A method for processing power dispatch strategy based on natural language processing

Abstract

The present invention relates to the field of electric power dispatching technology, and discloses a method for processing electric power dispatching strategies based on natural language processing, including: obtaining electric power dispatching processing information and performing preprocessing; extracting and identifying key features of the electric power dispatching processing information according to preset dispatching rules to obtain target dispatching analysis results; constructing a multi-task dispatching model through a deep learning algorithm, obtaining historical dispatching data of substations for model training, inputting the target dispatching analysis results into the dispatching model, and generating dispatching instructions; the user performs verification or modification, and outputs the dispatching results. The present invention automatically processes dispatching instructions through natural language to improve processing efficiency, extracts features and associates rules on dispatching information, and obtains accurate dispatching tasks to be processed. By constructing a multi-task model, multiple tasks can be processed simultaneously. It learns rules from historical data, generates adaptive dispatching instructions for each task, and improves dispatching accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power dispatching, and in particular to an electric power dispatching strategy processing method based on natural language processing. Background Art

[0002] In recent years, global energy consumption has continued to grow, and traditional power systems are facing more and more challenges, such as improving the stability, safety and economy of power systems. In order to meet these challenges, smart grid systems have emerged. By integrating various advanced technologies such as information technology, communication technology, and control technology, comprehensive monitoring and optimal management of power systems can be achieved, providing modern society with efficient, clean and sustainable energy services. The application of smart grid technology, especially the implementation of intelligent dispatching automation, has become an important means to improve the operating efficiency of power systems and ensure the security of power supply.

[0003] As an important representative of the new generation of information technology, artificial intelligence technology has powerful data processing and autonomous learning capabilities, providing strong support for the intelligent development of the power industry. However, the existing intelligent dispatching system lacks flexibility and accuracy in the smart grid dispatching mode, resulting in the failure to achieve multi-level dispatching and multi-layer professional coverage in the process of intelligent dispatching, and the lack of integrated architecture design, which limits the adaptability and response speed of dispatching and affects the efficiency and accuracy of power dispatching. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method for processing a power dispatching strategy based on natural language processing to solve the problem of how to improve the efficiency and accuracy of power dispatching.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for processing a power dispatch strategy based on natural language processing, comprising: obtaining power dispatch processing information and performing preprocessing;

[0008] Based on the pre-processed power dispatch processing information, according to the preset dispatch rules, key features of the power dispatch processing information are extracted and identified to obtain target dispatch analysis results;

[0009] Build a multi-task dispatching model through deep learning algorithm, obtain historical dispatching data of substation for model training, input the target dispatching analysis results into the dispatching model, and generate dispatching instructions;

[0010] Based on the scheduling instruction, the user verifies or modifies it and outputs the scheduling result.

[0011] As a preferred solution of the method for processing power dispatch strategy based on natural language processing described in the present invention, the obtaining of power dispatch processing information and preprocessing includes:

[0012] Obtain the power dispatch processing information input by the user and detect the accuracy of the information in real time;

[0013] The preprocessing includes text cleaning, data standardization, stop word removal, and then information segmentation processing.

[0014] As a preferred solution of the power dispatch strategy processing method based on natural language processing described in the present invention, the key feature extraction and identification of the power dispatch processing information includes:

[0015] Based on the pre-processed power dispatch processing information, key features are extracted and classified according to the preset keyword library;

[0016] The key feature classification includes: time feature, region feature, device feature and other keyword features, and the importance of each keyword in the other keyword features is calculated.

[0017] As a preferred solution of the power dispatch strategy processing method based on natural language processing described in the present invention, wherein: the target dispatch analysis result is obtained including:

[0018] According to the first rule in the dispatching rules, the other keyword features are associated with data, the successfully associated keywords are marked, a primary keyword set is generated, and the task type of the current power dispatching processing information is obtained according to the association result;

[0019] Compare the importance values ​​of other keyword features that are not successfully associated with the first rule with the importance threshold. If they are greater than the importance threshold, then sequentially associate data with the second rule in the power dispatching rule, mark the keywords that are successfully associated, generate a secondary keyword set, and obtain a first dispatching device set based on the association results;

[0020] Acquire device information in a corresponding device library according to the device characteristics, generate a second scheduling device set, and merge the first scheduling device set with the second scheduling device set to obtain a target device set;

[0021] The keywords that are unsuccessfully associated with the first rule and the second rule are marked to generate a third-level keyword set, and at least one task goal is determined by combining the keywords in the second-level keyword set with contextual semantic associations.

[0022] As a preferred solution of the method for processing power dispatch strategy based on natural language processing described in the present invention, the construction of a multi-task dispatch model includes:

[0023] The multi-task scheduling model is constructed by using the LSTM algorithm, wherein the first layer of the multi-task scheduling model is an input layer, an input shape of the input layer is set, and input data is received;

[0024] The second layer is a bidirectional LSTM layer with 128 neurons, which captures the temporal dependency of the input sequence, processes the sequence data, and returns the sequence;

[0025] The third layer is the Dropout layer, which sets the Dropout rate;

[0026] The fourth layer is a multi-head output layer, which is a fully connected layer constructed according to the task type, outputs the final scheduling instruction, the activation function is linear, and the output dimension of the fully connected layer is set;

[0027] Use the Adam optimizer, set the learning rate, the loss function is the mean square error, and the evaluation indicator is the mean absolute error.

[0028] As a preferred solution of the method for processing power dispatch strategy based on natural language processing described in the present invention, the acquisition of substation historical dispatch data for model training includes:

[0029] Acquire a data set, including historical power dispatch records, equipment data, and dispatch task objectives of the substation, wherein the historical power dispatch records include task classification, dispatch instructions, execution results, and environmental conditions;

[0030] Set the scheduling tag data and match the corresponding scheduling execution results for each scheduling instruction under each task category;

[0031] The data set is divided into training set, validation set and test set in a ratio of 7:2:1. The model is trained using the training set, the batch size and number of iterations are set, the hyperparameters are adjusted using the validation set data, the model performance is evaluated using the test set data, and the validation set is evaluated after each epoch. If the validation set performance does not improve within the preset rounds, the training is stopped to obtain a trained multi-task scheduling model.

[0032] As a preferred solution of the method for processing power dispatch strategy based on natural language processing described in the present invention, the generating of the dispatch command includes:

[0033] Classifying the key features and creating a mapping library, encoding the input data, generating a vector representation of a fixed dimension, and integrating all features into an input matrix, wherein the input data includes time features, regional features, task types, task targets, and target device sets;

[0034] The input matrix is ​​input into the trained multi-task scheduling model, the corresponding output layer is selected according to the task type, and the corresponding scheduling instruction is generated according to the scheduling label data mapped by the task target in the current task type.

[0035] In a second aspect, the present invention provides a power dispatch strategy processing system based on natural language processing, comprising:

[0036] The collection module is used to obtain power dispatch processing information and perform preprocessing;

[0037] A feature extraction module is used to extract and identify key features of the power dispatch processing information based on the pre-processed power dispatch processing information and according to preset dispatch rules to obtain a target dispatch analysis result;

[0038] A model building module is used to build a multi-task scheduling model through a deep learning algorithm, obtain historical scheduling data of the substation for model training, input the target scheduling analysis results into the scheduling model, and generate scheduling instructions;

[0039] The output module is used to verify or modify the scheduling instruction based on the scheduling instruction and output the scheduling result.

[0040] In a third aspect, the present invention provides an electronic device, comprising:

[0041] Memory and processor;

[0042] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the power dispatch strategy processing method based on natural language processing are implemented.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the power dispatch strategy processing method based on natural language processing.

[0044] Compared with the prior art, the present invention has the following beneficial effects: the present invention automatically processes scheduling instructions through natural language, improves processing efficiency, extracts features and associates rules on scheduling information, obtains accurate scheduling tasks to be processed, and can process multiple tasks simultaneously by building a multi-task model. It learns rules from historical data and generates adaptive scheduling instructions for each task, thereby improving scheduling accuracy. At the same time, it can handle more complex scheduling tasks, improve the overall intelligence level, and allow users to participate in verification and modification to ensure that the scheduling instructions meet actual needs and improve the adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0046] Figure 1 The present invention is a schematic diagram of the overall process of a method for processing a power dispatching strategy based on natural language processing according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0050] Reference Figure 1 Table 1 is an embodiment of the present invention, which provides a method for processing power dispatch strategy based on natural language processing, such as Figure 1 As shown, including:

[0051] S101, Obtain power dispatching processing information and perform preprocessing.

[0052] Preferably, obtain the power dispatching processing information input by the user and detect the accuracy of the information in real time;

[0053] The preprocessing includes text cleaning, data standardization, stop word removal, and then perform information word segmentation processing.

[0054] Specifically, text cleaning includes removing irrelevant characters in the text, such as punctuation marks, special symbols, etc. For numerical data, standardization (such as z-score standardization) or normalization processing (such as Min-Max scaling) is required to ensure that data with different dimensions can be compared on the same scale, or unify the vocabulary form (such as case conversion, abbreviation expansion, etc.) to reduce ambiguity; stop word removal includes removing common stop words, such as "de", "he", "shi", etc. These words usually do not affect the understanding of the text. Information word segmentation processing includes splitting sentences into words or phrases according to power industry terms, and processing through a word segmentation tool applicable to power industry terms. Among them, the power dispatching processing information can be "At 10 o'clock tomorrow morning, the power generation needs to be increased to meet the peak demand", "Rainstorm warning, it is expected that the power demand will increase tonight", "According to the plan, repair the equipment in the substation", etc.

[0055] It should be noted that in this step, all information related to power dispatching is first collected, which can be the information to be processed for power dispatching input by the staff or information such as weather forecasts, including but not limited to power grid load forecasting, the status of generating units, weather forecasts, etc. These raw data need to be preprocessed to facilitate subsequent feature extraction processing.

[0056] S102, Based on the preprocessed power dispatching processing information, according to the preset dispatching rules, extract and identify key features from the power dispatching processing information to obtain the target dispatching analysis result.

[0057] Preferably, the extraction and identification of key features from the power dispatching processing information include:

[0058] Based on the preprocessed power dispatching processing information, extract key features and classify the key features according to the preset keyword library;

[0059] The key feature classification includes: time feature, regional feature, equipment feature, and other keyword features. Calculate the importance of each keyword in the other keyword features, where the importance of the other keyword features is the frequency of the current keyword in the entire power dispatching processing information text, and the calculation method is the number of times the current keyword appears divided by the total number of all keywords in the entire power dispatching processing information text.

[0060] Specifically, the key features in the text information of power dispatching processing can be extracted through algorithms such as TF-IDF and TextRank, and the key variables that have a decisive influence on power dispatching can be identified through pattern recognition. The key features are classified according to the preset keyword library, which defines four keyword types represented by different text keywords. According to the keywords in the text information, they are matched with the meanings of the keyword library, and the key features are divided into time features, regional features, equipment features and other keyword features. Among them, if time features, regional features, and equipment features exist, they are extracted. If they do not exist, only other keyword features are extracted.

[0061] Preferably, obtaining the target scheduling analysis result includes:

[0062] According to the first rule in the dispatching rules, other keyword features are associated with data, the successfully associated keywords are marked, a first-level keyword set is generated, and the task type of the current power dispatching processing information is obtained based on the association results.

[0063] Specifically, first define the type of scheduling task, and define the category of scheduling task according to business needs, such as "planned maintenance", "fault abnormality", "power forecast scheduling", etc.

[0064] Exemplarily, the first rule is a set of rules defined for each scheduling task type, and the set of rules includes a specific set of keywords and the relationships between them. For example, "planned maintenance" involves keywords such as "plan", "maintenance", and "date".

[0065] In another possible implementation, the first rule is established through classifier training, and a classifier is trained considering that it is difficult for the rule to completely cover all situations. The training can be performed using a support vector machine, a neural network, etc., and the training can be performed using a labeled data set so that the model can automatically identify the type of scheduling tasks.

[0066] The importance values ​​of other keyword features that failed to be associated are compared with the importance threshold. If they are greater than the importance threshold, data association is performed with the second rule in the power dispatching rule in sequence, and the keywords that are successfully associated are marked to generate a secondary keyword set, and the first dispatching device set is obtained according to the association result;

[0067] Exemplarily, the importance threshold can be set according to the actual application scenario, and the equipment data that needs to be scheduled is judged based on the importance of the keywords, such as keywords such as "power generation" and "load", and association matching is performed according to the second rule. The association mapped by the second rule is the association relationship between the keywords and the settings, and finally multiple devices that need to be scheduled corresponding to the current keyword are obtained. Finally, the union of the devices corresponding to the multiple keywords that meet the threshold conditions is taken to obtain the first scheduling device set.

[0068] Acquire device information in a corresponding device library according to device characteristics, generate a second scheduling device set, and merge the first scheduling device set with the second scheduling device set to obtain a target device set;

[0069] Specifically, on the first hand, the corresponding equipment obtained according to the equipment characteristics in the key features is the target equipment that needs to be scheduled. On the second hand, the corresponding equipment identified according to the scheduling meaning of other keyword characteristics also belongs to the target equipment that needs to be scheduled. According to the target equipment set, the basic data of the equipment is obtained through the power grid system, including the location, model, status, operating parameters and other contents of the equipment.

[0070] The keywords that are not successfully associated with the first rule and the second rule are marked to generate a third-level keyword set, and at least one task goal is determined by combining the keywords in the second-level keyword set with contextual semantic associations.

[0071] Specifically, the third-level keyword set is generally a verb keyword, such as "increase", "adjust", "process", etc. Based on this keyword, combined with a keyword in the headphone keyword set associated with its context, the task objective of this dispatch is obtained, which can be specifically "increase power generation", "maintain power supply", etc. Finally, information such as time, location, equipment, task type, and task objective are combined to form a complete dispatch task description.

[0072] It should be noted that key features such as time information, location information, task type, task objective, and equipment name are extracted from the preprocessed data, and the preset scheduling rules are used to identify the features that are more important for the scheduling task. By analyzing the key features, specific scheduling task requirements, scheduling equipment and other information are obtained. This step can ensure that the model focuses on the most important features, improve the accuracy and robustness of the model, and make the scheduling task clearer through feature extraction and identification, which facilitates the generation of specific scheduling instructions.

[0073] S103, build a multi-task scheduling model through a deep learning algorithm, obtain historical scheduling data of the substation for model training, input the target scheduling analysis results into the scheduling model, and generate scheduling instructions.

[0074] Specifically, RNN (recurrent neural network) architectures such as LSTM (long short-term memory network) and GRU (gated recurrent unit) can be used to build a multi-task scheduling model to capture long-term dependencies in time series data.

[0075] Exemplarily, building a multi-task scheduling model includes:

[0076] A multi-task scheduling model is built through the LSTM algorithm. The first layer of the multi-task scheduling model is the input layer. The input shape of the input layer is set to receive input data.

[0077] The second layer is an LSTM encoding layer with 128 neurons, which captures the temporal dependency of the input sequence, processes the sequence data, and returns the sequence;

[0078] The third layer is the Dropout layer, which sets the Dropout rate, where the Dropout rate is 0.2 or 0.3;

[0079] The fourth layer is a multi-head output layer. The output layer is a fully connected layer built according to the task type. It outputs the final scheduling instructions. The activation function is linear, and the output dimension of the fully connected layer is set.

[0080] Use the Adam optimizer, set the learning rate, the loss function is the mean square error, and the evaluation indicator is the mean absolute error. The learning rate is set to 0.001 and can be adjusted as needed.

[0081] Specifically, the output dimension of the fully connected layer is set according to the task type and can be set to 3. Each operation instruction includes but is not limited to three parts: device ID, operation type and parameters. These instructions can be specific device operation commands, such as adjusting voltage, changing power output, etc.; among them, device ID: determines the device to be operated, operation type: specific operation instructions, such as adjusting load, increasing power generation, etc., parameter setting: specific value of the operation, such as target power level, voltage adjustment range, etc.

[0082] Obtaining substation historical dispatch data for model training includes:

[0083] Obtaining data sets, including historical power dispatch records, equipment data, and dispatch task objectives of substations. The historical power dispatch records include task classification, dispatch instructions, execution results, and environmental conditions.

[0084] Set the scheduling tag data and match the corresponding scheduling execution results for each scheduling instruction under each task category;

[0085] The dataset is divided into training set, validation set and test set in a ratio of 7:2:1. The model is trained using the training set, the batch size and number of iterations are set, the hyperparameters are adjusted using the validation set data, the model performance is evaluated using the test set data, and the validation set is evaluated after each epoch. If the validation set performance does not improve within the preset rounds, the training is stopped to obtain a trained multi-task scheduling model.

[0086] Specifically, for each scheduling task, the performance of the model is evaluated by calculating the corresponding evaluation indicators (such as accuracy, recall, F1 score, etc.).

[0087] For example, during the training phase, the loss functions of all task types are combined to form a total loss function. A weight factor can be set for each task type to balance the importance of different tasks. The total loss function is expressed as:

[0088] ;

[0089] in, Indicates The loss function for each task type is Indicates The weight coefficient of each task type.

[0090] Specifically, the batch size, the number of iterations, and the preset rounds can be set according to actual needs. In this embodiment, the batch size can be set to 64, the number of iterations to 50, and the preset rounds to 5.

[0091] It should be noted that historical data is used for training to ensure that the model can learn the patterns in past successful scheduling cases, cross-validation is performed to evaluate model performance, hyperparameters are adjusted to optimize the model, and the model can be retrained regularly to adapt to new data and environmental changes.

[0092] Generate scheduling commands including:

[0093] Create a mapping library for key feature classification, encode the input data, generate a fixed-dimensional vector representation, and integrate all features into an input matrix, where the input data includes time features, regional features, task type, task objectives, and target device sets;

[0094] The input matrix is ​​input into the trained multi-task scheduling model, the corresponding output layer is selected according to the task type, and the corresponding scheduling instructions are generated according to the scheduling label data mapped by the task target in the current task type. The generation of scheduling instructions is based on time characteristics, regional characteristics and other safety constraints of the power grid.

[0095] Specifically, if the model predicts that a certain area will experience high load in the next few hours, the dispatch instruction can be to increase the power generation in the area, adjust the generator power, or dispatch additional backup power sources.

[0096] It should be noted that this embodiment uses a deep learning algorithm to build a model that can handle multiple tasks simultaneously. Through multi-task learning, the model can share information between multiple related tasks to improve overall performance. Through training with a large amount of historical data, the model can have good generalization capabilities and is suitable for new scheduling scenarios. The model can directly generate scheduling instructions from input data, realize intelligence, and improve scheduling accuracy.

[0097] S104: Based on the scheduling instruction, the user verifies or modifies the scheduling instruction and outputs the scheduling result.

[0098] Preferably, the generated scheduling instructions are displayed to the user for review;

[0099] If the user finds an error in the instruction, he can make necessary modifications, such as adjusting the scheduling parameters or redefining the scheduling strategy;

[0100] If there are no errors, the final scheduling instructions will be sent and executed, and recorded in the system log.

[0101] Continuously monitor the status of the power grid, adjust the dispatch plan in time if any abnormality is found, and update the dispatch model

[0102] It should be noted that in this step, the user verifies the generated scheduling instructions to confirm the correctness and feasibility of the instructions, which can ensure that the scheduling instructions meet actual needs, avoid errors caused by model errors, and improve the adaptability of the scheduling plan.

[0103] The present invention automatically processes scheduling instructions through natural language to improve processing efficiency, extracts features and associates rules on scheduling information to obtain accurate scheduling tasks to be processed, and can learn rules from historical data by building a multi-task model to generate adaptive scheduling instructions for each task, thereby improving scheduling accuracy. At the same time, it can handle more complex scheduling tasks, improve the overall intelligence level, and allow users to participate in verification and modification to ensure that the scheduling instructions meet actual needs and improve the adaptability of the system.

[0104] The above is a schematic scheme of a power dispatch strategy processing method based on natural language processing in this embodiment. It should be noted that the technical scheme of the power dispatch strategy processing system based on natural language processing and the technical scheme of the power dispatch strategy processing method based on natural language processing belong to the same concept. For details not described in detail in the technical scheme of the power dispatch strategy processing system based on natural language processing in this embodiment, please refer to the description of the technical scheme of the power dispatch strategy processing method based on natural language processing.

[0105] The power dispatch strategy processing system based on natural language processing in this embodiment includes:

[0106] The collection module is used to obtain power dispatch processing information and perform preprocessing;

[0107] The feature extraction module is used to extract and identify key features of the power dispatch processing information based on the pre-processed power dispatch processing information and according to the preset dispatch rules to obtain the target dispatch analysis results;

[0108] The model building module is used to build a multi-task dispatching model through a deep learning algorithm, obtain historical dispatching data of substations for model training, input the target dispatching analysis results into the dispatching model, and generate dispatching instructions;

[0109] The output module is used to verify or modify the scheduling instructions and output the scheduling results.

[0110] This embodiment further provides an electronic device, which is applicable to a case where a power dispatching strategy is processed based on natural language processing, and includes:

[0111] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the power dispatch strategy processing method based on natural language processing as proposed in the above embodiment.

[0112] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for processing a power dispatch strategy based on natural language processing as proposed in the above embodiment is implemented.

[0113] The storage medium proposed in this embodiment and the method for implementing the power dispatching strategy processing based on natural language processing proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0114] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0115] This embodiment provides a method for processing a power dispatch strategy based on natural language processing. In order to verify its beneficial effects, comparison results of two solutions are provided.

[0116] By implementing the dispatching method of the present invention and other traditional dispatching methods, a power simulation experiment is carried out to obtain a power dispatching strategy and compare the dispatching results.

[0117] First, collect data sets and historical dispatch data, including dispatch instructions, execution results, environmental conditions, etc. The data sets should cover different types of dispatch tasks (such as load adjustment, fault anomalies, etc.), and use simulators or simulation tools to simulate the operating status of the power system, the behavior of equipment in the power system, changes in the state of the power grid, etc. Then use historical data to train the model and use the test set to evaluate the model performance.

[0118] The experimental process of the method of the present invention first performs data preprocessing to extract key features, such as task type, equipment, time, area, etc., and encodes them. According to the preset scheduling rules, the key features of the data are extracted and identified to obtain the target scheduling analysis results. The multi-task scheduling model is trained using historical scheduling data, and the scheduling instructions are generated using the LSTM model.

[0119] Finally, the effectiveness of the method is measured by the dispatch accuracy, that is, the matching degree between the dispatch instructions generated by the model and the actual optimal dispatch instructions, the average response time required for the model to generate dispatch instructions, the operating efficiency of the power system after the instructions are executed, that is, system stability and energy consumption, and the accuracy under abnormal conditions. The specific comparison data are shown in Table 1.

[0120] Table 1: Comparative experiment table

[0121] Scheduling method of the present invention Traditional rule-based scheduling method Scheduling accuracy (%) 92.6 84.1 Average scheduling response time (s) 0.3 75.0 System stability (%) 85.0 75.0 Energy consumption improvement rate (%) 12.3 9.8 Abnormal situation accuracy (%) 88.5 81.0

[0122] It can be seen from Table 1 that the LSTM-based multi-task scheduling method of the present invention is superior to the traditional scheduling method in terms of accuracy, response time and execution efficiency. In the power scheduling scenario, the accuracy and response speed of the scheduling instructions are improved, which helps to improve the overall operation efficiency and stability of the power system.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for processing power dispatch strategy based on natural language processing, characterized in that: include: Obtain power dispatch processing information and perform preprocessing; Based on the pre-processed power dispatch processing information, according to the preset dispatch rules, key features of the power dispatch processing information are extracted and identified to obtain target dispatch analysis results; Build a multi-task dispatching model through deep learning algorithm, obtain historical dispatching data of substation for model training, input the target dispatching analysis results into the dispatching model, and generate dispatching instructions; Based on the scheduling instruction, the user verifies or modifies it and outputs the scheduling result; Extracting and identifying key features of the power dispatch processing information includes: Based on the pre-processed power dispatch processing information, key features are extracted and classified according to the preset keyword library; The key feature classification includes: time feature, region feature, device feature and other keyword features, and calculating the importance of each keyword in the other keyword features; The target scheduling analysis result obtained includes: According to the first rule in the dispatching rules, the other keyword features are associated with data, the successfully associated keywords are marked, a primary keyword set is generated, and the task type of the current power dispatching processing information is obtained according to the association result; Compare the importance values ​​of other keyword features that are not successfully associated with the first rule with the importance threshold. If they are greater than the importance threshold, then sequentially associate data with the second rule in the power dispatching rule, mark the keywords that are successfully associated, generate a secondary keyword set, and obtain a first dispatching device set based on the association results; Acquire device information in a corresponding device library according to the device characteristics, generate a second scheduling device set, and merge the first scheduling device set with the second scheduling device set to obtain a target device set; Mark the keywords that are not successfully associated with the first rule and the second rule to generate a third-level keyword set, and determine at least one task goal by combining the keywords in the second-level keyword set with contextual semantic associations; The construction of the multi-task scheduling model includes: The multi-task scheduling model is constructed by using the LSTM algorithm, wherein the first layer of the multi-task scheduling model is an input layer, an input shape of the input layer is set, and input data is received; The second layer is a bidirectional LSTM layer with 128 neurons, which captures the temporal dependency of the input sequence, processes the sequence data, and returns the sequence; The third layer is the Dropout layer, which sets the Dropout rate; The fourth layer is a multi-head output layer, which is a fully connected layer constructed according to the task type, outputs the final scheduling instruction, the activation function is linear, and the output dimension of the fully connected layer is set; Use the Adam optimizer, set the learning rate, the loss function is the mean square error, and the evaluation indicator is the mean absolute error.

2. The method for processing power dispatch strategy based on natural language processing according to claim 1, characterized in that: The obtaining of power dispatch processing information and preprocessing includes: Obtain the power dispatch processing information input by the user and detect the accuracy of the information in real time; The preprocessing includes text cleaning, data standardization, stop word removal, and then information segmentation processing.

3. The method for processing electric power dispatching strategy based on natural language processing according to claim 1, characterized in that: The acquisition of substation historical dispatching data for model training includes: Acquire a data set, including historical power dispatch records, equipment data, and dispatch task objectives of the substation, wherein the historical power dispatch records include task classification, dispatch instructions, execution results, and environmental conditions; Set the scheduling tag data and match the corresponding scheduling execution results for each scheduling instruction under each task category; The data set is divided into training set, validation set and test set in a ratio of 7:2:

1. The model is trained using the training set, the batch size and number of iterations are set, the hyperparameters are adjusted using the validation set data, the model performance is evaluated using the test set data, and the validation set is evaluated after each epoch. If the validation set performance does not improve within the preset rounds, the training is stopped to obtain a trained multi-task scheduling model.

4. The method for processing electric power dispatching strategy based on natural language processing according to claim 3, characterized in that: The generating scheduling command comprises: Classifying the key features and creating a mapping library, encoding the input data, generating a vector representation of a fixed dimension, and integrating all features into an input matrix, wherein the input data includes time features, regional features, task types, task targets, and target device sets; The input matrix is ​​input into the trained multi-task scheduling model, the corresponding output layer is selected according to the task type, and the corresponding scheduling instruction is generated according to the scheduling label data mapped by the task target in the current task type.

5. A power dispatch strategy processing system based on natural language processing, applying the power dispatch strategy processing method based on natural language processing as described in any one of claims 1 to 4, characterized in that: include, The collection module is used to obtain power dispatch processing information and perform preprocessing; A feature extraction module is used to extract and identify key features of the power dispatch processing information based on the pre-processed power dispatch processing information and according to preset dispatch rules to obtain a target dispatch analysis result; A model building module is used to build a multi-task scheduling model through a deep learning algorithm, obtain historical scheduling data of the substation for model training, input the target scheduling analysis results into the scheduling model, and generate scheduling instructions; The output module is used to verify or modify the scheduling instruction based on the scheduling instruction and output the scheduling result.

6. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the power dispatch strategy processing method based on natural language processing as described in any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium, characterized in that: It stores computer executable instructions, which, when executed by a processor, implement the steps of the power dispatch strategy processing method based on natural language processing as described in any one of claims 1 to 4.

Images