Multi-source data processing and intelligent question-answering system based on power transmission and distribution

Abstract

The invention discloses a multi-source data processing and intelligent question-answering system based on power transmission and distribution, and relates to the technical field of power transmission and distribution data processing, and the system comprises the steps: collecting multi-source heterogeneous data of a power transmission and distribution production domain, and carrying out the standardization processing and fusion of the multi-source heterogeneous data; constructing a power transmission and distribution production field large model, and performing fine tuning and enhancement processing on the power transmission and distribution production field large model; constructing a standard evaluation system to perform standard evaluation on the power transmission and distribution production field large model to obtain an evaluated power transmission and distribution production field large model; obtaining user voice data and text data, converting the user voice data into a text, analyzing the text and the text data, and generating a user instruction; and calling the evaluated power transmission and distribution production field large model according to a user instruction, performing reasoning and analysis, generating a corresponding answer and a scheme, and feeding back a result to a user. According to the method, complex multi-source heterogeneous data in a power transmission and distribution production command scene can be processed, and the power transmission and distribution production domain professional knowledge question and answer ability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transmission and distribution data processing, in particular to a multi-source data processing and intelligent question-answering system based on power transmission and distribution. BACKGROUND

[0002] Power transmission and distribution is the "intermediate hub" of the power system, connecting the power generation side (power plant) and the user side (enterprise, resident), and bearing the core functions of power transmission, voltage transformation and load distribution. Its operation state directly affects the power supply reliability, power quality and power grid safety. It covers multiple data sources such as power equipment (transformer, line, switch, etc.), sensors, monitoring systems, etc. With the advancement of the "double carbon" goal and the construction of the new power system, the power transmission and distribution network is developing towards high reliability and low operation and maintenance cost. The traditional manual decision-making mode is difficult to meet the real-time analysis needs of massive data.

[0003] In a Chinese patent application with the application publication number CN118297228A, an optimization method and system for a power transmission and distribution conductor monitoring system are disclosed. According to a preset optimization objective function, the power transmission and distribution conductor system is sequentially subjected to zoned and graded layout optimization and backbone local layout optimization. The sensitivity coefficients corresponding to each grating are iteratively optimized through demodulation residuals until the sensitivity coefficients corresponding to each grating meet the preset requirements. Multi-scale correlation feature extraction, frequent item set and association rule mining are sequentially performed on the multi-source monitoring data output by the demodulation model to obtain a multi-dimensional spatio-temporal correlation fusion model. According to the preset knowledge optimization algorithm and the multi-dimensional spatio-temporal correlation fusion model, the power transmission and distribution conductor monitoring system is iteratively optimized to obtain an optimized power transmission and distribution conductor monitoring system. The problem of poor reliability of the power transmission and distribution conductor monitoring system due to single grating layout, large grating demodulation error and insufficient analysis of collected data in the prior art is solved. Through multi-directional adjustment and optimization strategies, the reliability of the power transmission and distribution conductor detection system is improved.

[0004] The existing technology has the following problems: the data generated in the field of power transmission and distribution production comes from various sources, including text (maintenance report), image (patrol video), table (device monitoring parameter), sensor data (monitoring current and voltage), etc. These data have different data source formats and large structural differences, and it is difficult to efficiently process massive, high-speed, multi-source heterogeneous data, especially unstructured text and image data. Moreover, there is a lack of unified standards to quantitatively evaluate the application capabilities of large models in the field of power transmission and distribution. Therefore, there is an urgent need for a system that can process and intelligently answer multi-source data of power transmission and distribution to meet the application requirements of high safety and high reliability of the power transmission and distribution system. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide a multi-source data processing and intelligent question-answering system based on power transmission and distribution, which comprises the following modules: data acquisition module one, multi-source data fusion module, large model construction module, model evaluation module, model evaluation module, data acquisition module two, interactive answering module.

[0006] The above-mentioned various modules are connected through wired and / or wireless connection to realize data transmission between the modules.

[0007] The data acquisition module one acquires multi-source heterogeneous data in the power transmission and distribution production domain and performs standardization processing on the multi-source heterogeneous data.

[0008] The multi-source data fusion module fuses the multi-source heterogeneous data after standardization processing.

[0009] The large model construction module constructs a large model in the power transmission and distribution production field, and performs fine-tuning and enhancement processing on the large model in the power transmission and distribution production field.

[0010] The model evaluation module constructs a standard evaluation system to perform standard evaluation on the large model in the power transmission and distribution production field, and obtains the evaluated large model in the power transmission and distribution production field.

[0011] The data acquisition module two acquires user voice data and text data, converts the user voice data into text, and parses the text and text data to generate user instructions.

[0012] The interactive answering module calls the evaluated large model in the power transmission and distribution production field according to the user instructions, performs reasoning and analysis, generates corresponding answers or solutions, and feeds back the results to the user.

[0013] In a preferred embodiment, the process of standardization processing on multi-source heterogeneous data is as follows:

[0014] The multi-source heterogeneous data is subjected to data cleaning preprocessing, the multi-source heterogeneous data after data cleaning preprocessing is uniformly converted into a specific time format; discrete data is converted into a string form and subjected to matrix processing; numerical data is unified in precision, unit and numerical range; text data is subjected to word segmentation, part-of-speech tagging and named entity recognition processing, key information is extracted, and text feature vectors are converted; image data is subjected to image preprocessing to enhance image features and utilize target detection algorithms to identify key information in the image data to form structured image description data; and finally, the multi-source heterogeneous data after standardization processing is obtained.

[0015] In a preferred embodiment, the process of fusing multi-source heterogeneous data is as follows:

[0016] The multi-source heterogeneous data is converted into a feature vector and input into a multi-modal Transformer model, the correlation between the feature vectors is learned through multiple Transformer encoder layers and modal specific encoder layers, the feature vectors are aligned in the same feature space through contrast learning, and the feature vectors are adaptively fused through a gated neural network to obtain fused multi-source heterogeneous data.

[0017] In a preferred embodiment, the process of building a power transmission and distribution production field large model includes:

[0018] The power transmission and distribution production field knowledge graph and the fused multi-source heterogeneous data are obtained to build a power transmission and distribution production field large model.

[0019] In a preferred embodiment, the process of fine-tuning and enhancing the power transmission and distribution production field large model includes:

[0020] The low-order adaptive fine-tuning technology (LoRA) and the reinforcement learning technology with human feedback are used to fine-tune and enhance the power transmission and distribution production field large model.

[0021] In a preferred embodiment, the standard evaluation system includes the BLEU index, the ROUNGE index, the precision, the recall, and the F1 value.

[0022] Compared with the prior art, the application has the beneficial effects that: the application collects multi-source heterogeneous data in the power transmission and distribution production field, standardizes the multi-source heterogeneous data, fuses the standardized multi-source heterogeneous data, builds a power transmission and distribution production field large model, fine-tunes and enhances the power transmission and distribution production field large model, builds a standard evaluation system to evaluate the power transmission and distribution production field large model, obtains user voice data and text data, converts the user voice data into text, analyzes the text and the text data, generates user instructions, calls the evaluated power transmission and distribution production field large model according to the user instructions, performs reasoning and analysis, generates corresponding answers or solutions, and feeds back the results to the user. On the one hand, the application standardizes and fuses the multi-source heterogeneous data to solve the problems of diverse data sources, different data sources, large structural differences, and inefficient processing of massive, high-speed, multi-source heterogeneous data in the power transmission and distribution production field. On the other hand, the low-order adaptive fine-tuning technology (LoRA) and the reinforcement learning technology with human feedback are used to fine-tune and enhance the power transmission and distribution production field large model, a standard evaluation system is established, the evaluated power transmission and distribution production field large model is called according to user instructions, reasoning and analysis are performed, corresponding answers or solutions are generated, and the results are fed back to the user. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 Figure 1 shows a schematic diagram of a power transmission and distribution based multi-source data processing and intelligent question answering system according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is merely meant to provide a better understanding of the subject matter described herein and can include changes, modifications, additions or omissions of the functions and arrangements of the elements discussed without departing from the scope of the present description. Various examples can omit, substitute, or add various procedures or components as appropriate, and the examples can not necessarily be performed in the order described. Also, the various features of some examples can be combined with features of other examples.

[0025] It should be noted that, unless otherwise defined, technical and scientific terms used in one or more embodiments of the present application have the meanings commonly understood by one of ordinary skill in the art in the field of the present application. The words "first", "second", and similar words do not necessarily have an ordinal or chronological significance, but are used only to distinguish one element from another. The words "include", "contain", and similar words do not exclude other elements not explicitly listed. The words "connected" or "coupled" do not necessarily mean direct connection or coupling, but can include indirect connection or coupling through one or more other elements. The words "above", "below", "left", "right", and similar words are used only to indicate relative positions, and can change when the absolute positions of the described objects change.

[0026] Referring to Figure 1, a power transmission and distribution based multi-source data processing and intelligent question answering system is shown. The system includes the following modules: data acquisition module one, multi-source data fusion module, large model construction module, model evaluation module, data acquisition module two, interactive answering module. Figure 1 The above modules are connected through wired and / or wireless connection to realize data transmission between the modules.

[0027] The above modules are connected through wired and / or wireless connection to realize data transmission between the modules.

[0028] Data acquisition module one: responsible for collecting multi-source heterogeneous data in the power transmission and distribution production domain, and performing standardization processing on the multi-source heterogeneous data.

[0029] On the basis of the above embodiment, the process of standardizing the multi-source heterogeneous data is: the multi-source heterogeneous data is preprocessed by data cleaning, the time type data after the data cleaning preprocessing is uniformly converted into a specific time format; the discrete type data is converted into a string form and matrix processing; the numerical type data is unified in precision, unit and numerical range; the text type data is processed by word segmentation, part-of-speech tagging and named entity recognition, key information is extracted, and a text feature vector is converted; the image data is preprocessed to enhance the image features and the key information in the image data is identified by using a target detection algorithm to form structured image description data; and finally the multi-source heterogeneous data after standardization is obtained.

[0030] It should be understood that the multi-source heterogeneous data includes but is not limited to: device running state data, device operation and maintenance data, device inspection image, device monitoring parameter data, device type data, sensor data, etc.; most of these data exist in various forms of structured, unstructured and semi-structured;

[0031] The key information in the image data includes: device components, defects and other target objects, and the class, position and size of the target objects are labeled to form structured image description data; the data cleaning preprocessing includes processing missing values, removing duplicate data, correcting error data and data normalization processing.

[0032] It should be noted that the multi-source heterogeneous data is obtained by a PLC (programmable logic controller), an EAM (enterprise asset management) system, an inspection robot equipped with a high-definition camera and a sensor network.

[0033] Multi-source data fusion module: the multi-source heterogeneous data after standardization is fused;

[0034] On the basis of the above embodiment, the process of fusing the multi-source heterogeneous data is: the multi-source heterogeneous data is converted into a feature vector and input into a multi-modal Transformer model, the correlation between the feature vectors is learned through multiple Transformer encoder layers and modal specific encoder layers, the feature vectors are aligned in the same feature space by using contrast learning, the feature vectors are adaptively fused by a gated neural network, and the fused multi-source heterogeneous data is obtained.

[0035] It should be noted that each Transformer encoder layer contains a self-attention mechanism and a feedforward neural network; the self-attention mechanism can simulate long-distance dependency relationships, allowing feature vectors to interact and influence each other, so that the multi-modal Transformer model can understand the relationship between feature vectors; the modal-specific encoder layer is used for features unique to each feature vector; contrastive learning maximizes the similarity between feature vectors of positive sample pairs (such as text descriptions and corresponding images), while minimizing the similarity between feature vectors of negative sample pairs, so that text and image feature vectors can be as close as possible to the corresponding semantic space, achieving feature alignment; the gating neural network can control the flow of information according to the feature patterns of the input feature vectors, thereby achieving more effective feature fusion.

[0036] A large model construction module is configured to construct a large model in the power transmission and distribution field, fine-tune and enhance the large model in the power transmission and distribution field.

[0037] Based on the above embodiment, the process of constructing the large model in the power transmission and distribution field includes: obtaining a knowledge graph in the power transmission and distribution field and a fused multi-source heterogeneous data to construct the large model in the power transmission and distribution field.

[0038] The process of fine-tuning and enhancing the large model in the power transmission and distribution field includes: using a low-order adaptive fine-tuning technology (LoRA) and a reinforcement learning technology with human feedback to fine-tune and enhance the large model in the power transmission and distribution field.

[0039] It should be noted that the knowledge graph in the power transmission and distribution field is a knowledge representation method that organizes and manages various concepts, entities, events and their relationships in the power transmission and distribution field in the form of a graph. This knowledge graph integrates information from multiple data sources, including sensor data, device parameters, maintenance records, technical standards and procedural specifications, to form a comprehensive and accurate knowledge base.

[0040] The low-order adaptive fine-tuning technology (LoRA) freezes the weight matrix of the large model in the power transmission and distribution field, and injects a trainable low-rank decomposition matrix into each layer of the large model in the power transmission and distribution field, each layer containing a matrix multiplication. The weight matrix in these layers usually has full rank. When adapting to the power transmission and distribution domain task, the large model in the power transmission and distribution field has a low "intrinsic dimensionality", which effectively learns by randomly projecting the weight matrix and the low-rank decomposition matrix to a smaller subspace. When fine-tuning the large model in the power transmission and distribution field using LoRA, the number of new parameters is relatively small, so the cost of the fine-tuning process can be significantly reduced.

[0041] The human feedback reinforced learning technology (Factually Augmented RLHF) enhances the reward of the power transmission and distribution production field large model by adding additional factual information (such as power equipment images, related text descriptions, and real equipment category information), thereby alleviating the reward manipulation phenomenon in RLHF and further improving the performance of the power transmission and distribution production field large model.

[0042] The model evaluation module: a standard evaluation system is constructed to evaluate the power transmission and distribution production field large model, and an evaluated power transmission and distribution production field large model is obtained.

[0043] On the basis of the above embodiment, the standard evaluation system includes BLEU index, ROUNGE index, precision, recall, and F1 value.

[0044] It should be noted that the power production field generation covers sub-business domains of power transformation, power transmission, and power distribution, and each sub-business domain has no less than 1000 high-quality and labeled question and answer pairs as reasoning data sets.

[0045] The power transmission and distribution production field large model is evaluated, including power transmission and distribution language understanding ability, power transmission and distribution image understanding ability, power transmission and distribution multi-modal data understanding ability, power transmission and distribution reasoning ability, mathematical ability, code generation ability, API service calling ability, safety, response speed, etc. The standard evaluation system is used as the evaluation standard of the power transmission and distribution production field large model.

[0046] The accuracy, completeness, context consistency, relevance, and coherence of the question generation result can be verified by artificial and expert evaluation.

[0047] The data acquisition module two: acquires user voice data and text data, converts the user voice data into text, and parses the text and text data to generate user instructions.

[0048] On the basis of the above embodiment, the process of generating user instructions is as follows: user voice data is acquired through a microphone device and converted into an audio signal, the audio signal is converted into acoustic features, and the acoustic features are input into a speech recognition model to be mapped into text; the text and text data are parsed using natural language processing technology to generate user instructions.

[0049] The user instructions include fault query, device state query, device parameter setting, inspection, maintenance, dispatching, etc.

[0050] It should be noted that the speech recognition model includes an end-to-end speech recognition model with LSTM or a speech recognition model based on a deep neural network (DNN) and the like; for example, the user's voice or text input: "What was the load of the substation yesterday?"; after parsing and extracting the entities "substation", "load", and "yesterday" using natural language processing technology, the user's instruction is obtained as a device state query.

[0051] Interactive answering module: according to the user's instruction, the evaluated power transmission and distribution production field large model is called to perform reasoning and analysis, and the corresponding answer or solution is generated, and the result is fed back to the user.

[0052] It should be noted that the generated answer or solution is fed back to the user through the output interface, and the corresponding answer or solution can be a text form of answer, a visual result in the form of image or chart; interact with the user, provide further explanation or help, and collect the user's feedback on the answer or solution; these feedbacks can be used for subsequent optimization and improvement of the power transmission and distribution production field large model.

[0053] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of the processes should be determined according to their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0054] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0055] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A multi-source data processing and intelligent question-and-answer system based on power transmission and distribution, characterized in that, The system includes the following modules: Data acquisition module one collects multi-source heterogeneous data from the power transmission and distribution production domain and performs standardization processing on the multi-source heterogeneous data; The multi-source data fusion module fuses standardized multi-source heterogeneous data. The large model building module constructs a large model for the power transmission and distribution production field, and performs fine-tuning and enhancement processing on the large model for the power transmission and distribution production field; The model evaluation module constructs a standard evaluation system to conduct standard evaluations of large-scale models in the field of power transmission and distribution production, resulting in the evaluated large-scale models in the field of power transmission and distribution production. Data acquisition module two acquires user voice data and text data, converts user voice data into text, parses the text and text data, and generates user commands; The interactive response module invokes the evaluated large-scale model of power transmission and distribution production based on user commands, performs reasoning and analysis, generates corresponding answers or solutions, and feeds the results back to the user.

2. The multi-source data processing and intelligent question-answering system based on power transmission and distribution according to claim 1, characterized in that, The process of standardizing multi-source heterogeneous data is as follows: Data cleaning and preprocessing are performed on multi-source heterogeneous data. For the pre-processed multi-source heterogeneous data, time-based data is uniformly converted to a specific time format; discrete data is converted into string format and matrix-based; numerical data has its precision, units, and numerical range standardized; text data undergoes word segmentation, part-of-speech tagging, and named entity recognition to extract key information and convert it into text feature vectors; image data undergoes image preprocessing to enhance image features and uses object detection algorithms to identify key information in the image data, forming structured image description data; finally, standardized multi-source heterogeneous data is obtained.

3. The multi-source data processing and intelligent question-answering system based on power transmission and distribution according to claim 1, characterized in that, The process of fusing multi-source heterogeneous data is as follows: Multi-source heterogeneous data is transformed into feature vectors and input into a multimodal Transformer model. The correlation between feature vectors is learned through multiple Transformer encoder layers and modality-specific encoder layers. Contrastive learning is used to align the feature vectors in the same feature space. The feature vectors are then adaptively fused through a gated neural network to obtain the fused multi-source heterogeneous data.

4. The multi-source data processing and intelligent question-answering system based on power transmission and distribution according to claim 1, characterized in that, The process of constructing a large-scale model for power transmission and distribution production includes: We acquire a knowledge graph of the power transmission and distribution production field and construct a large model of the power transmission and distribution production field using fused multi-source heterogeneous data.

5. The multi-source data processing and intelligent question-answering system based on power transmission and distribution according to claim 1, characterized in that, The process of fine-tuning and enhancing a large-scale model in the power transmission and distribution production field includes: Low-order adaptive fine-tuning (LoRA) and reinforcement learning techniques with human feedback are used to fine-tune and enhance large models in the field of power transmission and distribution production.

6. The multi-source data processing and intelligent question-answering system based on power transmission and distribution according to claim 1, characterized in that, The standard evaluation system includes: BLEU score, ROUNGE score, precision, recall, and F1 score.

Images