BERT-based new energy storage industry data semantic matching method and equipment
By using a BERT-based semantic matching model, the problems of low accuracy and high maintenance cost in data model matching for energy storage sites are solved, achieving efficient and secure data matching, which is suitable for offline data analysis in the energy storage industry.
Patent Information
- Application Number
- CN202511670356.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-01-30
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Data model matching for energy storage sites suffers from high maintenance costs and low matching accuracy. In particular, due to differences in Chinese description rules among different sites and the limitations of regular expressions, it is impossible to effectively identify keywords with different word orders.
By adopting a BERT-based semantic matching model, a Chinese semantic dataset of data point codes in the energy storage industry is constructed and trained using the BERT model to achieve intelligent matching of offline operating data, thereby reducing maintenance costs and improving matching accuracy.
It significantly improves the matching accuracy of data models, reduces maintenance costs, and meets the data security requirements of the energy storage industry, while also having the ability to iteratively optimize for new equipment and data indicators.
Smart Images
Figure CN121435985A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the intersection of the energy storage industry and natural language processing, and in particular to a semantic matching method and device for new energy storage industry data based on BERT. Background Technology
[0002] The operational data of new energy storage stations includes critical information such as battery status, voltage and current, charging and discharging power, switch status, AC and DC power, and environmental and fire protection data. While energy storage stations are equipped with complete data acquisition systems and analysis platforms, these platforms cannot meet the needs of certain specialized, large-scale data analyses. Therefore, it is necessary to utilize off-site professional cloud analysis platforms to analyze the operational data. Furthermore, cloud analysis platforms can add additional scientific research value, digital asset value, and certain financial attributes to the station's operational data. Because the operational data of energy storage stations involves confidential issues such as data security and production safety, the station networks often use encrypted local area networks, isolated from external networks to ensure data and operational security. Direct data transmission with public cloud analysis platforms is not permitted; data transmission and analysis can only be achieved through an "offline export-local import" method.
[0003] However, when maintenance personnel at different energy storage sites maintain and collect data point codes, there are significant differences in the Chinese descriptions and rules for the same indicators. For example, for the "operating power" of a battery cluster, the keywords maintained by the site include "current operating power," "power," and "operating power." For the "total voltage" of a battery cluster, the keywords include "total voltage," "BUS voltage," "BUS total voltage," and "battery total voltage." There are also overlapping keywords for different indicators, such as "maximum module voltage," "maximum module voltage hub number," "maximum module voltage network card parallel sequence number," and "maximum module voltage network card series sequence number." In some cases, regular expression matching is commonly used to solve the problem of recognizing the meaning of the Chinese characters in the data point codes. Developers write a large number of customized regular expressions and matching templates for the Chinese description formats of different sites. As shown in the keyword example above, this method has obvious shortcomings. On the one hand, there are numerous energy storage sites, and the rules for the Chinese descriptions of the sites are flexible and varied, requiring customized templates to be maintained for each site, and the regular expressions need to be iteratively maintained. On the other hand, regular expressions can only recognize text with fixed formats based on character matching rules, which can lead to recognition failures and misinterpretations. For descriptions with the same meaning but different word order, such as "cell voltage" and "cell battery voltage," as well as keywords with repetitive descriptions, such as "maximum module voltage" and "maximum module voltage hub number," effective matching cannot be achieved. This results in the data model matching accuracy generally being lower than expected, making it difficult to meet the precision requirements of data analysis in the energy storage industry.
[0004] Therefore, there is an urgent need for a data model matching method that can reduce maintenance costs, improve semantic understanding capabilities, and increase matching accuracy in order to address the inherent shortcomings of existing regular expression matching technologies. Summary of the Invention
[0005] The purpose of this application is to provide a BERT-based data semantic matching method and device for the new energy storage industry, which can meet the offline data security processing requirements of the energy storage industry, reduce the maintenance cost of data semantic matching, and improve the accuracy of semantic matching.
[0006] To achieve the above objectives, this application provides the following solution.
[0007] Firstly, this application provides a BERT-based semantic matching method for new energy storage industry data. The method includes: acquiring historical offline operation data of different energy storage stations, and constructing a Chinese semantic dataset of data point codes for the energy storage industry using the historical offline operation data; the historical offline operation data consists of operation status parameter data recorded by the energy storage station during operation, which contains different header row names; constructing a semantic matching model for energy storage industry data based on the BERT model, and training the semantic matching model using the Chinese semantic dataset of data point codes to obtain a trained semantic matching model; reading the current offline operation data of the target energy storage station, and determining whether the header row name is correctly identified; if so, reading the data in the column where the header row name is located; if not, matching the unidentifiable header row name using the trained semantic matching model; determining whether the matching probability value is greater than a preset probability; if so, reading the data in the column where the matched header row name is located; if not, outputting a recognition failure signal; the recognition failure signal is used to indicate that there is no data in the current offline operation data of the target energy storage station that matches the trained semantic matching model.
[0008] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the BERT-based data semantic matching method for the new energy storage industry described above.
[0009] Based on the specific embodiments provided in this application, the following technical effects are disclosed.
[0010] This application constructs a Chinese semantic dataset for data point codes specific to the energy storage industry by collecting historical offline operation data from different energy storage sites. Based on this dataset, a semantic matching model is built and trained using a BERT pre-trained model, enabling it to deeply understand energy storage industry terminology. During the application of the semantic matching model, after reading the offline operation data of the target site, the model first attempts to identify the title line using conventional methods; if this fails, the trained BERT model is invoked for semantic matching, and intelligent mapping is achieved by calculating the matching probability. This application effectively solves the problems of word order differences and format variations that traditional regular expressions cannot handle by leveraging BERT's semantic understanding capabilities, significantly improving matching accuracy. Furthermore, due to the use of a unified semantic template and offline processing mode, there is no need to maintain matching rules separately for each site, nor is it necessary to upload sensitive data to the public network, greatly reducing maintenance costs and meeting industry data security requirements. In addition, the model designed in this application supports continuous iterative optimization, can adapt to the access of new equipment and new data indicators, and has the potential to be extended to offline data analysis scenarios in other industries. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1 A flowchart illustrating the BERT-based semantic matching method for new energy storage industry data provided in this application embodiment. Figure 1 .
[0013] Figure 2 A flowchart illustrating the BERT-based semantic matching method for new energy storage industry data provided in this application embodiment. Figure 2 .
[0014] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0016] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0017] Example 1, such as Figures 1-2 As shown in the figure, this embodiment provides a semantic matching method for new energy storage industry data based on BERT, and the method includes the following steps.
[0018] S1. Obtain historical offline operation data of different energy storage sites, and use the historical offline operation data to construct a Chinese semantic dataset of data point codes for the energy storage industry; the historical offline operation data is the operation status parameter data recorded by the energy storage sites during operation, which has different header row names.
[0019] Step S1 specifically includes the following steps.
[0020] S11. Collect historical offline operation data of different energy storage sites.
[0021] S12. Extract the Chinese description text corresponding to all data point codes in the historical offline running data, and use it as the original text set.
[0022] S13. Perform data cleaning on the original text set to obtain the cleaned text set.
[0023] Furthermore, the data cleaning includes: removing duplicates, correcting typos, standardizing formats, correcting simplified and traditional Chinese characters, and semantic normalization.
[0024] Furthermore, the semantic normalization process includes: removing characters irrelevant to the energy storage industry, removing stop words in the energy storage industry, and standardizing the word order of energy storage industry-specific terms.
[0025] S14. In the cleaned text set, label the standard data model field labels corresponding to each Chinese descriptive text to form a labeled dataset, and output the labeled dataset as the dot code Chinese semantic dataset.
[0026] Furthermore, the standard data model field labels are determined based on the preset model fields of the energy storage industry data analysis platform, and each field label corresponds to a unique field.
[0027] S2. Construct a semantic matching model for energy storage industry data based on the BERT model, and train the semantic matching model using the Chinese semantic dataset of data point codes to obtain the trained semantic matching model.
[0028] Furthermore, step S2 specifically includes the following steps.
[0029] S21. Divide the Chinese semantic dataset of data points into a training set, a validation set, and a test set according to a preset ratio.
[0030] Optionally, the preset ratio is 7:2:1.
[0031] S22. Use the Python development language transformers library to build a semantic matching model for energy storage industry data based on the BERT model.
[0032] Optionally, the BERT model is a BERT-Base-Chinese pre-trained model.
[0033] Optionally, the BERT-Base-Chinese pre-trained model has been pre-trained on a Chinese corpus and has basic Chinese semantic understanding capabilities.
[0034] S23. Use the AdamW optimizer to set the model training parameters, use the standard data model field labels as labels, and input the training set into the semantic matching model for training.
[0035] S24. Use the validation set to select the optimal parameters of the lightweight model, and use the model corresponding to the minimum loss function as the trained semantic matching model.
[0036] S25. Use the test set to evaluate the performance of the trained semantic matching model. If the accuracy of the validation set does not improve for n consecutive rounds, stop training and save the trained semantic matching model; n is greater than or equal to 1.
[0037] Furthermore, the training parameters include: learning rate, batch size, number of training epochs, weight decay coefficient, and maximum input length.
[0038] Furthermore, n=3.
[0039] S3. Read the current offline operation data of the target energy storage station and determine whether the title row name is correctly recognized. If so, read the data in the column where the title row name is located. If not, use the trained semantic matching model to match the unrecognizable title row name.
[0040] S4. Determine whether the matching probability value is greater than the preset probability. If yes, read the data of the matched title row name sitting in the column. If no, output the recognition failure signal. The recognition failure signal is used to indicate that there is no data in the current offline operation data of the target energy storage station that matches the trained semantic matching model.
[0041] Furthermore, the preset probability is 0.75.
[0042] In practical applications, the specific process of step S4 is as follows.
[0043] 1) Maintain a unified data model template on the analysis platform, which includes Chinese semantic descriptions in the training model, data model encoding on the analysis platform, and other data model-related information such as data type and data precision.
[0044] 2) Site maintenance personnel export offline data files with Chinese titles, usually in plain Excel or CSV format.
[0045] 3) First, use a general method to identify the data title row that can accurately match the Chinese tags of the analysis platform's data model. If the identification fails, use the semantic matching model. The model outputs the probability value of the data title to be matched corresponding to the Chinese tags of the analysis platform's data model. Select the optimal matching title column with the highest probability value and a probability value greater than 0.75 to complete the data model matching.
[0046] This application also provides an implementation method, as detailed below.
[0047] 1) Construct a Chinese semantic dataset of data point codes for the energy storage industry.
[0048] 1.1 As shown in the table, collect offline data from energy storage stations. The data file for this example is a CSV file, as shown in the table below. Extract the Chinese description text from the data header row.
[0049] Table 1 Offline Data Table for Energy Storage Sites
[0050] 1.2 The Chinese descriptions in the offline data title rows are cleaned and semantically normalized, as follows.
[0051] Operating power: [Current operating power, power, operating power, current power].
[0052] BUS Total Voltage: [BUS Total Voltage, BUS Voltage].
[0053] Total battery voltage: [Total battery voltage, total battery cluster voltage, total battery stack voltage].
[0054] Maximum module voltage: [Maximum module voltage].
[0055] 1.3 As shown in Table 2, the normalized semantic text set is organized into a CSV format training set, as follows.
[0056] Table 2 Training Set in CSV Format
[0057] Step 2: Construct a semantic matching model for energy storage industry data based on the BERT model.
[0058] 2.1 Load the BERT base model using the Python-transformers library.
[0059] 2.2 Set training parameters TrainingArguments, using the AdamW optimizer, with the learning rate (learning_rate) set to... The batch size (per_device_train_batch_size / per_device_eval_batch_size) is set to 32, the number of training epochs (num_train_epochs) is set to 10, the weight decay coefficient (weight_decay) is set to 0.01, and the maximum input length (max_seq_length) is set to 64.
[0060] 2.3 If there is no improvement after 3 consecutive rounds of model training, stop training and save the optimal model.
[0061] Step 3. Real-time matching and application of offline data with the data model of the analysis platform.
[0062] 3.1 As shown in Table 3, maintain a set of data model templates related to the semantics of the above-mentioned energy storage industry keywords in the analysis platform database.
[0063] Table 3. Data Model Templates Related to Keyword Semantics in the Energy Storage Industry
[0064] 3.2 Real-time Offline Data Matching to the Data Model: Read the offline data CSV. As shown in step 1, first directly match data columns with the same header. If a match is successful, save it directly as data that matches the data model of the analysis platform. Otherwise, load the BERT-based training model from step 2 for matching, selecting the optimal matching data column with the highest matching probability value (greater than 0.75). If the identification is successful, save it as data that matches the data model of the analysis platform; if the identification fails, it is assumed that there is no data in the file that matches the data model of the analysis platform.
[0065] The technical effects of this application are as follows.
[0066] 1) Significantly reduced maintenance costs: This invention uses a BERT pre-trained model instead of traditional regular expressions. Only one set of Chinese semantic templates related to fine-tuning the model needs to be maintained on the analysis platform. There is no need to maintain the data model template for each site separately, which greatly reduces maintenance costs.
[0067] 2) Improve the accuracy of data model matching: The BERT model is based on semantic understanding and can effectively identify descriptions with the same semantics but different word order and format, which is far superior to traditional regular expressions.
[0068] 3) Applicable to the data security needs of the energy storage industry: This invention is designed based on offline data import scenarios. All model training and data matching processes are completed on the analysis platform, eliminating the need to directly transmit the energy storage site operation data to the public network, which meets the data security requirements of the energy storage industry.
[0069] 4) Possesses sustainable optimization capabilities: As energy storage stations continue to innovate and connect to new equipment, they will also continuously connect to new data indicators, which can be used to iteratively optimize and train the model. At the same time, Chinese labels can be added to the analysis platform iteratively.
[0070] 5) Scalability: This method can be extended to offline data analysis scenarios in other industries. It only requires training the BERT model with Chinese labels from other industries.
[0071] Example 2: This application also provides a computer device, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores and processes data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the methods described above.
[0072] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0073] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0074] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0075] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A BERT-based data semantic matching method for the new energy storage industry, characterized in that, The method comprises: acquiring historical offline operation data of different energy storage stations, and constructing a data point code Chinese semantic data set of the energy storage industry by using the historical offline operation data; the historical offline operation data are operation state parameter data recorded by the energy storage station in the operation process and exist in different title line names; constructing a semantic matching model of the energy storage industry data based on a BERT model, and training the semantic matching model by using the data point code Chinese semantic data set to obtain a trained semantic matching model; reading current offline operation data of a target energy storage station, and judging whether the title line name is normally recognized; if yes, reading data in the column where the title line name is located; if no, matching the title line name that cannot be recognized by using the trained semantic matching model; judging whether the matching probability value is greater than a preset probability; if yes, reading data in the column where the matched title line name is located; if no, outputting a recognition failure signal; the recognition failure signal is used to represent that there is no data matched with the trained semantic matching model in the current offline operation data of the target energy storage station.
2. The BERT-based new energy storage industry data semantic matching method according to claim 1, characterized in that, acquiring historical offline operation data of different energy storage stations, and constructing a data point code Chinese semantic data set of the energy storage industry by using the historical offline operation data, specifically comprising: collecting historical offline operation data of different energy storage stations; extracting Chinese description texts corresponding to all data point codes in the historical offline operation data as an original text set; performing data cleaning on the original text set to obtain a cleaned text set; in the cleaned text set, labeling a corresponding standard data model field label for each Chinese description text to form a labeled data set, and outputting the labeled data set as a point code Chinese semantic data set.
3. The BERT-based new energy storage industry data semantic matching method according to claim 1, characterized in that, The data cleaning comprises deleting duplicates, correcting wrong characters, unifying formats, correcting simplified and traditional Chinese, and performing semantic normalization processing.
4. The BERT-based new energy storage industry data semantic matching method according to claim 3, characterized in that, The semantic normalization processing comprises removing energy storage industry irrelevant characters, removing energy storage industry stop words, and standardizing energy storage industry special word sequences.
5. The BERT-based new energy storage industry data semantic matching method according to claim 2, characterized in that, The standard data model field label is determined based on a preset model field of an energy storage industry data analysis platform, and each field label corresponds to a unique field.
6. The BERT-based new energy storage industry data semantic matching method according to claim 2, characterized in that, constructing a semantic matching model of the energy storage industry data based on a BERT model, and training the semantic matching model by using the data point code Chinese semantic data set to obtain a trained semantic matching model, specifically comprising: dividing the data point code Chinese semantic data set into a training set, a validation set and a test set according to a preset proportion; constructing a semantic matching model of the energy storage industry data based on the BERT model by using a Python development language transformers library; setting model training parameters by using an AdamW optimizer, taking the standard data model field label as a label, and inputting the training set into the semantic matching model for training; using the validation set to select optimal parameters of a lightweight model, and taking a model corresponding to a minimum value of a loss function as the trained semantic matching model; evaluating the performance of the trained semantic matching model by using the test set, stopping training when the accuracy of the validation set does not improve for n consecutive rounds, and saving the trained semantic matching model; n is greater than or equal to 1.
7. The BERT-based new energy storage industry data semantic matching method according to claim 6, characterized in that, The training parameters include: a learning rate, a batch size, a training round, a weight decay coefficient, and a maximum input length.
8. The BERT-based new energy storage industry data semantic matching method according to claim 6, characterized in that, n=3。 9. The BERT-based new energy storage industry data semantic matching method according to claim 1, characterized in that, The preset probability is 0.
75.
10. A computer device comprising: Memory, processor and computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the BERT-based new energy storage industry data semantic matching method of any one of claims 1-9.