Data processing method of distributed energy storage digital platform

Abstract

The invention relates to the technical field of energy storage, and aims to solve the problem that a distributed energy storage digital platform is difficult to meet large-scale, multi-source and heterogeneous data integration and analysis requirements in the aspect of data processing. In order to achieve the purpose, the data processing method of the distributed energy storage digital platform comprises the steps that battery data of energy storage equipment, geographic position data of the energy storage equipment, power load data, user power consumption behavior data and environment monitoring data are collected; performing data feature extraction and classification on the data based on an SVM decision function, wherein parameters of the SVM decision function are dynamically updated according to a predetermined period; performing data cleaning on each category of data after classification by adopting an adaptive cleaning algorithm, and storing the data according to a distributed data storage architecture; and performing multi-dimensional analysis on various data. According to the method, large-scale multi-source heterogeneous data can be collected, processed and analyzed, and the data processing efficiency and the intelligent level of the whole distributed energy storage digital platform are improved.

Description

Technical Field

[0001] The present invention relates to the field of energy storage technology, and in particular to a data processing method for a distributed energy storage digital platform. Background Art

[0002] With the transformation of the global energy structure and the increasing emphasis on the use of renewable energy, the application of distributed energy storage systems in the power sector has become more and more extensive. As a key technical means to integrate and manage distributed energy storage resources, the distributed energy storage digital platform has emerged and plays an extremely important role in the modern energy management system. Distributed energy storage systems are characterized by decentralization and diversity. They cover a variety of energy storage device types, such as lithium-ion batteries, lead-acid batteries and other battery systems with different chemical compositions, as well as new energy storage components such as supercapacitors. These energy storage devices are distributed in different geographical locations, including residential communities, commercial areas, industrial plants, and distributed power generation sites in remote areas. They are closely connected to the power load and are significantly affected by the user's electricity consumption behavior and surrounding environmental factors.

[0003] On the one hand, the power load exhibits complex fluctuation characteristics in different time periods. For example, in urban areas, commercial activities and industrial production are busy during the day, and the power load is at its peak; at night, most commercial and industrial electricity consumption decreases, but residential electricity consumption may remain at a certain level due to the use of household electrical appliances, resulting in significant peak and valley differences in power load. This dynamic change in power load poses a severe challenge to the energy storage system, requiring it to be able to accurately perform charging and discharging operations at the right time to achieve peak shaving and valley filling, and improve the stability of the power system and energy efficiency.

[0004] On the other hand, users' electricity consumption behavior is highly uncertain and personalized. Different user groups, such as residential users, commercial users and industrial users, have different electricity consumption habits, power requirements and electricity consumption time distribution. Residential users' electricity consumption behavior may be affected by factors such as daily routines, seasonal changes and holidays; commercial users are closely related to business hours and business formats; industrial users' electricity consumption depends on production plans, equipment operation cycles, etc. Therefore, in-depth analysis of user electricity consumption behavior is crucial to optimize the operation strategy of energy storage systems and meet users' personalized electricity needs.

[0005] Furthermore, environmental factors have an impact on the performance and life of energy storage equipment that cannot be ignored. Changes in environmental parameters such as ambient temperature, humidity, and light intensity will directly or indirectly affect the charging and discharging efficiency of energy storage equipment, the rate of battery capacity decay, and the reliability of the equipment. For example, a high temperature environment may accelerate the chemical reaction inside the battery, causing the battery capacity to drop too quickly, or even cause safety hazards; while a low temperature environment may reduce the activity of the battery and affect its charging and discharging performance.

[0006] In such a complex context, data processing in distributed energy storage digital platforms faces many challenges. Existing data processing methods often have difficulty in coping with such large-scale, multi-source heterogeneous data integration and analysis needs. The lack of an effective integration mechanism for data collection, storage, and management leads to low data utilization and the inability to fully tap the value behind the data. In addition, with the continuous development and expansion of distributed energy storage systems, the amount of data has exploded, placing higher requirements on the real-time and accuracy of data processing. Existing data processing technologies have certain limitations in processing speed and data analysis accuracy, and cannot meet the urgent needs of distributed energy storage digital platforms for efficient data processing and intelligent decision support.

[0007] In view of this, the art needs a data processing method for a distributed energy storage digital platform to solve the above problems. Summary of the invention

[0008] In order to solve the above technical problems, that is, to solve the problem that distributed energy storage digital platforms are difficult to cope with large-scale, multi-source heterogeneous data integration and analysis needs in data processing.

[0009] The present invention provides a data processing method for a distributed energy storage digital platform, the data processing method comprising:

[0010] Collect battery data of energy storage equipment, geographic location data of energy storage equipment, power load data, user power consumption behavior data and environmental monitoring data;

[0011] Based on the SVM decision function, data features are extracted and classified for the collected battery data of the energy storage device, the geographical location data of the energy storage device, the power load data, the user's power consumption behavior data and the environmental monitoring data, wherein the parameters of the SVM decision function are dynamically updated according to a predetermined period;

[0012] Use the appropriate cleaning algorithm to clean the classified data in each category;

[0013] The cleaned data is stored according to the distributed data storage architecture, where each type of data corresponds to a database node cluster that is adapted to it;

[0014] Analyze various types of data, and the analysis dimensions include at least energy storage system performance evaluation, energy storage battery failure prediction, charging and discharging strategy optimization analysis, user electricity consumption behavior analysis, energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis.

[0015] In some preferred embodiments, the SVM decision function is:

[0016]

[0017] Among them, x is the feature vector of the data to be classified, m is the number of training samples, and x i is the feature vector of the training sample, y i is the category label, α i is the Lagrange multiplier, K(x i ,x) is the kernel function and b is the bias term.

[0018] In some preferred embodiments, the kernel function K(x i ,x) is calculated as:

[0019]

[0020] Among them, σ is the kernel function parameter.

[0021] In some preferred embodiments, the parameters of the SVM decision function are dynamically updated according to a predetermined period, specifically including:

[0022] Initiate a dynamic classification rule update mechanism according to a predetermined period;

[0023] Extract a preset proportion of various data samples from the classified data warehouse as a training set;

[0024] Retrain using the SVM algorithm to update the parameters and decision boundaries of the SVM decision function;

[0025] The updated SVM decision function is applied to subsequent classification operations.

[0026] In certain preferred embodiments, the classified data includes a battery performance data class, and the cleaning step of the battery performance data class includes cleaning outliers based on a normal distribution model and smoothing the data using a moving average method;

[0027] Cleaning outliers based on the normal distribution model includes:

[0028] Calculate the battery voltage data sequence {V b1 ,V b2 ,…,V bn The mean of and standard deviation

[0029]

[0030] The calculated mean and standard deviation Determine the reasonable range Traverse the battery voltage data sequence {V b1 ,V b2 ,…,V bn Each data point V inbi , outliers outside the reasonable range are removed from the data sequence;

[0031] Where n is the battery voltage data sequence {V b1 ,V b2 ,…,V bn The total number of data points in V bi is the battery voltage data sequence {V b1 ,V b2 ,…,V bn}The voltage value of the i-th data point in;

[0032] The data smoothed by moving average include:

[0033] Determine the simple moving average window size m1;

[0034] The smoothed battery voltage data is calculated using the following formula:

[0035]

[0036] in, It is the battery voltage data at index i after being smoothed by the moving average method.

[0037] In certain preferred embodiments, the classified data includes a power load data class, and the cleaning step of the power load data class includes filling missing data using a data interpolation method and removing noise using a data smoothing algorithm;

[0038] Data interpolation methods to fill missing data include:

[0039] When it is detected that there are missing data points in the power load data, determine the two adjacent data points P load(i-1) and P load(i+1) ,

[0040] The padding value is calculated according to the following linear interpolation formula:

[0041]

[0042] in, is the power load value obtained by linear interpolation to fill the missing data points, P load(i-1) is the power load value of the adjacent data point before the missing data point, P load(i+1) is the power load value of the data point adjacent to the missing data point;

[0043] The use of data smoothing algorithms to remove noise includes:

[0044] Set the smoothing coefficient α;

[0045] When calculating the smoothed power load data, starting from the first data point of the power load data sequence, iterative calculation is performed according to the following formula:

[0046]

[0047] in, is the power load data at index i after being smoothed by the exponentially weighted moving average algorithm, P load (i) is the load value of the data point at index i in the original power load data sequence, It is the power load data at index i-1 after being smoothed by the exponentially weighted moving average algorithm.

[0048] In some preferred embodiments, the classified data includes a user electricity consumption behavior data class, and the cleaning of the user electricity consumption behavior data class includes processing electricity consumption time series data and processing electricity consumption power data;

[0049] The processing of electricity consumption time series data includes:

[0050] Sort the timestamps in ascending order, and then check whether the time intervals between adjacent timestamps are as expected;

[0051] For missing time points, the following cubic spline interpolation function S(z) is used to repair them:

[0052] S(z)=a zi +b zi (zz i )+c zi (zz i ) 2 +d zi (zz i ) 3

[0053] Among them, z is the timestamp, a zi 、b zi 、c zi and d zi are parameter coefficients, z i is the i-th time point;

[0054] The processing of power consumption data includes:

[0055] For daily power consumption data, the daily power consumption data is taken as a subset;

[0056] The K-Means clustering algorithm is used to determine the normal power range for each subset.

[0057] In certain preferred embodiments, the classified data includes an environmental data class, and the cleaning of the environmental data class includes analyzing the environmental data time series pattern using an autocorrelation function and a partial autocorrelation function and filling in missing data and smoothing data according to the time series law;

[0058] The use of autocorrelation functions and partial autocorrelation functions to analyze environmental data time series patterns includes:

[0059] Calculate the mean value of the ambient temperature, calculate the autocorrelation coefficient and draw an autocorrelation function graph, wherein the calculation formula of the autocorrelation coefficient ACF(k) is:

[0060]

[0061] Among them, n e is the ambient temperature time series The total number of data points, T ei is the ambient temperature value at the i-th moment, is the ambient temperature time series The average value of the ambient temperature, k is the lag order;

[0062] Use the Durbin-Levinson algorithm to calculate the partial autocorrelation function and draw the partial autocorrelation function graph;

[0063] Filling missing data and smoothing data according to time series rules include:

[0064] For missing data points, linear interpolation method was used to fill them;

[0065] The smoothed data is calculated using the following weighted moving average formula:

[0066]

[0067] in, is the ambient temperature value at index i after smoothing by weighted moving average method, m e is the moving average window size, w j is the weighting coefficient, T ej is the ambient temperature value at the jth moment.

[0068] In some preferred embodiments, the classified data includes energy storage device geographic location data, and cleaning the energy storage device geographic location data includes:

[0069] Determine the original coordinate system of the acquired geographic location data of the energy storage device;

[0070] For data in different original coordinate systems, use the coordinate conversion algorithm to uniformly convert them into the WGS84 coordinate system;

[0071] Check whether the latitude and longitude values ​​are within a reasonable geographical range;

[0072] Repair data for abnormal coordinate values ​​that are not within a reasonable geographical range;

[0073] Coordinate values ​​that cannot be repaired are marked as invalid data and discarded.

[0074] In some preferred embodiments, the data processing method further comprises:

[0075] Obtain the data selection operation entered by the user on the visual interface;

[0076] According to the input data selection operation, data is retrieved from the corresponding database and an intuitive chart form is generated for display.

[0077] It can be seen from the above that the data processing method of the distributed energy storage digital platform provided by the present invention has the following beneficial technical effects:

[0078] The present invention adopts SVM decision function to classify and extract features from multi-source data, taking into full account the characteristics of battery data, geographic location data, power load data, user power consumption behavior data and environmental monitoring data of energy storage equipment. Through the powerful classification ability of SVM, the key features in different types of data can be effectively distinguished, providing a solid foundation for subsequent accurate analysis and decision-making. For example, in the classification of battery data, the key features related to battery performance can be accurately identified, which helps to timely discover potential battery problems and take measures in advance to improve the safety and reliability of the energy storage system. Moreover, the SVM decision function parameters are dynamically updated according to a predetermined period, so that the data classification and analysis model can adapt to the changing trend of the data. With the passage of the operating time of the energy storage system, the change of user power consumption habits and the change of environmental conditions, the distribution characteristics and internal relationships of the data will also change. By regularly updating the SVM decision function parameters, it can be ensured that the model always maintains the ability to classify and effectively analyze the data with high accuracy, providing continuous and reliable support for intelligent decision-making. For example, when users' electricity consumption behavior changes significantly during seasonal changes or when social activities change, the dynamically updated model can capture these changes in a timely manner and adjust the analysis strategy, and optimize the operation strategy of the energy storage system accordingly to better meet user needs and adapt to changes in the electricity market, thereby improving the intelligence level and resilience of the entire distributed energy storage digital platform.

[0079] Furthermore, the use of adaptive cleaning algorithms for different types of data can remove noise, outliers, missing data and other problems in various types of data to the greatest extent. The cleaned data is stored according to the distributed data storage architecture, and each type of data corresponds to a dedicated database node cluster, which greatly improves the storage efficiency and query speed of the data. When a specific type of data needs to be called, the relevant data can be quickly obtained, and the data can be quickly located and extracted, which reduces the data retrieval time and improves the efficiency of the entire data processing process.

[0080] In addition, the present invention realizes multi-dimensional data analysis, covering energy storage system performance evaluation, fault prediction, charge and discharge strategy optimization analysis, user electricity behavior analysis, energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis. In terms of energy storage system performance evaluation, comprehensive multi-source data can comprehensively consider the performance of the energy storage system under different working conditions, obtain more accurate performance evaluation results through precise data analysis, and timely discover components or links with performance degradation and perform maintenance or optimization. For energy storage battery fault prediction, using classified and cleaned data and then analyzing it, battery failure can be predicted in advance, avoiding problems such as energy storage system shutdown caused by sudden battery failure, reducing maintenance costs and improving system availability. In the charge and discharge strategy optimization analysis, the impact of power load fluctuations, changes in user electricity demand and environmental factors on energy storage equipment is fully considered, and a more reasonable charge and discharge strategy is formulated to achieve peak shaving and valley filling, improve the economic benefits of the energy storage system, and also help stabilize the operation of the power network. Through user electricity behavior analysis, personalized service solutions can be provided to power companies to improve user satisfaction and the operational efficiency of the power market. In terms of energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis, combined with geographic location data and other relevant data, it is possible to determine the optimal layout location of energy storage equipment, improve the energy storage equipment's ability to regulate regional energy supply and demand, promote efficient use and reasonable allocation of regional energy, reduce energy transmission losses and improve the reliability of energy supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0082] Figure 1 The present invention is a flow chart of the data processing method of the distributed energy storage digital platform. DETAILED DESCRIPTION

[0083] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0084] Based on the problem pointed out in the background technology that the distributed energy storage digital platform is difficult to cope with the problem of large-scale, multi-source heterogeneous data integration and analysis needs in data processing. The present invention provides a data processing method for a distributed energy storage digital platform, aiming to enable the distributed energy storage digital platform to collect, process and analyze large-scale multi-source heterogeneous data, thereby improving the data processing efficiency and intelligence level of the entire distributed energy storage digital platform.

[0085] like Figure 1 As shown, the data processing method of the distributed energy storage digital platform of the present invention includes:

[0086] Collect battery data of energy storage equipment, geographic location data of energy storage equipment, power load data, user power consumption behavior data and environmental monitoring data;

[0087] Based on the SVM decision function, data features are extracted and classified for the collected battery data of the energy storage device, the geographical location data of the energy storage device, the power load data, the user's power consumption behavior data and the environmental monitoring data, wherein the parameters of the SVM decision function are dynamically updated according to a predetermined period;

[0088] Use the appropriate cleaning algorithm to clean the classified data in each category;

[0089] The cleaned data is stored according to the distributed data storage architecture, where each type of data corresponds to a database node cluster that is adapted to it;

[0090] Analyze various types of data, and the analysis dimensions include at least energy storage system performance evaluation, energy storage battery failure prediction, charging and discharging strategy optimization analysis, user electricity consumption behavior analysis, energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis.

[0091] Preferably, when the distributed energy storage digital platform performs data collection, the distributed energy storage digital platform can be connected to the battery management system (BMS) data interface, environmental monitoring sensor interface, power conversion system (PCS) data interface, power load monitoring interface, user power consumption data interface and geographic information system (GIS) data interface of the energy storage device from the hardware aspect, and a suitable adapter driver is set: for BMS data collection, initialize CAN controller parameters (such as baud rate set to 500kbps, filter setting, etc.), register the receiving interrupt function to process data frame reception, such as parsing battery voltage, current, temperature, power and other information from a specific data frame position. The environmental monitoring sensor interface sets serial port parameters (such as baud rate 9600bps, data bit 8 bits, stop bit 1 bit, no parity check), constructs a data request frame (such as function code 03 request under the Modbus protocol) and initializes the receiving buffer. The power load monitoring interface, if the smart meter communicates through Ethernet, configures network parameters and parses the specific data format of power load data (such as real-time power, active power, reactive power, etc.). The user electricity data interface develops corresponding adapters according to different data sources (such as smart meter data, electricity APP data, etc.) to analyze the user's electricity usage time, electricity consumption, power consumption and other information. The GIS data interface connects to the geographic information database or service to obtain the latitude and longitude coordinates of the energy storage device, the location of the area and other geographic location data, and converts the data format for subsequent processing. After all interfaces are initialized, the data acquisition thread is started. According to the preset acquisition frequency: obtain battery data from the BMS interface: including real-time data such as battery voltage (unit: volt, V), battery current (unit: ampere, A), battery temperature (unit: Celsius, ℃), battery power (expressed as a percentage), etc. The acquisition frequency can be set to once every 10 milliseconds. Collect data from the environmental monitoring sensor interface: such as ambient temperature (℃), ambient humidity (expressed as a percentage), light intensity (unit: lux, lx), etc. The acquisition frequency can be once every 30 seconds. Collect power load data from the power load monitoring interface: real-time power (unit: watt, W), active power, reactive power, etc. The acquisition frequency is set to once every 5 minutes according to the requirements of the power system. Obtain user electricity consumption behavior data from the user electricity consumption data interface: electricity consumption time series (accurate to minutes or seconds), electricity consumption (unit: kilowatt-hour, kWh), electricity power curve, etc. The collection frequency depends on the data source. For example, smart meter data can be collected every 15 minutes, and electricity APP data can be collected in real time or hourly. Obtain energy storage device geographic location data from the GIS data interface: energy storage device longitude, latitude and regional information (such as city, block, etc.), collected when the equipment is installed or the location information is updated. The collected raw data is immediately transmitted to the cache area of ​​the data collection end for temporary storage.Continuously monitor the amount of data in the cache area, and when the amount of data reaches a preset percentage of the cache capacity (for example, 80%), start the data transmission process. Sort the data by priority and type. For example, battery safety-related data has the highest priority, with a weight of 5; power load data is used for real-time control, with a weight of 4; user power behavior data has a weight of 3; environmental data has a weight of 2; and the geographic location data of energy storage equipment is relatively stable, with a weight of 1. Use an efficient data transmission protocol (such as an optimized TCP / IP protocol) to transmit the grouped data to the data classification module. During the transmission process, use mechanisms such as cyclic redundancy check (CRC) to ensure data reliability. If the check fails, request retransmission.

[0092] Preferably, when data classification is performed on the distributed energy storage digital platform, data feature extraction is performed first. After the data classification module receives the original data, it parses the data packet header to obtain the source identifier: for BMS data, the battery module number in the packet header determines the battery to which the data belongs; power load data can be distinguished based on the meter number or monitoring point identifier; user power consumption behavior data is classified according to the user account or meter identifier; environmental data is classified by sensor location identifier; geographic location data is associated with the device unique identifier; data format structure is analyzed: structured data (such as the database table format of power load data) directly extracts field information; semi-structured data (such as user power consumption APP data log) is converted into a structured form through regular expressions, etc.; basic data statistics are calculated: power load data calculates mean and variance, such as real-time power mean and variance; user power consumption behavior data statistics power consumption mean, power consumption peak, etc. Then, the data classification is determined, and the features are input into a classification model based on an automatic data feature recognition algorithm, using an SVM (support vector machine) decision function. For example, optionally, the data is classified into battery performance data, power load data, user power consumption behavior data, environmental impact data, equipment failure data, energy storage equipment geographic location data, etc. according to the model output, so that the acquired data of different categories can be further refined and classified after feature extraction, which is convenient for subsequent storage and analysis.

[0093] Further preferably, the SVM decision function is:

[0094]

[0095] Among them, x is the feature vector of the data to be classified, m is the number of training samples, and x i is the feature vector of the training sample, y i is the category label, α i is the Lagrange multiplier, K(x i ,x) is the kernel function, and b is the bias term. In the above, preferably, the kernel function K(x i ,x) is calculated as:

[0096]

[0097] Among them, σ is the kernel function parameter.

[0098] In some preferred embodiments, the parameters of the SVM decision function are dynamically updated according to a predetermined period, specifically including:

[0099] Start the dynamic classification rule update mechanism according to the scheduled period:

[0100] For example, the scheduled period can be weekly. Choosing weekly as the update period is a trade-off between data timeliness and computing resource consumption. In the distributed energy storage digital platform, the distribution and change patterns of data have certain regularity, but will not change dramatically and fundamentally in a very short time. A week is enough to accumulate enough new data to reflect some possible data feature change trends. For example, the user's electricity consumption behavior will experience different patterns of weekdays and weekends in a week, and the power load will also have corresponding periodic fluctuations. Through a week of data accumulation, the subtle changes in these cycles can be captured. At the same time, updating the model too frequently (such as daily updates) will consume a lot of computing resources, including data reading, model training calculations, and possible storage resource adjustments. Weekly updates can keep the model adaptable to data changes while being within an affordable resource consumption range. In addition, a scheduled task scheduler is set, which can be a scheduled task function based on the operating system (such as cron tasks in the Linux system) or a timer specially designed in the task management module of the distributed energy storage digital platform itself. When the weekly set time point is reached (for example, 2 a.m. every Sunday, when the overall system load is relatively low), the timer triggers the update mechanism start signal to notify the data processing module to start the classification rule update process.

[0101] Extract a preset proportion of various data samples from the classified data warehouse as a training set:

[0102] Preferably, for each type of data, the total amount of data of this type is first calculated according to its storage structure or index information in the data warehouse. For example, assuming that the total amount of battery data is 10,000, according to a sampling ratio of 10%, 1,000 pieces of data are randomly selected from the 10,000 pieces of data. In the random selection process, to ensure the randomness of the sampling, a pseudo-random number generator can be used in combination with the unique identifier of the data (such as the number of the data record) to determine which data records are selected. In this way, a certain proportion of samples are extracted from each type of data, and the combination constitutes the final training set, so that the training set contains information on various types of data, and the model training will not be biased towards one aspect due to too much or too little data in a certain type.

[0103] Retrain using the SVM algorithm and update the parameters and decision boundaries of the SVM decision function:

[0104] Preferably, when initializing the SVM model, for the kernel function parameters, if there is a certain empirical value or a better value range obtained through cross-validation, it can be initialized according to this range first. i and bias term b, which are usually set to default values ​​(such as α i =0, b=0), they will be adjusted according to the data samples during the training process. For the SVM model, the training process is to solve the optimization problem to determine the appropriate α i value, so that the classification interval is maximized while minimizing the penalty for misclassified samples. In this process, α is continuously adjusted according to the sample point distribution and category labels of the training data. i The value of , which in turn affects the decision function The summation term in is used to determine the position and shape of the classification hyperplane, that is, to adjust the decision boundary. At the same time, the kernel function parameter σ will also be adjusted according to the distribution of the data. If it is found that the data is still difficult to distinguish or overfitting occurs after mapping in high-dimensional space, the σ value can be appropriately reduced to make the mapped space more complex; conversely, if underfitting occurs, the σ value can be increased.

[0105] Apply the updated SVM decision function to subsequent classification operations:

[0106] After completing the model retraining and parameter adjustment, the updated model needs to be deployed to the data classification module of the distributed energy storage digital platform. This involves the transmission of model files (if the model is trained on an independent training server) or the update of the model object in memory (if the training and application are in the same process space). When switching models, ensure the continuity and correctness of the data classification operation. A strategy of running two models in parallel for a period of time can be adopted, that is, the old model and the new model classify the newly arrived data at the same time, and compare the classification results of the two. If the consistency rate of the results reaches a predetermined ratio (such as more than 95%), you can completely switch to the new model; if the inconsistency rate is high, you need to further analyze the cause. It may be that the new model has defects or the data has undergone abnormal changes during the switching process. Targeted adjustments are required, such as rechecking the training data, adjusting the model parameters, or optimizing the model structure, and then deploying and switching tests again until the new model can be stably and accurately applied to subsequent classification operations.

[0107] Preferably, the classified data categories include at least battery performance data, power load data, user power consumption behavior data, environmental data, and energy storage device geographic location data, etc. Different data categories have different characteristics, and should be cleaned using appropriate cleaning algorithms.

[0108] In some preferred embodiments, for the battery performance data, a normal distribution model is used to clean outliers and a moving average method is used to smooth the data:

[0109] Cleaning outliers based on the normal distribution model: First, calculate the battery voltage data series {V b1 ,V b2 ,…,V bn The mean of and standard deviation To calculate the mean, all the battery voltage data are added together and then divided by the total number of data points, which is To calculate the standard deviation, first calculate the sum of the squares of the difference between each data point and the mean, then divide it by the total number of data points minus 1, and finally take the square root. The formula is According to the calculated mean and standard deviation, the reasonable range is determined as Traverse each data point in the battery voltage data sequence. If a data point V bi satisfy or It is marked as an outlier and removed from the data sequence. In the process of removing outliers, the relevant information of the outliers, such as the timestamp of occurrence, battery number, etc., is recorded to facilitate subsequent analysis of the causes of the outliers, which may be caused by battery sensor failure, internal battery short circuit, etc.

[0110] Use the moving average method to smooth the data: Determine the simple moving average window size m1. The selection of this value needs to be determined based on the characteristics of the battery data and the frequency of data fluctuations. If the battery data fluctuates frequently, a smaller window size (such as 5-10) can more quickly reflect short-term changes in the data; if the data is relatively stable, a larger window size (such as 20-50) can better smooth out the noise and highlight the long-term trend of the data. For the calculation of the smoothed battery voltage data, when calculating V bsmooth (i), take out m1 consecutive data points centered on the current index i from the data sequence (when i is close to the beginning or end of the data sequence, adjust the data point selection range in the window according to the actual situation), according to the formula During the calculation process, pay attention to the order of data points and the accuracy of indexes to avoid missing data or repeated calculations.

[0111] In some preferred implementations, the power load data is processed by using data interpolation to fill in missing data and using data smoothing algorithms to remove noise:

[0112] Use data interpolation to fill missing data: When it is detected that there is a missing data point in the power load data (at time index i), first determine its two adjacent data points P load(i-1) and P load(i+1) , quickly locate the positions of these two data points through the data storage structure or index information, and follow the linear interpolation formula Calculate the filling value. During the calculation process, ensure that the difference of the time index is calculated correctly and check the rationality of the calculation result. For example, the power value after filling cannot exceed the normal power range of the power system (according to the rated power, minimum power and other parameter settings of the power system). If it exceeds the range, further analysis of the abnormal data is required. It may be caused by a failure of the data acquisition equipment or a special event in the power system. loadfill is the power load value obtained by linear interpolation to fill the missing data points, P load(i-1) is the power load value of the adjacent data point before the missing data point, P load(i+1) is the power load value of the data point adjacent to the missing data point.

[0113] Use data smoothing algorithm to remove noise: select smoothing coefficient α, which is between 0 and 1. Smaller α value (such as 0.1-0.3) will make the smoothed data more inclined to the long-term trend of historical data, slower response to new data, and better remove high-frequency noise; larger α value (such as 0.7-0.9) will pay more attention to the impact of current data. The smoothed data can reflect data changes more quickly, but some noise will be retained. When calculating the smoothed power load data, start from the first data point of the data sequence and follow the formula Perform iterative calculations, initial value During the calculation process, it is important to retain enough decimal places to avoid error accumulation due to data accuracy issues, which will affect the smoothing effect. At the same time, for the smoothed data, a comparison chart of the power load curve before and after smoothing can be drawn to visually observe the smoothing effect and adjust the α value according to actual needs. is the power load data at index i after being smoothed by the exponentially weighted moving average algorithm, P load (i) is the load value of the data point at index i in the original power load data sequence, It is the power load data at index i-1 after being smoothed by the exponentially weighted moving average algorithm.

[0114] In some preferred implementations, the user's electricity consumption behavior data includes processing electricity consumption time series data and processing electricity consumption power data:

[0115] The processing of electricity consumption time series data includes: checking the continuity of timestamps of electricity consumption time series data. First, sort the timestamps in ascending order, and then check whether the time intervals between adjacent timestamps are in line with expectations. For example, for electricity consumption data recorded in minutes, the time interval between adjacent timestamps should be 1 minute under normal circumstances. If the time interval is found to be abnormal (such as greater than 1 minute or a negative number or other unreasonable situations), it is determined that there are missing or abnormal time points.

[0116] For missing time points, the cubic spline interpolation method is used to repair them. The cubic spline interpolation function S(z) is used in each subinterval [z i ,z i+1 ](i=0,1,…,n e -1, z0 to is a cubic polynomial on the known time points before and after the missing time point:

[0117] S(z)=a zi +b zi (zz i )+c zi (zz i ) 2 +d zi (zz i ) 3

[0118] Among them, z is the timestamp, a zi , b zi 、c zi and d zi are parameter coefficients, z i is the i-th time point.

[0119] In order to determine these coefficients, the following conditions need to be met: Interpolation condition: S(z i )=q i , S(z i+1 )=q i+1 (q i and q i+1 is the corresponding time point z i and z i+1 The power consumption of zi =q i 、a zi +b zi (z i+1 -z i )+c zi (z i+1 -zi ) 2 +d zi (z i+1 -z i ) 3 =q i+1 ; First-order derivative continuity condition: S′(z i+1 )=S ′ (z i+2 ), taking the derivative of S(z) yields: S′(z)=b zi +2c zi (zz i )+3d zi (zz i ) 2 ; Second-order derivative continuity condition: S″(z i+1 )=S″(z i+2 ), taking the derivative of S′(z) yields: S″(z)=2c zi +6d zi (zz i );Arrange the above conditions into a matrix form:

[0120]

[0121] where h i =z i+1 -z i By solving this linear equation system, we can get the coefficient c of the cubic spline interpolation function. zi and d zi , combined with a zi and b zi The relevant conditions (derived from the above formula) can be used to obtain the complete cubic spline interpolation function S(z), and then the timestamp of the missing time point is substituted into the interpolation function to calculate the filled power consumption value. When constructing the coefficient matrix and solving the linear equations, numerical computing libraries (such as Python's NumPy library) should be used to improve computing efficiency and accuracy.

[0122] The processing of power consumption data includes:

[0123] Collect the historical power consumption data of users and divide the data into different subsets according to the time period (such as day, week, month). For example, for daily power consumption data, the power consumption data of each day is taken as a subset. For each subset, a cluster analysis method (such as K-Means clustering algorithm) is used to determine the normal power consumption range. First, determine the number of clusters K, which can be determined by the elbow rule (ElbowMethod), that is, draw the cluster error sum of squares curve under different cluster numbers, and select the cluster number at the inflection point of the curve as the optimal cluster number. Then randomly initialize K cluster centers, assign each data point to the cluster to which the nearest cluster center belongs, update the cluster center to the mean of the data points in each cluster, and repeat this process until the cluster center no longer changes or the maximum number of iterations is reached. Calculate the center and radius of each cluster, and treat the data beyond the cluster radius as abnormal data. In the clustering process, the data should be standardized to make data of different magnitudes comparable. For example, the Z-Score standardization method is used to convert the data into standard normal distribution data with a mean of 0 and a standard deviation of 1.

[0124] In some preferred embodiments, for environmental data, autocorrelation function and partial autocorrelation function are used to analyze the time series pattern of environmental data, and missing data are filled and data is smoothed according to the law of time series.

[0125] Take ambient temperature as an example: For the ambient temperature time series Calculate the autocorrelation function. First calculate the mean ambient temperature Then for different lag orders k (usually starting from 1 and gradually increasing to a certain proportion of the data sequence length, such as n e / 10), according to the formula Calculate the autocorrelation coefficient, where n e is the ambient temperature time series The total number of data points, T ei is the ambient temperature value at the i-th moment, is the ambient temperature time series The average value of the ambient temperature, k is the lag order. With the lag order k as the horizontal axis and the autocorrelation coefficient ACF(k) as the vertical axis, a scatter plot or a line graph is drawn. By observing the graph, the periodic characteristics of the ambient temperature data can be intuitively discovered. If the autocorrelation coefficient has an obvious peak at a certain lag order, for example, there is a peak at (corresponding to the number of hours in a day), it may indicate the existence of a daily cycle; if there is a peak at (corresponding to the number of days in a month, assuming that data is collected once a day), there may be a monthly cycle, etc.

[0126] Calculate the partial autocorrelation function (PACF), which measures the correlation between two time points while eliminating the influence of intermediate variables. A recursive algorithm, such as the Durbin-Levinson algorithm, can be used to calculate the PACF.

[0127] Specifically, the PACF is calculated using the Durbin-Levinson algorithm:

[0128] Initialization: Set This is to calculate the estimated autocorrelation coefficient at lag 1;

[0129] For k = 2, 3, ..., calculate the partial autocorrelation coefficients of different lag orders in turn:

[0130] calculate is the coefficient calculated in the previous step;

[0131] calculate

[0132] calculate This is the value of the partial autocorrelation coefficient at lag k.

[0133] renew j=1,2,…,k-1, preparing for the next round of calculation.

[0134] Similarly, a graph is drawn with lag order k as the horizontal axis and partial autocorrelation coefficient as the vertical axis. The autoregressive order of the time series is determined by observing the graph. For example, if the partial autocorrelation coefficient approaches 0 rapidly after lag order 2, then when constructing an autoregressive model, only lag 1 and lag 2 terms may need to be considered, that is, the autoregressive order is 2.

[0135] Fill missing data and smooth data according to the rules of time series: For missing data points, linear interpolation method is used to fill them. Determine the adjacent data points of the missing data points, and calculate the filling value according to the linear interpolation formula, which is similar to the interpolation method of power load data. However, in environmental data, it is necessary to take into account that the changes in environmental data are relatively slow, and the interpolation results should conform to the overall change trend of environmental data. For example, if the temperature changes of adjacent data points are relatively slow, the filling value should not fluctuate too much. The weighted moving average method is used to smooth the filled data. Determine the weighting coefficient, which is determined according to the distance of the data time. Generally, an exponential decreasing method is used, that is, the closer the data point is to the current time, the greater the weight, and the farther the data point is, the smaller the weight. According to the weighted moving average formula Calculate the smoothed data, where is the ambient temperature value at index i after smoothing by weighted moving average method, m e is the moving average window size, wj is the weighting coefficient, T ej is the ambient temperature value at the jth moment. During the calculation process, the attenuation factor and window size should be adjusted according to the characteristics of the environmental data. For example, for environmental data that changes slowly, a smaller attenuation factor and a larger window size can be selected.

[0136] In some preferred implementations, the geographical location data of the energy storage device is processed in the following manner:

[0137] Determine the original coordinate system of the acquired geographic location data, which can be determined by the data header information, data source description or specific identification field. If the data comes from different geographic information systems or sensors, there may be multiple coordinate systems, such as Beijing 54 coordinate system, Xi'an 80 coordinate system, etc. For data in different coordinate systems, use a coordinate conversion algorithm to uniformly convert them into the WGS84 coordinate system. Coordinate conversion involves complex mathematical calculations, including translation, rotation, scaling and other operations. For example, for plane coordinate conversion, a similarity transformation model or an affine transformation model can be used; for geodetic coordinate conversion, a more accurate geodetic coordinate conversion formula is required, such as the Bursa-Wolf Model or the Molodensky Model. In the conversion process, accurate earth parameters (such as the earth's major semi-axis, minor semi-axis, flattening, etc.) must be used. These parameters can be obtained from authoritative data sources such as the International Terrestrial Reference System (ITRS).

[0138] Verify the rationality of geographic location data: Check whether the longitude and latitude range is within a reasonable geographic range. The longitude range should be between -180° and 180°, and the latitude range should be between -90° and 90°. If the longitude and latitude of the data point exceed this range, it is determined to be an abnormal coordinate value. For the abnormal coordinate value, further analyze its possible causes, which may be data collection errors, interference during data transmission, or data corruption during data storage. If it is a data collection error, it may be necessary to check the settings and positioning accuracy of the data collection equipment; if it is a data transmission or storage problem, you can try to repair the data through the data verification and recovery mechanism. For abnormal coordinate values ​​that cannot be repaired, they can be marked as invalid data and related information can be recorded for subsequent analysis and processing. During the verification process, the visualization function of the geographic information system (GIS) can be combined to display the geographic location data on the map, intuitively observe the distribution of the data, and quickly discover the location and distribution characteristics of abnormal data points.

[0139] Preferably, storing the cleaned data in a distributed data storage architecture includes:

[0140] Battery performance data storage: In the battery database node cluster, create independent storage partitions according to the battery number. Each partition can store data in the form of a database table. The table structure includes fields such as battery number, timestamp, battery voltage, battery current, battery temperature, and battery power. When storing data, quickly route the data to the corresponding partition table according to the battery number. In order to improve query efficiency, you can create an index on the battery number field, so that when querying the historical data of a specific battery, you can quickly locate the partition and table where the data is located through the index, reducing the time for data search. For example, when you need to query the voltage changes of a battery with the battery number "B001" in the past month, the database engine can directly find the partition table corresponding to the "B001" battery based on the index, and then filter out the required voltage data in the table based on the timestamp range.

[0141] Power load data storage: In the power load database node cluster, partition storage is performed according to the monitoring area and time range. First, the monitoring area is divided into different geographical areas, such as cities, blocks, industrial parks, etc., and independent storage partitions are created for each area. In each partition, the storage directory or table structure is further subdivided according to the time range (such as year, month, and day). For example, under the partition of "City A", there are year directories such as "2024" and "2023", and under the "2024" directory, there are month subdirectories such as "January" and "February". Under each month subdirectory, there can be a database table storing power load data. The table structure includes fields such as timestamp, monitoring point number, real-time power, active power, and reactive power. When storing data, the data is accurately stored in the corresponding partitions and tables according to the monitoring area and time information. In order to facilitate the query of power load data in different time ranges and regions, indexes can be established on the timestamp field and the monitoring area field. For example, when you need to query the peak power load data of City A in May 2024, the database can quickly locate the "City A" partition through the monitoring area index, and then filter out the peak data in the "May 2024" table through the timestamp index.

[0142] User electricity usage behavior data storage: In the user electricity usage database node cluster, data is stored according to user accounts and time partitions. Create an independent storage directory or partition for each user account, and store data in each user partition according to time periods (such as days, weeks, and months). Data can be stored in the form of files (such as CSV files) or database tables. The file or table structure includes fields such as timestamp, electricity consumption, and electricity power. During the storage process, data is routed to the corresponding partition according to the user account, and data is stored in chronological order in each partition. In order to facilitate the analysis of the evolution of user electricity usage habits, indexes can be established on the user account field and the timestamp field. For example, when it is necessary to analyze the changes in the electricity usage behavior of user "U001" in the past year, the database can find the partition of user "U001" through the user account index, and then obtain data in different time periods through the timestamp index for analysis.

[0143] Environmental data storage: In the environmental database node cluster, data is stored by monitoring point location and time partition. First, determine the location information of the environmental monitoring point, such as the monitoring point number, geographical location (latitude and longitude), etc., and create an independent storage partition for each monitoring point. In each partition, store data according to the time range (such as year, month, day). The data can be stored in a database table, and the table structure includes fields such as timestamp, monitoring point number, ambient temperature, ambient humidity, and light intensity. When storing, the data is stored in the corresponding partition and table according to the monitoring point number and time information. In order to facilitate the query of environmental data of a specific monitoring point within a specific time range, an index can be established on the monitoring point number field and the timestamp field. For example, when it is necessary to query the ambient temperature data of the monitoring point "M001" in June 2024, the database can find the "M001" partition through the monitoring point number index, and then obtain the temperature data in June through the timestamp index.

[0144] Storage of geographical location data of energy storage equipment: In the geographic information database node cluster, the geographical location data of energy storage equipment is associated with the device number and stored. A spatial database (such as PostGIS) can be used to store geographical location data, and a table containing fields such as device number, longitude, latitude, and region is created. During the storage process, the device number is used as the primary key to ensure the uniqueness and relevance of the data. In order to facilitate geographic spatial analysis, such as querying the number of energy storage devices in a certain area, calculating the distance between energy storage devices, etc., a spatial index can be established in the spatial database, such as a spatial index based on an R-tree or a quadtree, so that when performing spatial queries, data points that meet the conditions can be quickly located, thereby improving the efficiency of geographic spatial analysis. For example, when it is necessary to query energy storage devices within a radius of 10 kilometers from a certain coordinate point (such as longitude and latitude (116°E, 39°N)), the spatial database can use the spatial index to quickly filter out the device number and location information that meet the conditions. At the same time, during the storage process, the accuracy and precision of the geographical location data must be ensured, such as using a suitable geographical coordinate data type (such as double type), and performing necessary verification and validation on the data to prevent incorrect geographical location data from being stored in the database.

[0145] Preferably, when analyzing various types of data, if the performance of the energy storage system is to be evaluated, the evaluation index system can be determined first, except for the battery capacity decay rate. (C b0 is the initial capacity of the battery, C b is the current battery capacity), charge and discharge efficiency (P out is the power of the energy storage device when it is discharged, P in Power when charging energy storage equipment), energy utilization (E out (i) is the energy released by the energy storage device during the i-th discharge, E in (i) is the energy stored in the energy storage device during the i-th charging, n k is the number of charge and discharge times of the energy storage device), self-discharge rate (SOC b1 SOC is the state of charge of the battery at the start time. b2 is the battery charge state after time t), and consider the power load matching (Among them, P load (t) is the power demand of the power load at time t, P storage (t) is the charging and discharging power of the energy storage system in the time interval [t1, t2], which is used to measure the energy storage system's ability to track and regulate power load).

[0146] Extract relevant data from the distributed database, perform data preprocessing (such as standardization and normalization), use data mining algorithms (such as association rule mining and cluster analysis) to analyze the internal relationship of indicators, and build an energy storage system performance evaluation model. For example, use the analytic hierarchy process (AHP) to determine the indicator weights, and set the weights of each indicator to be w1, w2, w3, w4, and w5 (and ), comprehensive evaluation score S = w1γ + w2η cd +w3μ+w4δ+w5ξ.

[0147] Preferably, when analyzing various types of data, if energy storage battery failure prediction is required, for battery failure data, extract features such as battery voltage change rate Abnormal temperature fluctuation characteristics (the variance or standard deviation of temperature data can be calculated), and abnormal power load fluctuation characteristics (such as the number of times the power load power change rate exceeds the threshold), etc. Use the LSTM algorithm to train the fault prediction model. The LSTM unit calculation formula is:

[0148] i t =σ(W ii x t +W hi h t-1 +b i )

[0149] f t =σ(W if x t +W hf h t-1 +b f )

[0150] g t =tanh(W ig x t +W hg h t-1 +b g )

[0151] o t =σ(W io x t +W ho h t-1 +b o )

[0152]

[0153] Among them, x t is the input feature vector at time step t, i t ,f t ,g t ,o tare the activation values ​​of the input gate, forget gate, cell state update gate and output gate respectively, σ is the sigmoid function, represents element-wise multiplication, W ii ,W hi ,W if ,W hf ,W ig ,W hg ,W io ,W ho is the weight matrix, b i ,b f ,b g ,b o is the bias vector, h t-1 and c t-1 are the hidden state and cell state of the previous time step t-1, respectively, and h t and c t are the hidden state and cell state at time step t respectively. Finally, the hidden layer output is converted into the battery failure probability P through the fully connected layer f ,The model predicts the type and time of future battery failures based on the ,characteristic patterns of historical failure data, providing a basis for ,preventive maintenance.

[0154] Preferably, when analyzing various types of data, if the charging and discharging strategy is optimized, the charging and discharging strategy optimization: the objective function takes the economic benefits of the energy storage system (maximizing the benefits of peak shaving and valley filling, reducing electricity expenses) and operational stability (avoiding excessive battery charging and discharging, maintaining power balance) as considerations, assuming that the benefits of peak shaving and valley filling are R, the electricity expenses are C, and the operational stability index is S st , objective function J = ω1R - ω2C + ω3S st , where ω1, ω2, ω3 are weight coefficients. The peak shaving and valley filling benefits are calculated by multiplying the peak-valley electricity price difference by the charge and discharge power. The electricity fee expenditure is determined according to the actual electricity price and the charge and discharge power. The operation stability index is quantified by limiting the fluctuation range of battery charge and discharge power and preventing overcharge and over-discharge. Obtain the real-time status data of the energy storage system (such as battery power, charge and discharge power limit, electricity price information (time-of-use electricity price data), load demand data (real-time power demand curve) and user electricity consumption behavior data (predicting the future trend of user electricity demand changes) from the database as state variables. Use reinforcement learning algorithms (such as deep Q network, DQN) to build a charging and discharging strategy optimization model, and the Q value function update formula is: where s t is the state at time step t (including the above state variables), a t is the action in this state (charge and discharge power adjustment amount), α q is the learning rate, r t+1 is the immediate reward (such as the increase in peak load-shaving revenue minus the increase in electricity costs), γ qBy iteratively updating the Q value function, the action that maximizes the Q value is selected as the charging and discharging power adjustment strategy according to the current state, thereby improving the economic benefits and operating efficiency of the energy storage system.

[0155] Preferably, when analyzing various types of data, if the user's electricity consumption behavior is analyzed, a user electricity consumption pattern clustering model is constructed, and a clustering algorithm (such as the K-Means algorithm) is used to cluster the user's electricity consumption behavior data into different electricity consumption patterns, such as peak electricity consumption type, valley electricity consumption type, balanced electricity consumption type, etc. The K-Means algorithm minimizes the objective function (where K w is the number of clusters, x i is the user's electricity consumption data point, μ k is the kth cluster center, n w is the number of cluster centers), divide the user's electricity consumption data into different clusters, analyze the electricity consumption characteristics and patterns of each cluster user, such as the peak period of electricity consumption and peak power consumption of peak power users, and provide a basis for power companies to formulate personalized marketing strategies or energy storage systems to optimize user-side energy storage configuration. Establish a user electricity demand prediction model and use time series analysis methods (such as ARIMA model or LSTM model in deep learning) to predict the user's future electricity demand. For the ARIMA model, by performing differential processing on the user's historical electricity consumption data, determine the autoregressive order p1, moving average order q1 and differential order d1, build a model and make predictions. Taking the LSTM model as an example, the user's historical electricity consumption data sequence is used as input, and the model is trained to learn the time dependency in the data, so as to predict the user's power or electricity consumption trend in the future, and help the energy storage system plan charging and discharging strategies in advance to better meet user needs and reduce costs.

[0156] Preferably, when analyzing various types of data, if the energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis are performed, the energy storage equipment layout optimization analysis is performed, and the coverage capacity and response efficiency of energy storage equipment in different geographical locations to the surrounding power demand are analyzed by combining power load data and user power consumption behavior data. For example, by calculating the distance between the energy storage equipment and the surrounding power load center, power transmission loss and other factors, the rationality of the energy storage equipment layout is evaluated, and the spatial analysis function of the geographic information system (GIS) is used to intuitively display the distribution of energy storage equipment in geographic space and the matching of power demand distribution, so as to provide decision support for the addition or migration of energy storage equipment. Carry out regional energy supply and demand balance analysis, divide different regions according to the geographical location data of energy storage equipment, and count the total capacity, charging and discharging capacity of energy storage equipment in each region, as well as the total power load and total power consumption of users in the region. By comparing and analyzing the energy supply and demand situation in each region, identify areas with tight or surplus energy supply, so as to formulate energy allocation strategies between regions or adjust the charging and discharging plans of energy storage equipment, and promote the balance and efficient use of regional energy. For example, in areas with tight energy supply, priority is given to discharging energy storage equipment to meet electricity demand; in areas with energy surplus, the charging capacity of energy storage equipment is increased to store excess electricity.

[0157] Preferably, the data processing method of the present invention further includes:

[0158] Obtain the data selection operation entered by the user on the visual interface;

[0159] According to the input data selection operation, data is retrieved from the corresponding database and an intuitive chart form is generated for display.

[0160] Specifically, according to the user's operation selection on the visualization interface, data is obtained from the corresponding database and generated in the form of intuitive charts. For battery performance data, such as displaying the decay curve of battery capacity over time (line graph), comparing the battery charge and discharge efficiency, etc., the health status and performance of the battery are intuitively reflected. The power load data can display the real-time power change trend over time through a line graph, or use a bar graph to compare the load peaks in different time periods (such as day, week, month), to help users understand the dynamic change law of power load. The user's power consumption behavior data uses a pie chart of the cluster analysis results to display the proportional distribution of users with different power consumption modes, or uses a line graph to present the change curve of the power consumption of a specific user over time, to assist power companies in analyzing user power consumption habits and formulating marketing strategies. Environmental data uses a line graph to display the changes in ambient temperature, humidity, etc. over time, and analyzes the potential impact of environmental factors on the energy storage system. The geographic location data of energy storage equipment marks the location of energy storage equipment on the geographic information system (GIS) map, and uses different colors or icon sizes to represent the capacity, operating status and other information of the equipment, intuitively presenting the geographical distribution of energy storage equipment and its relationship with the surrounding environment (such as power load centers and user distribution areas). In addition, it can also provide a variety of chart interaction functions, such as hovering the mouse to display data details, zooming, panning, data filtering, etc., to facilitate users to explore data in depth and discover potential problems. For example, on the battery capacity decay curve chart, hovering the mouse over a certain point can display the battery capacity, timestamp and related battery parameter information at that moment; on the geographic information system map, users can zoom the map to view the distribution details of energy storage equipment in different regions, or use the filtering function to only display energy storage equipment in a specific state (such as normal operation, failure).

[0161] Preferably, the data has an export function, and a data export button is set in the visualization interface. When the user clicks the export button, a file format selection menu (such as Excel, PDF and other common formats) pops up. According to the file format selected by the user, the currently displayed data is obtained from the visualization interface or the original data is re-obtained from the data analysis module (according to user needs). If the user chooses the Excel format, the backend uses Python's openpyxl library or other related libraries to convert the data into an Excel worksheet format, set the header, data format, etc., and then generate an Excel file and provide a download link; if the PDF format is selected, the relevant PDF generation library (such as ReportLab) is used to convert the data into a PDF document, including charts and data tables, etc., and a download link is also provided to facilitate users to conduct further data analysis and report generation.

[0162] It should be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0163] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0164] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0165] While the invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications and variations of these embodiments will be apparent to those skilled in the art in light of the foregoing description.

[0166] One or more embodiments of the present invention are intended to cover all such replacements, modifications and variations that fall within the broad scope of the present invention. Therefore, any omissions, modifications, equivalent replacements, improvements, etc. made within the spirit and principles of one or more embodiments of the present invention should be included in the protection scope of the present invention.

Claims

1. A data processing method for a distributed energy storage digital platform, characterized in that: The data processing method comprises: Collect battery data of energy storage equipment, geographic location data of energy storage equipment, power load data, user power consumption behavior data and environmental monitoring data; Based on the SVM decision function, data features are extracted and classified for the collected battery data of the energy storage device, the geographical location data of the energy storage device, the power load data, the user's power consumption behavior data and the environmental monitoring data, wherein the parameters of the SVM decision function are dynamically updated according to a predetermined period; Use the appropriate cleaning algorithm to clean the classified data in each category; The cleaned data is stored according to the distributed data storage architecture, where each type of data corresponds to a database node cluster that is adapted to it; Analyze various types of data, and the analysis dimensions include at least energy storage system performance evaluation, energy storage battery failure prediction, charging and discharging strategy optimization analysis, user electricity consumption behavior analysis, energy storage equipment layout optimization analysis and regional energy supply and demand balance analysis.

2. The data processing method of the distributed energy storage digital platform according to claim 1 is characterized in that: The SVM decision function is: Among them, x is the feature vector of the data to be classified, m is the number of training samples, and x i is the feature vector of the training sample, y i is the category label, α i is the Lagrange multiplier, K(x i ,x) is the kernel function and b is the bias term.

3. The data processing method of the distributed energy storage digital platform according to claim 2 is characterized in that: The kernel function K(x i ,x) is calculated as: Among them, σ is the kernel function parameter.

4. The data processing method of the distributed energy storage digital platform according to claim 3 is characterized in that: The parameters of the SVM decision function are dynamically updated according to a predetermined period and specifically include: Initiate a dynamic classification rule update mechanism according to a predetermined period; Extract a preset proportion of various data samples from the classified data warehouse as a training set; Retrain using the SVM algorithm to update the parameters and decision boundaries of the SVM decision function; The updated SVM decision function is applied to subsequent classification operations.

5. The data processing method of the distributed energy storage digital platform according to claim 1 is characterized in that: The classified data includes a battery performance data class, and the cleaning steps for the battery performance data class include cleaning outliers based on a normal distribution model and smoothing data using a moving average method; Cleaning outliers based on the normal distribution model includes: Calculate the battery voltage data sequence {V b1 ,V b2 ,…,V bn The mean of and standard deviation The calculated mean and standard deviation Determine the reasonable range Traverse the battery voltage data sequence {V b1 ,V b2 ,…,V bn } for each data point V bi , outliers outside the reasonable range are removed from the data sequence; Where n is the battery voltage data sequence {V b1 ,V b2 ,…,V bn The total number of data points in V bi is the battery voltage data sequence {V b1 ,V b2 ,…,V bn }The voltage value of the i-th data point in; The data smoothed by moving average include: Determine the simple moving average window size m1; The smoothed battery voltage data is calculated using the following formula: in, It is the battery voltage data at index i after being smoothed by the moving average method.

6. The data processing method of the distributed energy storage digital platform according to claim 1 is characterized in that: The classified data includes a power load data class, and the cleaning steps for the power load data class include using data interpolation to fill in missing data and using a data smoothing algorithm to remove noise; Data interpolation methods to fill missing data include: When it is detected that there are missing data points in the power load data, determine the two adjacent data points P load(i-1) and P load(i+1) , The padding value is calculated according to the following linear interpolation formula: in, is the power load value obtained by linear interpolation to fill the missing data points, P load(i-1) is the power load value of the adjacent data point before the missing data point, P load(i+1) is the power load value of the data point adjacent to the missing data point; The use of data smoothing algorithms to remove noise includes: Set the smoothing coefficient α; When calculating the smoothed power load data, starting from the first data point of the power load data sequence, iterative calculation is performed according to the following formula: in, is the power load data at index i after being smoothed by the exponentially weighted moving average algorithm, P load (i) is the load value of the data point at index i in the original power load data sequence, It is the power load data at index i-1 after being smoothed by the exponentially weighted moving average algorithm.

7. The data processing method of the distributed energy storage digital platform according to claim 1 is characterized in that: The classified data includes user electricity consumption behavior data, and the cleaning of user electricity consumption behavior data includes processing of electricity consumption time series data and processing of electricity consumption power data; The processing of electricity consumption time series data includes: Sort the timestamps in ascending order, and then check whether the time intervals between adjacent timestamps are as expected; For missing time points, the following cubic spline interpolation function S(z) is used to repair them: S(z)=a zi +b zi (z-z i )+c zi (z-z i ) 2 +d zi (z-z i ) 3 Among them, z is the timestamp, a zi , b zi 、c zi and d zi are parameter coefficients, z i is the i-th time point; The processing of power consumption data includes: For daily power consumption data, the daily power consumption data is taken as a subset; The K-Means clustering algorithm is used to determine the normal power range for each subset.

8. The data processing method of the distributed energy storage digital platform according to claim 1 is characterized in that: The classified data include environmental data, and the cleaning of environmental data includes analyzing the time series pattern of environmental data using autocorrelation function and partial autocorrelation function, and filling in missing data and smoothing data according to the law of time series; The use of autocorrelation functions and partial autocorrelation functions to analyze environmental data time series patterns includes: Calculate the mean value of the ambient temperature, calculate the autocorrelation coefficient and draw an autocorrelation function graph, wherein the calculation formula of the autocorrelation coefficient ACF(k) is: Among them, n e is the ambient temperature time series The total number of data points, T ei is the ambient temperature value at the i-th moment, is the ambient temperature time series The average value of the ambient temperature, k is the lag order; Use the Durbin-Levinson algorithm to calculate the partial autocorrelation function and draw the partial autocorrelation function graph; Filling missing data and smoothing data according to time series rules include: For missing data points, linear interpolation method was used to fill them; The smoothed data is calculated using the following weighted moving average formula: in, is the ambient temperature value at index i after smoothing by weighted moving average method, m e is the moving average window size, w j is the weighting coefficient, T ej is the ambient temperature value at the jth moment.

9. The data processing method of the distributed energy storage digital platform according to claim 1, characterized in that: The classified data includes the energy storage device location data class. The cleaning of the energy storage device location data class includes: Determine the original coordinate system of the acquired geographic location data of the energy storage device; For data in different original coordinate systems, use the coordinate conversion algorithm to uniformly convert them into the WGS84 coordinate system; Check whether the latitude and longitude values ​​are within a reasonable geographical range; Repair data for abnormal coordinate values ​​that are not within a reasonable geographical range; Coordinate values ​​that cannot be repaired are marked as invalid data and discarded.

10. The data processing method of the distributed energy storage digital platform according to any one of claims 1 to 9, characterized in that: The data processing method further includes: Obtain the data selection operation entered by the user on the visual interface; According to the input data selection operation, data is retrieved from the corresponding database and an intuitive chart form is generated for display.

Images