. Real-time acquisition of operating data via IoT devices requires a combination of sensor networks and embedded algorithms; the hardware system architecture is as follows:

[0055] The sensor layer includes:

[0056] Voltage / current sensor: measures battery terminal voltage V and charging / discharging current I;

[0057] Temperature sensor: Collects the battery surface temperature T;

[0058] Coulomb counter chip: Calculates the cumulative charge / discharge capacity Q cum .

[0059] Edge computing units include:

[0060] The running data is filtered in real time using an embedded MCU.

[0061] Data is uploaded to the cloud platform via LoRa / Wi-Fi / NB-IoT.

[0062] In some embodiments, the operational data of the energy storage system is preprocessed based on the Spark big data platform.

[0063] In some embodiments, the method further includes using the embedded MCU of the edge computing unit to perform real-time battery state of charge estimation based on preprocessed operating data to obtain an estimated battery state of charge value.

[0064] Specifically, it includes:

[0065] The basic values ​​for the state of charge are calculated using the ampere-hour integration method.

[0066] The calculation formula is as follows:

[0067] ;

[0068] In the formula: SOC0 represents the initial SOC value, and SOC(t) is the estimated state of charge at time t; This indicates the nominal capacity of the battery, measured in Ah. This represents the coulombic efficiency, where η < 1 during charging and η < 1 during discharging. .

[0069] The extended Kalman filter algorithm is used to dynamically correct the base values ​​to obtain the estimated state of charge.

[0070] The state equations and observation equations are shown below:

[0071] ;

[0072] Where: state variables u q Indicates control input, z q Represents the observed value, x q The state variable (including SOC, battery internal resistance, etc.) is represented by q, which represents the time index. The charging and discharging current I is used as the control input, and the battery terminal voltage V is used as the observed value. The internal resistance of the battery is represented and dynamically updated to compensate for the effects of temperature and aging; the estimated state of charge is iteratively corrected by real-time voltage V and current I.

[0073] In some embodiments, the battery state-of-charge estimation further includes a temperature compensation model, which dynamically updates the battery internal resistance to compensate for the effects of temperature and aging. and capacity As temperature changes, a compensation relationship needs to be established:

[0074] ;

[0075] ;

[0076] In the formula: α represents the material coefficient, such as α≈0.02 / ℃ for NCM batteries; β represents the capacity temperature coefficient, such as β≈0.005 / ℃.

[0077] Charge and discharge efficiency calculation:

[0078] Charge and discharge efficiency Defined as the ratio of output energy to input energy:

[0079] ;

[0080] After considering internal resistance losses, the correction is as follows:

[0081] ;

[0082] Among them, V d Indicates the discharge voltage, I d Represents the discharge current, V c Indicates the charging voltage, I c Let t represent the charging current, t represent the time, t1 represent the start time of the charging / discharging process, and t2 represent the end time of the charging / discharging process.

[0083] Real-time data processing flow:

[0084] ① Raw data acquisition: Read the raw V, I, and T signals at a fixed sampling frequency; eliminate high-frequency noise through moving average filtering.

[0085] ;

[0086] Among them, V f (t) represents the voltage value after moving average filtering, M represents the length of the moving window, m represents the index variable, and Δt represents the sampling time interval.

[0087] ② Edge-end SOC pre-calculation: Run the lightweight Extended Kalman Filter (EKF) algorithm in the MCU to output the battery state of charge estimate (SOC estimate); when the rate of change of the battery state of charge estimate exceeds the first preset threshold (e.g., 5% / second), an abnormal alarm is triggered.

[0088] ③ Cloud data fusion and optimization: Combine historical data and multi-battery pack information to optimize global parameters.

[0089] Real-time data from a single battery can only reflect its current instantaneous state, which has limitations. By fusing historical data and data from similar batteries, including preprocessed real-time data uploaded from the edge (steps ① and ②), such as filtered voltage / current, SOC estimates, and temperature, as well as massive historical operating data stored in the cloud database, including past charge-discharge cycle records, performance degradation trends, and fault logs, the algorithm used at the edge (such as the noise parameters of the EKF algorithm) is retrained or fine-tuned with more comprehensive data to make the model more accurate before being distributed to the edge.

[0090] S120. Construct a dataset based on the preprocessed running data, perform cluster analysis on the data in the dataset, and select a training sample set based on the clustering results.

[0091] In some embodiments, cluster analysis is performed on the data in the dataset, including:

[0092] S121. Initialize class centers by randomly selecting K vectors from the dataset as initial class centers;

[0093] Dataset D s It consists of n eigenvectors, represented as In D s K vectors are randomly selected as the initial class centers. .

[0094] S122. Calculate the Euclidean distance of each vector to each initial class center, and classify all feature vectors into their corresponding classes according to the nearest neighbor principle;

[0095] The formula for calculating Euclidean distance is as follows:

[0096] ;

[0097] Where r represents the i-th vector Di With the kth class center C k The Euclidean distance between them, where t represents the dimension of the t-th time-series load feature vector, T represents the total number of dimensions of the time-series load feature vectors, and D it D represents the i-th vector. i The attribute value C on the t-th time-series load feature vector dimension kt Represents the k-th class center C k The attribute value in the t-th time-series load feature vector dimension.

[0098] S123. For each class, calculate the sum of distances from all vectors within the class to other vectors within the class, and update the vector with the smallest sum of distances as the new class center.

[0099] The calculation formula is as follows:

[0100] ;

[0101] In the formula: For the i-th vector D i up to the j-th vector D j The distance between the i-th vector D and the vector J is the distance between the i-th vector D and the vector J. i The number of feature vectors of the class; For the i-th vector D i The sum of distances to all vectors of the current class.

[0102] S124. Repeat S122-S123 to iteratively update the cluster centers until the cluster centers no longer change, and obtain the clustering results.

[0103] S125. Calculate the sum of distances from each vector in each class to other vectors in that class, and filter out vectors whose sum of distances is greater than a second preset threshold. The remaining vectors form a training sample set. The second preset threshold can be set according to the actual situation.

[0104] These rejected vectors are typically data that differs significantly from other samples within the class, possibly due to measurement errors, transient disturbances, or special operating conditions. Retaining them would "pollute" the training dataset, causing subsequent model training to learn incorrect patterns. Removing them results in a higher-quality, smaller-scale, and more representative training sample set, significantly improving the accuracy and robustness of the final trained model.

[0105] S130. Based on the selected training sample set, the improved Adaboost algorithm is used for iterative training to build a strong classifier; wherein, the improved Adaboost algorithm includes a weight threshold adjustment mechanism.

[0106] In some embodiments, an improved Adaboost algorithm is used for iterative training to build a strong classifier, specifically including:

[0107] Based on the selected training sample set, initialize the weights of each training sample;

[0108] Assuming there are N training samples, in the initial stage, each training sample is assigned the same weight value, as shown in the following formula:

[0109] ;

[0110] ;

[0111] ;

[0112] In the formula: The training weights of the first weak classifier on the i-th sample are represented by ; N represents the number of training samples. represents the output weight of the first weak classifier; i represents the sample sequence number.

[0113] Train a weak classifier and calculate its error rate;

[0114] ① In this round of classification, determine whether the sample points have been correctly classified.

[0115] ② If a sample point is correctly classified, its weight value is reduced in subsequent training; if a training sample point is misclassified, its weight value is increased in subsequent training.

[0116] ③ Update the weight values ​​of all sample points and proceed to the next round of training.

[0117] The entire training process repeats the above steps iteratively.

[0118] The error rate of the weak classifier is calculated using the following formula:

[0119] ;

[0120] In the formula: This represents the probability that the prediction of a weak classifier for samples in the training set is inconsistent with the true label. This represents the t-th weak classifier; This represents the i-th training sample; This represents the true label value of the i-th training sample; This represents the weight of the i-th sample in the t-th iteration; The weight of the i-th sample in the t-th iteration is adjusted based on the performance of the weak classifiers in the previous iteration, and is used to control the importance of different samples in the training process; the number of weak classifiers is equal to the number of iterations.

[0121] Update the weights of the weak classifier and the sample weights based on the error rate;

[0122] The weights of the weak classifiers in the final classifier are calculated as follows:

[0123] ;

[0124] In the formula: This represents the weight of the t-th weak classifier; Let represent the weighted error rate of the t-th weak classifier on the training set.

[0125] The formula for calculating sample update weights is:

[0126] ;

[0127] In the formula: This represents the weight of the i-th sample in the (t+1)-th iteration; Z represents the weight of the i-th sample in the t-th iteration; t Let represent the normalization factor for the t-th iteration, used to ensure that the sum of the weights of all samples is 1.

[0128] When a sample is misclassified Then the sample weights are updated. The calculation formula is:

[0129] ;

[0130] When the sample is correctly classified Then the sample weights are updated. The calculation formula is:

[0131] .

[0132] Furthermore, when the sample weights exceed the weight threshold, the sample weights are reset to their initial values.

[0133] Update sample weight distribution Because weight updates depend on ,and It also depends on the error rate. Therefore, the weight update formula can be directly applied. To prevent overfitting, which results in noisy data due to unrestricted weight increases or decreases, this study introduces a weight threshold during the weight update process. ( This weight threshold is used to limit weight updates, preventing weights from becoming too large or too small, thus avoiding an overly complex model and capturing noise in the training data. When the sample weights exceed... When this happens, the weights of the sample point are reset to their initial values. Sample weight update. The normalization factor Z for t iterations t The calculation formula is as follows:

[0134] ;

[0135] .

[0136] A strong classifier is formed by combining multiple weak classifiers.

[0137] The strong classifier is obtained by weighted summation of all weak classifiers. The calculation formula is as follows:

[0138] ;

[0139] In the formula: This represents the final strong classifier used to classify the input x; t represents the number of weak classifiers, i.e., the number of iterations. This represents the weight of the t-th weak classifier, reflecting its importance in the strong classifier; This represents the classification result output by the t-th weak classifier given input x.

[0140] S140. Input the real-time collected energy storage system operation data into the strong classifier, output the classification prediction results of the energy storage status, and perform intelligent scheduling and management of the energy storage system based on the classification prediction results.

[0141] For example, the classification prediction results may include: health status assessment, such as: normal, warning, fault; operation mode recognition, such as: idle, charging, discharging; abnormal behavior detection, such as: insulation fault, internal short circuit warning, etc.; and generating and executing control strategies based on the model prediction results.

[0142] This invention combines the Spark big data platform with the AdaBoost algorithm to significantly improve the efficiency and accuracy of the algorithm when processing large-scale energy storage data. The Spark platform provides powerful distributed computing capabilities, enabling the partitioning of large datasets into smaller chunks and parallel processing across multiple nodes, thereby drastically reducing data processing time. Simultaneously, the improved AdaBoost algorithm effectively avoids overfitting while maintaining classification accuracy, thus enhancing the model's generalization ability.

[0143] In the specific implementation process, the energy storage data is first preprocessed using the Spark platform, including data cleaning and feature extraction, to ensure the quality and consistency of the input data. Then, the preprocessed data is clustered using the aforementioned clustering method to select representative training samples. Next, the selected training samples are iteratively trained using an improved AdaBoost algorithm to gradually build a strong classifier. During training, a weight threshold mechanism is introduced to dynamically adjust the weights of the weak classifiers to prevent overfitting caused by excessively large or small weights. Finally, the trained strong classifier is applied to the classification and prediction of actual energy storage data to achieve accurate monitoring and early warning of energy storage status.

[0144] Figure 2 This is a schematic diagram of the structure of a novel intelligent energy storage management system based on big data, as provided in an embodiment of this application. Figure 2 As shown, the system includes:

[0145] The data acquisition module is used to collect the operating data of the energy storage system in real time and preprocess the operating data to obtain preprocessed operating data.

[0146] The training sample set construction module is used to construct a dataset based on the preprocessed running data, perform cluster analysis on the data in the dataset, and select the training sample set based on the clustering results.

[0147] The classifier training module is used to iteratively train a strong classifier based on the selected training sample set using the improved Adaboost algorithm; the improved Adaboost algorithm includes a weight threshold adjustment mechanism.

[0148] The output module is used to input the real-time collected energy storage system operation data into the strong classifier, output the classification result of the energy storage status, and generate early warning information based on the classification result.

[0149] The execution process of the system part of this application embodiment is the same as that of the method part of the embodiment described above, and will not be repeated here.

[0150] This application also provides an electronic device, including: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute any of the new energy-based power grid multi-resource coordinated control methods.

[0151] This application also proposes a computer storage medium storing a computer program, which, when executed by a processor, implements any one of the new energy-based power grid multi-resource coordinated control methods.

[0152] Computer storage media may be simply referred to as media. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Dual Data SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus Direct RAM (RDRAM), Direct Memory Bus Dynamic RAM (DRDRAM), and Memory Bus Dynamic RAM (RDRAM). The various embodiments described in this specification are presented in a progressive manner, and similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, for embodiments of apparatus, devices, and non-volatile computer storage media, since they are substantially similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to the description of the method embodiments.

[0153] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A new energy storage intelligent management method based on big data, characterized in that, The application relates to a method for predicting the state of a storage energy system. The method comprises the following steps: collecting and preprocessing operation data of the storage energy system to obtain preprocessed operation data; constructing a data set according to the preprocessed operation data, performing cluster analysis on the data in the data set, and screening a training sample set according to the cluster result; based on the screened training sample set, performing iterative training by using an improved Adaboost algorithm to construct a strong classifier; wherein the improved Adaboost algorithm comprises a weight threshold adjustment mechanism; inputting the real-time collected operation data of the storage energy system into the strong classifier, outputting a classification prediction result of the storage state, and intelligently scheduling and managing the storage energy system according to the classification prediction result.

2. The method of claim 1, wherein, The operation data of the storage energy system comprises voltage data, current data, temperature data and cumulative charge and discharge power data.

3. The method of claim 2, wherein, After the operation data of the storage energy system is collected and preprocessed to obtain preprocessed operation data, the method further comprises the following steps: based on the preprocessed operation data, performing real-time battery state of charge estimation to obtain a battery state of charge estimation value; when the change rate of the battery state of charge estimation value exceeds a first preset threshold, triggering an abnormal alarm.

4. The method of claim 3, wherein, The real-time battery state of charge estimation based on the preprocessed operation data to obtain a battery state of charge estimation value comprises the following steps: calculating a basic value of the state of charge by using an ampere-hour integral method; dynamically correcting the basic value by using an extended Kalman filter algorithm to obtain a state of charge estimation value.

5. The method of claim 4, wherein, The battery state of charge estimation further comprises a temperature compensation model for adjusting the battery internal resistance and the battery nominal capacity according to temperature changes.

6. The method of claim 1, wherein, The cluster analysis on the data in the data set comprises the following steps: initializing class centers, and randomly selecting K vectors from the data set as initial class centers; calculating the Euclidean distance of each vector to each initial class center, and dividing all feature vectors into corresponding classes according to the nearest neighbor principle; for each class, calculating the sum of distances of all vectors in the class to other vectors in the class, and updating the vector with the smallest distance sum as a new class center; iteratively updating the class centers until the class centers no longer change, and obtaining a cluster result.

7. The method of claim 6, wherein, The calculation formula of the Euclidean distance is as follows: ； wherein r represents the Euclidean distance between the ith vector D i and the kth class center C k , t represents the tth time-series load feature vector dimension, T represents the total number of time-series load feature vector dimensions, D it represents the ith vector D i at the tth time-series load feature vector dimension, C kt represents the kth class center C k at the tth time-series load feature vector dimension.

8. The method of claim 6, wherein, According to the cluster result, the training sample set is screened, specifically: calculating the sum of distances of each vector in each class to other vectors in the class, and screening out the vectors with a distance sum greater than a second preset threshold, and the remaining vectors constitute the training sample set.

9. The method of claim 1, wherein, Based on the screened training sample set, the improved Adaboost algorithm is used for iterative training to construct a strong classifier, specifically comprising the following steps: initializing the weights of each training sample; training a weak classifier and calculating the error rate of the weak classifier; updating the weight and sample weight of the weak classifier based on the error rate; the strong classifier is composed of multiple weak classifiers.

10. The method of claim 9, wherein, When the sample weight exceeds the weight threshold, the sample weight is reset to the initial value when the weight and sample weight of the weak classifier are updated based on the error rate.

11. The method of claim 10, wherein, The weight threshold ζ is in the range of 0.4 < ζ < 0.

6.

12. The method of claim 9, wherein, The strong classifier is obtained by weighted summation of all weak classifiers.

13. The method of claim 1, wherein, The operation data of the energy storage system are preprocessed based on a Spark big data platform.

14. A new energy storage intelligent management system based on big data, characterized in that, The method comprises the following steps: a data collection module is configured to collect operation data of the energy storage system in real time, preprocess the operation data, and obtain preprocessed operation data; a training sample set construction module is configured to construct a data set according to the preprocessed operation data, perform cluster analysis on data in the data set, and filter out a training sample set according to a cluster result; a classifier training module is configured to perform iterative training on the filtered training sample set by using an improved Adaboost algorithm to construct a strong classifier; wherein the improved Adaboost algorithm comprises a weight threshold adjustment mechanism; an output module is configured to input the real-time collected operation data of the energy storage system into the strong classifier, output a classification result of the energy storage state, and generate early warning information based on the classification result.

15. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of any one of claims 1-13.

16. A computer readable storage medium characterized by: A computer program is stored on the memory, and when the computer program is executed in the computer, the computer executes the method of any one of claims 1-13.

Images