A power hybrid model construction method and system, electronic equipment and storage medium

Abstract

The application discloses a power hybrid model construction method and system, an electronic device and a storage medium, belongs to the technical field of power automation, and comprises the following steps: constructing a power system dataset; performing task division and supervision training on the power system dataset according to a power business scene, to obtain a plurality of specialized power sub-models; extracting a power system feature set, and performing correlation analysis on the power system feature set and the plurality of specialized power sub-models to determine a mapping relationship; constructing a gating network based on the mapping relationship, and placing the gating network in front of the specialized power sub-models to obtain a power hybrid model; acquiring a target power task, performing feature matching on the target power task in combination with the gating network, and determining an adapted specialized power sub-model; and calling the power hybrid model and processing the task by using the adapted specialized power sub-model. The application solves the problems of single and inflexible traditional power system analysis methods, and realizes efficient integration and utilization of multi-source heterogeneous data of a power system.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of electric power automation, and particularly relates to a power hybrid model construction method and system, an electronic device, and a storage medium. BACKGROUND

[0002] In the power system, there are various business scenarios, such as the power generation side, the power transmission side, the power distribution side, and the power market, etc. A large amount of multi-source heterogeneous data is generated in these scenarios. The power generation side is the source of the power system, and the data of the power generation side is rich and diverse. The traditional thermal power plant will record the operating parameters such as boiler temperature, pressure, and fuel consumption, etc. These data reflect the real-time state of the power generation equipment and the energy utilization efficiency. The hydropower plant focuses on water level, flow rate, and water turbine speed, etc. They are closely related to the utilization of water resources and power generation. With the large-scale access of new energy, the data of wind power plants and photovoltaic power stations is more challenging. Wind power generation needs to monitor parameters such as wind speed, wind direction, and pitch angle in real time to optimize power generation efficiency and ensure equipment safety; photovoltaic power stations record data such as light intensity, component temperature, and inverter output power, which are greatly affected by weather and day-night changes, and have obvious intermittency and volatility. The power transmission side is the channel for power transmission, and the data mainly focuses on the operation state and safety of the power grid. The current, voltage, and power factor of the transmission line are the focus of real-time monitoring, which reflect the load condition and transmission capacity of the power grid. The power distribution side directly faces the end users, and its data is characterized by dispersion and diversity. The operating data of the distribution transformer, such as load rate and oil temperature, reflect the health status and power supply capacity of the transformer. The user's power consumption data, including power consumption, power consumption time, and power factor, are important basis for understanding the user's power consumption behavior and demand. At the same time, with the development of smart grids, a large amount of distributed energy data has emerged at the power distribution side, such as distributed photovoltaic power generation and electric vehicle charging piles, etc. The access of these data further increases the complexity and uncertainty of the data at the power distribution side. The power market is the economic regulator in the power system, and its data mainly involves power trading and prices. The data of the power market, including the bidding data of power generation enterprises, the power demand data of users, the trading rules and settlement data of the power market, etc., together form the data system of the power market. These data reflect the supply and demand relationship and price fluctuations of the power market, and are of great significance for optimizing power resource allocation and improving market efficiency. These data from different business scenarios not only differ in format, such as structured data, semi-structured data, and unstructured data, but also differ in time scale, spatial distribution, and semantic connotation, forming the characteristics of multi-source heterogeneous data.

[0003] Traditional power system analysis methods gradually expose many limitations when facing such complex multi-source heterogeneous data, mainly in that they are usually trained for a single task, and lack consideration of data correlation and model generality in different business scenarios. Because it is unable to effectively integrate and utilize multi-source heterogeneous data from different business scenarios, the correlation between data is ignored, and the information value is not fully mined. At the same time, the model trained for a single task is difficult to reuse and expand, and cannot adapt to the dynamic changes and diversified needs of the power system, increasing the cost of model development and maintenance. Therefore, in the face of the explosive growth of data volume and the continuous improvement of business complexity, traditional power system analysis methods are not up to the task and cannot provide timely and accurate decision support for the operation and management of the power system. SUMMARY

[0004] The purpose of the present application is to solve the problems in the prior art, and to provide a power hybrid model construction method, system, electronic device and storage medium, which realizes efficient integration and utilization of multi-source heterogeneous data of the power system by constructing a power hybrid model, and flexibly calls and adapts according to the task requirements of different power business scenarios, improving the accuracy and efficiency of task processing.

[0005] In order to achieve the above purpose, the present application has the following technical solutions:

[0006] In a first aspect, a power hybrid model construction method is provided, comprising:

[0007] Obtaining multi-source heterogeneous data generated during the operation of the power system and constructing a power system data set, dividing the power system data set into tasks according to the power business scenario and supervising and training to obtain a plurality of specialized power sub-models;

[0008] Extracting a power system feature set and performing correlation analysis on the power system feature set and the obtained plurality of specialized power sub-models to determine the mapping relationship between the power system features and the specialized power sub-models;

[0009] Based on the mapping relationship between the power system features and the specialized power sub-models, a gating network is constructed, and the gating network is placed in front of the specialized power sub-models to obtain a power hybrid model;

[0010] Obtaining a target power task, combining the constructed gating network to perform feature matching on the target power task, and determining the adapted specialized power sub-model;

[0011] Calling the power hybrid model and using the adapted specialized power sub-model to process the target power task, outputting the specialized power sub-model analysis information, and determining the target power task processing result according to the specialized power sub-model analysis information.

[0012] As a preferred scheme, the step of task division and supervised training of the power system dataset according to the power business scenarios to obtain a plurality of specialized power sub-models comprises:

[0013] Abnormal data identification and data preprocessing are performed on the power system dataset to obtain a standard power system dataset;

[0014] A scene demand task set is obtained according to the power business scenarios, and the scene demand task set comprises an analysis sub-task set of each power business scenario;

[0015] The obtained standard power system dataset is divided into a plurality of power scene task datasets according to the scene demand task set;

[0016] The plurality of power scene task datasets are respectively supervised trained and parameter optimized to obtain a plurality of specialized power sub-models.

[0017] As a preferred scheme, the step of abnormal data identification and data preprocessing of the power system dataset to obtain a standard power system dataset comprises:

[0018] The power business data application standard is determined according to the power business scenarios;

[0019] The power system dataset is subjected to abnormal data identification by using the power business data application standard to obtain abnormal power system data;

[0020] The data preprocessing process is called according to the abnormal power system data, and the data preprocessing process comprises data normalization, missing data filling and data cleaning;

[0021] The power system dataset is subjected to data preprocessing based on the called data preprocessing process to obtain a standard power system dataset.

[0022] As a preferred scheme, the step of respectively supervised training and parameter optimization of the plurality of power scene task datasets to obtain a plurality of specialized power sub-models comprises:

[0023] Task characteristics and data characteristics of the plurality of power scene task datasets are obtained;

[0024] Model architecture selection and model parameter initialization are performed based on the task characteristics and the data characteristics to build a plurality of power task model frameworks;

[0025] The plurality of power task model frameworks are respectively used for supervised training of the plurality of power scene task datasets to obtain a plurality of initial power task models;

[0026] The obtained multiple initial power task models are sequentially subjected to verification loss calculation and parameter iteration optimization to obtain multiple specialized power sub-models.

[0027] As a preferred scheme, the step of obtaining multiple initial power task models by supervising training of the multiple power scene task data sets in the multiple power task model frameworks includes:

[0028] The token generation allocation strategy and the load adjustment strategy are determined based on the token load balancing, and the task routing strategy is determined.

[0029] The multiple power scene task data sets are subjected to integrated supervision training based on the token generation allocation strategy and the load adjustment strategy, and the task routing strategy.

[0030] As a preferred scheme, the step of obtaining the target power task and combining the constructed gating network to perform feature matching on the target power task to determine the adapted specialized power sub-model includes:

[0031] The target power task is decomposed according to the scene demand task set to obtain multiple power sub-tasks;

[0032] The feature extraction layer and the model matching layer are determined according to the gating network;

[0033] The obtained multiple power sub-tasks are sequentially subjected to feature extraction and sub-model matching based on the feature extraction layer and the model matching layer to determine the adapted specialized power sub-model.

[0034] As a preferred scheme, the step of extracting the power system feature set includes:

[0035] A power system feature node set is obtained, and the power system feature node set includes a power generation side, a power transmission side, a power distribution side, and a power market;

[0036] The power system data set is subjected to associated feature mining according to the power system feature node set to obtain an associated power node feature set;

[0037] The obtained associated power node feature set is subjected to chi-square calculation evaluation by chi-square test to obtain a power node feature correlation coefficient set;

[0038] The associated power node feature set is filtered based on the power node feature correlation coefficient set to obtain a power system feature set within a preset correlation threshold.

[0039] In a second aspect, a power hybrid model construction system is provided, including:

[0040] The power sub-model construction module is configured to acquire multi-source heterogeneous data generated in the operation process of the power system and construct a power system dataset, perform task division and supervised training on the power system dataset according to a power business scenario, and obtain a plurality of specialized power sub-models;

[0041] The mapping relationship determination module is configured to extract a power system feature set, and perform correlation analysis on the power system feature set and the obtained plurality of specialized power sub-models to determine a mapping relationship between the power system features and the specialized power sub-models;

[0042] The power hybrid model construction module is configured to construct a gating network based on the mapping relationship between the power system features and the specialized power sub-models, and place the gating network in front of the specialized power sub-models to obtain a power hybrid model.

[0043] The power sub-model matching module is configured to acquire a target power task, perform feature matching on the target power task in combination with the constructed gating network, and determine an adapted specialized power sub-model.

[0044] The processing result acquisition module is configured to call the power hybrid model, process the target power task by using the adapted specialized power sub-model, output specialized power sub-model analysis information, and determine a target power task processing result according to the specialized power sub-model analysis information.

[0045] In a third aspect, an electronic device is provided, which includes a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the power hybrid model construction method according to the first aspect.

[0046] In a fourth aspect, a computer readable storage medium is provided, which stores at least one instruction, and the at least one instruction is executed by a processor to implement the power hybrid model construction method according to the first aspect.

[0047] Compared with the prior art, the first aspect of the present application has at least the following beneficial effects:

[0048] The traditional power system analysis method can usually only process a single task and cannot flexibly process tasks in different business scenarios, and the power hybrid model is constructed to realize efficient integration and utilization of multi-source heterogeneous data of the power system, and according to the task requirements of different power business scenarios, the professional power sub-models are flexibly called to process the tasks. In the technical principle, the concept of the power hybrid model is introduced, the power hybrid model is composed of multiple professional power sub-models, and the task distribution is realized through the gating network, and the flexible processing of different power business scenarios is realized. The power system dataset is divided into tasks according to the power business scenario, and multiple professional power sub-models are obtained through supervised training, the mapping relationship between the professional power sub-models and the power system feature set is determined through correlation analysis, and the gating network is constructed based on the mapping relationship, which is in front of the professional power sub-model to form the power hybrid model. When the target power task is obtained, the gating network is combined to perform feature matching on the target power task, and the adaptive professional power sub-model is determined, the power hybrid model is called and the adaptive professional power sub-model is used to process the target power task, the professional power sub-model analysis information is output, and the target power task processing result is determined according to the professional power sub-model analysis information. The power hybrid model is constructed, multiple professional power sub-models are integrated, and the task is flexibly distributed through the gating network, thereby solving the problem of single and inflexible traditional power system analysis method, and improving the accuracy and efficiency of power system task processing.

[0049] It can be understood that the beneficial effects of the second aspect to the fourth aspect can be referred to the related description in the first aspect, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 The power hybrid model construction method flowchart of the embodiment of the present application;

[0052] Figure 2 The power hybrid model construction system structure schematic diagram of the embodiment of the present application. DETAILED DESCRIPTION

[0053] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, technologies, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0054] Referring to Figure 1 The power hybrid model construction method according to the embodiment of the present application mainly includes the following steps:

[0055] S1, acquiring multi-source heterogeneous data generated in the operation process of a power system and constructing a power system dataset, task division and supervised training of the power system dataset according to a power business scenario, and obtaining a plurality of specialized power sub-models;

[0056] S2, extracting a power system feature set, and performing correlation analysis on the power system feature set and the plurality of specialized power sub-models obtained, to determine a mapping relationship between the power system features and the specialized power sub-models;

[0057] S3, constructing a gating network based on the mapping relationship between the power system features and the specialized power sub-models, and placing the gating network in front of the specialized power sub-models to obtain a power hybrid model;

[0058] S4, acquiring a target power task, and performing feature matching on the target power task in combination with the constructed gating network to determine an adapted specialized power sub-model;

[0059] S5, calling the power hybrid model and processing the target power task by using the adapted specialized power sub-model, outputting specialized power sub-model analysis information, and determining a target power task processing result according to the specialized power sub-model analysis information.

[0060] In a possible implementation, the multi-source heterogeneous data generated in the operation process of the power system in step S1 includes data of a power generation side, a power transmission side, a power distribution side, and a power market; wherein the data of the power generation side includes unit output, fuel consumption, etc., the data of the power transmission side includes line load, voltage stability, etc., the data of the power distribution side includes user load, equipment state, etc., and the data of the power market includes electricity price, transaction volume, etc. The power business scenario in step S1 includes load forecasting, fault diagnosis, electricity price forecasting, etc., and further, the step of obtaining a plurality of specialized power sub-models by task division and supervised training of the power system dataset according to the power business scenario in step S1 includes:

[0061] performing abnormal data identification and data preprocessing on the power system dataset to obtain a standard power system dataset;

[0062] obtaining a scenario requirement task set according to the power business scenarios (a power generation side business scenario, a power transmission side business scenario, and a power distribution side business), the scenario requirement task set including an analysis subtask set of each power business scenario;

[0063] performing task division on the obtained standard power system data set according to the scenario requirement task set, to obtain a plurality of power scenario task data sets (such as unit commitment optimization, fuel cost prediction, and the like in the power generation side business scenario, line fault diagnosis, and the like in the power transmission side business scenario, user load prediction, equipment life prediction, and the like in the power distribution side business scenario);

[0064] respectively performing supervised training and parameter tuning on the plurality of power scenario task data sets, to obtain a plurality of specialized power submodels.

[0065] In a possible implementation, the step of performing abnormal data identification and data preprocessing on the power system data set to obtain a standard power system data set includes:

[0066] determining power business data application standards (including data format, value range, timeliness, and the like constraint conditions, and the specific power business data application standards are the conventional data standards in the corresponding scenario) according to the power business scenarios;

[0067] performing abnormal data identification on the power system data set by using the power business data application standards (the abnormal data identification in this embodiment can be performed by setting a threshold or using a 3σ principle (σ represents a standard deviation), that is, data exceeding the mean value ± 3 times the standard deviation is regarded as abnormal, and the identified abnormal data is removed), to obtain abnormal power system data;

[0068] calling a data preprocessing process according to the abnormal power system data, the data preprocessing process including data normalization, missing data filling (data filling is performed by using a mean / median filling method, a time series interpolation method, and the like), and data cleaning (data cleaning includes deleting duplicate or invalid record data);

[0069] performing data preprocessing (data normalization, missing data filling, and data cleaning) on the power system data set based on the called data preprocessing process, to obtain a standard power system data set.

[0070] In a possible implementation, the step of respectively performing supervised training and parameter tuning on the plurality of power scenario task data sets, to obtain a plurality of specialized power submodels includes:

[0071] obtaining task characteristics and data characteristics of a plurality of power scene task data sets (the task characteristics are specific problem types to be solved by the sub-models, such as classification, regression, clustering, and time series prediction; the data characteristics are the forms of data, such as structured data, unstructured data, time series data, image data, and data size; different model frameworks are adapted to different task characteristics and data characteristics);

[0072] selecting a model architecture and initializing model parameters based on the task characteristics and the data characteristics (for example, if the task characteristics are time series prediction and the corresponding data characteristics are time series data, the preferred model can be a time convolution network; if the task characteristics are classification and the corresponding data characteristics are image data, a CNN convolutional neural network can be used as the model architecture selection; different task characteristics and data characteristics have corresponding model architecture selection. For example, a power transformer life prediction model has the task characteristics of a regression task and the data characteristics of unstructured data and time series data), and a plurality of power task model frameworks are built;

[0073] using the plurality of power task model frameworks built to perform supervised training on the plurality of power scene task data sets (the corresponding power scene task data sets are divided into a training set of 70%, a validation set of 20%, and a test set of 10% in proportion; the plurality of power task model frameworks are trained using the data in the training set), and a plurality of initial power task models are obtained;

[0074] the plurality of initial power task models are sequentially verified for loss calculation and parameter iteration optimization (using the validation set and the test set), and a plurality of specialized power sub-models are obtained.

[0075] Further, the step of using the plurality of power task model frameworks built to perform supervised training on the plurality of power scene task data sets to obtain the plurality of initial power task models includes:

[0076] determining a token generation allocation strategy and a load adjustment strategy based on token load balancing, and determining a task routing strategy;

[0077] integrating and supervising the training of the plurality of power scene task data sets based on the token generation allocation strategy and the load adjustment strategy, and the task routing strategy.

[0078] In the process of supervised training, the computing power resources of each model are allocated and adjusted based on the token generation allocation strategy and the load adjustment strategy to ensure balanced allocation of computing power resources, and the task routing strategy is a high-priority task preemption according to the model training priority. Thus, the balanced allocation of computing power resources and the priority of higher models are ensured.

[0079] In a possible implementation, the step S4 of obtaining the target power task, in combination with the constructed gating network, performs feature matching on the target power task to determine the adaptive specialized power sub-model, which includes:

[0080] According to the scene demand task set, the target power task is decomposed to obtain a plurality of power sub-tasks (a complex power task is decomposed into a plurality of sub-tasks that can be independently processed, each sub-task corresponds to the business scope of a specialized power sub-model, such as load forecasting and voltage stability analysis decomposed from power grid safety evaluation, and the power system features corresponding to the task are determined according to the plurality of power sub-tasks);

[0081] According to the gating network, the feature extraction layer and the model matching layer are determined.

[0082] Based on the feature extraction layer and the model matching layer, the plurality of power sub-tasks obtained are sequentially subjected to feature extraction and sub-model matching to determine the adaptive specialized power sub-model.

[0083] In a possible implementation, the step S2 of extracting the power system feature set includes:

[0084] A power system feature node set is obtained, which includes a power generation side, a power transmission side, a power distribution side, and a power market (the power generation side nodes correspond to thermal power plant units, wind turbine generators, photovoltaic inverters, etc.; the power transmission side nodes correspond to substations, high-voltage lines, circuit breakers, etc.; the power distribution side nodes correspond to power distribution transformers, smart meters, ring network cabinets, etc.; and the power market nodes correspond to transaction center nodes and power price collection points);

[0085] According to the power system feature node set, the power system data set is subjected to associated feature mining to obtain an associated power node feature set (obtaining power system data associated with each power system feature node, and storing various data combinations associated in the associated power node feature set, for example, in a power distribution transformer fault event, data combinations of frequent line voltage drop and current surge occur. The data combinations can be obtained by principal component analysis and frequent item mining);

[0086] The chi-square test is used to perform chi-square calculation and evaluation on the obtained associated power node feature set to obtain a power node feature correlation coefficient set (the power node feature correlation coefficient set is a chi-square value for evaluating each data combination in the associated power node feature set, and the greater the chi-square value, the stronger the correlation between the corresponding feature and the label);

[0087] Based on the power node feature correlation coefficient set, the associated power node feature set is filtered (obtaining the associated power node features with a chi-square value greater than a preset correlation threshold), and a power system feature set within the preset correlation threshold (a preset chi-square value parameter) is obtained.

[0088] Further, in step S2, when the power system feature set and the obtained plurality of specialized power sub-models are associated and analyzed, the data in the power system feature set and the input requirement data and output data of the plurality of specialized power sub-models are respectively acquired, the data features (such as data categories) of the two are compared, the objects with consistent data features between the power system feature set and the specialized power sub-models are acquired, the association relationship between the two is determined, and the mapping relationship between the power system features and the specialized power sub-models is obtained.

[0089] In a possible implementation, in step S5, the power hybrid model is called and the target power task is processed by using the adapted specialized power sub-model. The power hybrid model is a multi-modal and multi-task large model architecture for power systems, which performs different processing tasks. Real-time monitoring data of the current target power task is acquired and input into the specialized power sub-model, the specialized power sub-model analysis information is output, and the target power task processing result is determined according to the specialized power sub-model analysis information. The present application solves the technical problem that the conventional power system analysis method in the prior art usually only processes a single task and cannot flexibly process tasks in different business scenarios. By constructing the power hybrid model, efficient integration and utilization of multi-source heterogeneous data of the power system are realized, the adapted specialized power sub-model is flexibly called for processing according to the task requirements of different power business scenarios, and the accuracy and efficiency of power system task processing are improved.

[0090] Please refer to Figure 2 Another embodiment of the present application also proposes a power hybrid model construction system, which comprises:

[0091] The power sub-model construction module 11 is configured to acquire multi-source heterogeneous data generated in the operation process of the power system and construct a power system data set, perform task division and supervised training on the power system data set according to the power business scenario, and obtain a plurality of specialized power sub-models.

[0092] The mapping relationship determination module 12 is configured to extract a power system feature set, and perform association and analysis on the power system feature set and the obtained plurality of specialized power sub-models, and determine the mapping relationship between the power system features and the specialized power sub-models.

[0093] The power hybrid model construction module 13 is configured to construct a gating network based on the mapping relationship between the power system features and the specialized power sub-models, and place the gating network in front of the specialized power sub-models to obtain a power hybrid model.

[0094] The power sub-model matching module 14 is configured to acquire a target power task, perform feature matching on the target power task in combination with the constructed gating network, and determine an adapted specialized power sub-model.

[0095] The processing result acquisition module 15 is configured to call the power hybrid model, process the target power task by using the adaptive specialized power sub-model, output specialized power sub-model analysis information, and determine the target power task processing result according to the specialized power sub-model analysis information.

[0096] In a possible implementation, the power sub-model construction module 11 performs abnormal data identification and data preprocessing on the power system dataset to obtain a standard power system dataset; acquires a scene demand task set according to the power business scene, the scene demand task set including an analysis sub-task set of each power business scene; performs task division on the obtained standard power system dataset according to the scene demand task set to obtain a plurality of power scene task datasets; and performs supervised training and parameter tuning on the plurality of power scene task datasets respectively to obtain a plurality of specialized power sub-models.

[0097] Further, when the power sub-model construction module 11 performs abnormal data identification and data preprocessing on the power system dataset to obtain a standard power system dataset, the power sub-model construction module 11 determines a power business data application standard according to the power business scene; performs abnormal data identification on the power system dataset by using the power business data application standard to obtain abnormal power system data; calls a data preprocessing process according to the abnormal power system data, the data preprocessing process including data normalization, missing data filling and data cleaning; and performs data preprocessing on the power system dataset based on the called data preprocessing process to obtain a standard power system dataset.

[0098] Further, when the power sub-model construction module 11 performs supervised training and parameter tuning on the plurality of power scene task datasets respectively to obtain a plurality of specialized power sub-models, the power sub-model construction module 11 acquires task characteristics and data characteristics of the plurality of power scene task datasets; performs model architecture selection and model parameter initialization based on the task characteristics and the data characteristics to build a plurality of power task model frameworks; performs supervised training on the plurality of power scene task datasets by using the built plurality of power task model frameworks respectively to obtain a plurality of initial power task models; and performs verification loss calculation and parameter iterative tuning on the obtained plurality of initial power task models in sequence to obtain a plurality of specialized power sub-models.

[0099] Further, when the power sub-model construction module 11 performs supervised training on the plurality of power scene task datasets by using the built plurality of power task model frameworks respectively to obtain a plurality of initial power task models, the power sub-model construction module 11 determines a token generation distribution strategy and a load adjustment strategy based on token load balancing, and determines a task routing strategy; and performs integrated supervised training on the plurality of power scene task datasets based on the token generation distribution strategy and the load adjustment strategy, and the task routing strategy respectively.

[0100] In a possible implementation, the power sub-model matching module 14 acquires the target power task, performs feature matching on the target power task in combination with the constructed gating network, determines the adaptive specialized power sub-model, and decomposes the target power task according to the scene requirement task set to obtain a plurality of power sub-tasks; determines the feature extraction layer and the model matching layer according to the gating network; and performs feature extraction and sub-model matching on the obtained plurality of power sub-tasks in sequence based on the feature extraction layer and the model matching layer to determine the adaptive specialized power sub-model.

[0101] In a possible implementation, when the mapping relationship determining module 12 extracts the power system feature set, acquires a power system feature node set, the power system feature node set includes a power generation side, a power transmission side, a power distribution side, and a power market; performs associated feature mining on the power system data set according to the power system feature node set to obtain an associated power node feature set; performs chi-square calculation and evaluation on the obtained associated power node feature set by using chi-square test to obtain a power node feature correlation coefficient set; and filters the associated power node feature set based on the power node feature correlation coefficient set to obtain the power system feature set within a preset correlation threshold.

[0102] Another embodiment of the present application also provides an electronic device, including a processor and a memory, the processor is used to execute the computer program stored in the memory to realize the power hybrid model construction method.

[0103] Another embodiment of the present application also provides a computer readable storage medium, the computer readable storage medium stores at least one instruction, the at least one instruction is executed by the processor to realize the power hybrid model construction method.

[0104] The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable storage medium can include any entity or device, medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium, etc. capable of carrying the computer program code. It should be noted that the content contained in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals. For the convenience of description, the above content only shows the part related to the embodiment of the present application, and the specific technical details are not disclosed, please refer to the method part of the embodiment of the present application. The computer readable storage medium is non-transitory and can be stored in the storage device formed by various electronic devices, and can realize the execution process recorded in the embodiment of the present application.

[0105] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, apparatus such as a system, or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0106] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0107] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0108] The computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart or block diagram block or blocks. Figure 1 one or more functions specified in one or more of the flowchart or block diagram block or blocks. Figure 1 means for carrying out each of the functionality specified in the flowchart or block diagram block or blocks.

[0109] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing and illustrating, not limiting, the technical solution of the present application. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered in the protection scope of the claims of the present application.

Claims

1. A method for constructing a hybrid power model, characterized in that, include: Acquire multi-source heterogeneous data generated during the operation of the power system and construct a power system dataset. Divide the power system dataset into tasks and conduct supervised training according to the power business scenario to obtain multiple specialized power sub-models. Extract the power system feature set, and perform correlation analysis on the power system feature set and the obtained multiple specialized power sub-models to determine the mapping relationship between the power system features and the specialized power sub-models; A gated network is constructed based on the mapping relationship between power system characteristics and specialized power sub-models, and the gated network is placed in front of the specialized power sub-models to obtain a hybrid power model; The target power task is obtained, and the feature matching of the target power task is performed by combining the constructed gating network to determine the appropriate specialized power sub-model. The system calls upon the hybrid power model and employs an adapted specialized power sub-model to process the target power task, outputs specialized power sub-model analysis information, and determines the processing result of the target power task based on the specialized power sub-model analysis information. The multi-source heterogeneous data generated during the operation of the power system includes data from the generation side, transmission side, distribution side, and electricity market; among which, the generation side data includes unit output and fuel consumption, the transmission side data includes line load and voltage stability, the distribution side data includes user load and equipment status, and the electricity market data includes electricity price and transaction volume; the power business scenarios include load forecasting, fault diagnosis, and electricity price forecasting. The process of acquiring the target power task, and combining the constructed gating network to perform feature matching on the target power task to determine the appropriate specialized power sub-model includes: The target power task is decomposed according to the task set of scenario requirements to obtain multiple power sub-tasks; each power sub-task corresponds to the business scope of a specialized power sub-model, and the power grid security assessment is decomposed into load forecasting and voltage stability analysis; The feature extraction layer and model matching layer are determined based on the gating network; Based on the feature extraction layer and the model matching layer, feature extraction and sub-model matching are performed on the multiple power sub-tasks obtained in sequence to determine the appropriate specialized power sub-model. The steps for extracting the power system feature set include: A set of characteristic nodes of the power system is obtained, which includes the generation side, transmission side, distribution side, and power market; the generation side nodes include thermal power plant units, wind turbine units, and photovoltaic inverters; the transmission side nodes include substations, high-voltage lines, and circuit breakers; the distribution side nodes include distribution transformers, smart meters, and ring main units; and the power market nodes include trading center nodes and electricity price collection points. Based on the set of characteristic nodes of the power system, we perform correlation feature mining on the power system dataset to obtain the feature set of associated power nodes; The chi-square test is used to evaluate the obtained characteristic set of related power nodes and obtain the set of power node characteristic correlation coefficients. The power system feature set is obtained by filtering the power node feature set based on the power node feature correlation coefficient set.

2. The method for constructing a hybrid power model according to claim 1, characterized in that, The steps of dividing the power system dataset into tasks and supervising training based on power business scenarios to obtain multiple specialized power sub-models include: Anomaly identification and data preprocessing were performed on the power system dataset to obtain a standard power system dataset. Based on the power business scenarios, obtain the scenario requirement task set, which includes the set of analysis sub-tasks for each power business scenario; The standard power system dataset is divided into tasks according to the task set of the scenario requirements to obtain multiple power scenario task datasets. Supervised training and parameter tuning were performed on multiple power scenario task datasets to obtain several specialized power sub-models.

3. The method for constructing a hybrid power model according to claim 2, characterized in that, The steps of identifying and preprocessing abnormal data in the power system dataset to obtain a standard power system dataset include: Determine the application standards for power business data based on the power business scenario; The power system dataset is anomaly identified using the power business data application standard to obtain abnormal power system data; Based on abnormal power system data, a data preprocessing process is invoked, which includes data normalization, missing data imputation, and data cleaning. The power system dataset is preprocessed based on the invoked data preprocessing process to obtain a standard power system dataset.

4. The method for constructing a hybrid power model according to claim 2, characterized in that, The steps of supervising and tuning parameters on multiple power scenario task datasets to obtain multiple specialized power sub-models include: Acquire the task characteristics and data characteristics of multiple power scenario task datasets; Based on task characteristics and data characteristics, model architecture selection and model parameter initialization are performed to build multiple power task model frameworks. Multiple power task model frameworks were constructed and supervised training was performed on multiple power scenario task datasets to obtain multiple initial power task models. The obtained initial power task models were successively verified by loss calculation and parameter iterative optimization to obtain multiple specialized power sub-models.

5. The method for constructing a hybrid power model according to claim 4, characterized in that, The steps of using the constructed multiple power task model frameworks to perform supervised training on multiple power scenario task datasets to obtain multiple initial power task models include: Based on token load balancing, determine the token generation and distribution strategy and the load adjustment strategy, and determine the task routing strategy; Based on token generation and allocation strategies, load adjustment strategies, and task routing strategies, integrated supervised training was performed on multiple power scenario task datasets.

6. A system for constructing a hybrid power model, characterized in that, include: The power sub-model construction module is used to acquire multi-source heterogeneous data generated during the operation of the power system and construct a power system dataset. Based on the power business scenario, the power system dataset is divided into tasks and supervised training is performed to obtain multiple specialized power sub-models. The mapping relationship determination module is used to extract the power system feature set and perform correlation analysis on the power system feature set and the obtained multiple specialized power sub-models to determine the mapping relationship between the power system features and the specialized power sub-models. The power hybrid model construction module is used to construct a gated network based on the mapping relationship between power system characteristics and specialized power sub-models, and to prepend the gated network on the specialized power sub-model to obtain the power hybrid model; The power sub-model matching module is used to acquire the target power task, combine the constructed gating network to perform feature matching on the target power task, and determine the appropriate specialized power sub-model. The processing result acquisition module is used to call the power hybrid model and use the adapted specialized power sub-model to process the target power task, output the specialized power sub-model analysis information, and determine the processing result of the target power task based on the specialized power sub-model analysis information. The multi-source heterogeneous data generated during the operation of the power system includes data from the generation side, transmission side, distribution side, and electricity market; among which, the generation side data includes unit output and fuel consumption, the transmission side data includes line load and voltage stability, the distribution side data includes user load and equipment status, and the electricity market data includes electricity price and transaction volume; the power business scenarios include load forecasting, fault diagnosis, and electricity price forecasting. The process of acquiring the target power task, and combining the constructed gating network to perform feature matching on the target power task to determine the appropriate specialized power sub-model includes: The target power task is decomposed according to the task set of scenario requirements to obtain multiple power sub-tasks; each power sub-task corresponds to the business scope of a specialized power sub-model, and the power grid security assessment is decomposed into load forecasting and voltage stability analysis; The feature extraction layer and model matching layer are determined based on the gating network; Based on the feature extraction layer and the model matching layer, feature extraction and sub-model matching are performed on the multiple power sub-tasks obtained in sequence to determine the appropriate specialized power sub-model. The steps for extracting the power system feature set include: A set of characteristic nodes of the power system is obtained, which includes the generation side, transmission side, distribution side, and power market; the generation side nodes include thermal power plant units, wind turbine units, and photovoltaic inverters; the transmission side nodes include substations, high-voltage lines, and circuit breakers; the distribution side nodes include distribution transformers, smart meters, and ring main units; and the power market nodes include trading center nodes and electricity price collection points. Based on the set of characteristic nodes of the power system, we perform correlation feature mining on the power system dataset to obtain the feature set of associated power nodes; The chi-square test is used to evaluate the obtained characteristic set of related power nodes and obtain the set of power node characteristic correlation coefficients. The power system feature set is obtained by filtering the power node feature set based on the power node feature correlation coefficient set.

7. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement the electric hybrid model construction method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction that, when executed by a processor, implements the electric hybrid model construction method as described in any one of claims 1 to 5.

Images