Electric power resource intelligent control system based on digital twinning

Abstract

The invention relates to the technical field of electric power system control, in particular to an electric power resource intelligent control system based on digital twinning, which comprises a data acquisition module, a digital twinning model module, an electric power dispatching module, a fault early warning module and a three-dimensional visualization module. According to the method, the load demand prediction model is established based on the hybrid RNN, the RNN and full-connection network hybrid neural network model is constructed, the historical time sequence, the time feature, the historical statistical feature and the similarity feature are learned, and the problem that the power load demand is difficult to predict due to renewable energy source grid connection and power demand increase is solved; a power dispatching decision is made through a multi-objective function optimization algorithm based on an improved particle swarm, the particle swarm optimization algorithm is improved by adding a variation strategy for high-fitness particles, the problem that the particle swarm optimization algorithm is prone to falling into a local optimal solution in the practice process is solved, and therefore the problem that power dispatching of multi-objective optimization is difficult to achieve is solved.

Description

Technical Field

[0001] The present invention relates to the field of power system control technology, and specifically to an intelligent control system for power resources based on digital twins. Background Art

[0002] With the development of smart grid technology, efficient management of power resources has become particularly important. Traditional power resource management systems have limitations in data processing and system optimization, and are difficult to meet the needs of modern power grids. As an emerging technology, digital twin technology can achieve real-time monitoring and in-depth analysis of power systems by creating virtual copies of physical entities, thereby improving the management efficiency and intelligent control level of power resources. However, the current power resource intelligent control system based on digital twins has the problem that the grid connection of renewable energy and the increasing power demand make it difficult to predict the power load demand; the high complexity and strong coupling of the power system make it difficult to achieve power dispatching with multiple objective optimizations. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent control system for electric power resources based on digital twins: in view of the problem that the grid connection of renewable energy and the increasing power demand make it difficult to predict the power load demand, the present invention establishes a load demand prediction model based on a hybrid RNN, specifically constructs a neural network model that is a mixture of RNN and a fully connected network, and learns historical time series, time characteristics, historical statistical characteristics and similarity characteristics; in view of the problem that the high complexity and strong coupling of the power system make it difficult to achieve power dispatching with multiple objectives optimized, the present invention formulates power dispatching decisions by making use of a multi-objective function optimization algorithm based on an improved particle swarm, improves the particle swarm algorithm, and adds a mutation strategy for particles with high fitness, thereby solving the problem that the particle swarm algorithm is prone to fall into a local optimal solution in practice, thereby achieving optimal power dispatching for economic, environmental and safety goals.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent control system for electric power resources based on digital twins, including a data acquisition module, a digital twin model module, a power dispatching module, a fault warning module, and a three-dimensional visualization module. The output end of the data acquisition module is connected to the input end of the digital twin model module, the output end of the digital twin model module is respectively connected to the input ends of the power dispatching module, the fault warning module, and the three-dimensional visualization module, the output end of the power dispatching module is connected to the input end of the three-dimensional visualization module, and the output end of the fault warning module is connected to the input end of the three-dimensional visualization module.

[0005] The data acquisition module monitors power resources in real time and collects entity data;

[0006] The digital twin model module builds and updates the digital twin model of power resources in real time, and predicts the load demand and fault diagnosis of the power system through machine learning algorithms;

[0007] The power dispatch module makes power dispatch decisions and converts them into control commands;

[0008] The fault warning module provides fault warning for power resources;

[0009] The three-dimensional visualization module displays the digital twin model, power dispatching decision and fault warning in three dimensions.

[0010] The data acquisition module is composed of distributed sensors, intelligent instruments and detection equipment. Distributed sensors, intelligent instruments and detection equipment are distributed in key nodes of the power system, including power generation equipment, energy storage equipment, transformer facilities, transmission lines, distribution devices, and load centers, and detect the current, voltage, power, and energy consumption parameters of key nodes in real time.

[0011] The digital twin model module includes a model building unit, a state update unit, and a behavior prediction unit:

[0012] The model building unit builds a digital twin model for the power system. Specifically, it creates a three-dimensional model of the power resources based on the physical characteristics, geometric properties, material and texture characteristics of the power resources, and maps the entity data in the digital twin model.

[0013] The status update unit updates the digital twin model based on the entity data transmitted in real time by the data acquisition module.

[0014] The behavior prediction unit first uses feature engineering technology to preprocess the entity data to obtain preprocessed data, and then establishes a load demand prediction model based on the hybrid RNN model and a fault diagnosis model based on the LSTM model.

[0015] Feature engineering technology is used to preprocess entity data, including: deleting duplicate items in entity data; converting each feature in entity data into a unified unit and format; using clustering algorithms to detect and process distorted data; and using linear interpolation and polynomial interpolation algorithms to fill in missing values ​​for each feature in entity data.

[0016] A load demand forecasting model is established based on the hybrid RNN model. The specific steps are as follows:

[0017] Step K1: Select four types of features from the preprocessed data, including historical time series, time features, historical statistical features, and similarity features;

[0018] Step K2: Construct a hybrid RNN model: The hybrid RNN model consists of an RNN network, a fully connected network, and an output module. The RNN is responsible for processing time series data, including historical time series. The fully connected neural network processes non-time series features, including time features, historical statistical features, and similarity features. The output module includes 1 series layer and 1 fully connected layer. The overall formula is as follows:

[0019] ;

[0020] in represents the load demand forecast value, represents a fully connected layer, RNN represents an RNN network, and FCNN represents a fully connected network. represents the concatenation operation, represents a historical time series, Represents time characteristics, Represents historical statistical characteristics, Indicates similarity characteristics;

[0021] Step K3: Train the hybrid RNN model: Use gradient regularization and back propagation to optimize the model parameters. Specifically, use the mean square error as the loss function and add a regularization term to the loss function. The formula of the regularization term is as follows:

[0022] ;

[0023] in represents the hybrid RNN model parameters, express The corresponding regularization term is, represents the regularization coefficient.

[0024] The formula of the mean square error loss function after the introduction of the regularization term is as follows:

[0025] ;

[0026] in represents the hybrid RNN model parameters, express The corresponding regularization term is, represents the calculation result of the mean square error loss function, represents the number of samples, represents the load demand forecast value, Indicates the actual value of load demand.

[0027] The back-propagation algorithm is used to calculate the gradient of the loss function for each parameter, and the hybrid RNN model parameters are updated according to the gradient and learning rate, and iterate to the preset maximum number of iterations.

[0028] A fault diagnosis model is established based on the LSTM model. Specifically, the power resource status time series data and maintenance records in the preprocessed entity data are used to train the LSTM model. Accuracy, recall rate, and F1 value are selected as performance evaluation indicators, and cross entropy is used as the loss function. The LSTM model structure and hyperparameters are adjusted according to the evaluation results, and the fault diagnosis model is verified through the cross-validation algorithm.

[0029] The power dispatch module includes a decision support unit and an execution unit:

[0030] The decision support unit calls the load demand prediction model of the digital twin model module to obtain the prediction of load demand, and then makes power dispatch decisions through the multi-objective function optimization algorithm based on the improved particle swarm.

[0031] The steps of the multi-objective function optimization algorithm based on improved particle swarm are as follows:

[0032] S1: Define the objective function: including minimization of economic cost, gas emission and voltage deviation, where economic cost includes fuel cost, operation cost and maintenance cost, gas emission is the sum of gas emission of each network node, and power deviation is the sum of voltage deviation of each network node;

[0033] S2: Define constraints: including total power generation, generator output power, network node voltage and line current carrying capacity. The total power generation needs to be greater than or equal to the load demand forecast result, and the generator output power, network node voltage and line current carrying capacity need to be between the upper and lower limits of the corresponding physical quantities;

[0034] S3: Initialize particle swarm: randomly generate a group of particle swarms, each particle represents a solution for power dispatch, and randomly initialize the position and speed of each particle;

[0035] S4: Calculate the fitness of each particle according to the objective function, and update the individual optimal solution and the global optimal solution of each particle according to the fitness;

[0036] S5: Update the speed and position of each particle. The formula for calculating the particle update speed is as follows:

[0037] ;

[0038] in represents the velocity of the individual particle at the tth iteration, represents the inertia weight, and is the learning factor, and is a random number between 0 and 1, is the individual optimal solution, is the global optimal solution, Represents the position of the individual particle at the tth iteration.

[0039] Sort the particles from small to large according to their fitness. The formula for updating the positions of the individuals in the half of the population with low fitness is as follows:

[0040] ;

[0041] in is the position of the individual particle at the tth iteration, represents the velocity at the position at the t+1th iteration.

[0042] For half of the population with high fitness, a mutation strategy is adopted, and the formula for updating the position of individual particles is as follows:

[0043] ;

[0044] in is the position of the individual particle at the tth iteration, and is a random number between 0 and 1, is the global optimal solution, and are the positions of two randomly selected particle individuals in the half of the population with high fitness;

[0045] S6: Repeat S4-S5 until the maximum number of iterations is reached. The global optimal solution at this time is the power dispatch decision.

[0046] The execution unit is responsible for receiving the power dispatching decisions provided by the decision support unit, converting the power dispatching decisions into specific control instructions, adjusting the relevant power generation equipment and energy storage equipment, monitoring the execution of the control instructions, comparing the real-time entity data of the relevant power generation equipment and energy storage equipment with the expected results, and promptly discovering deviations or faults in the execution process.

[0047] The fault warning module includes a fault warning unit and a response coordination unit:

[0048] The fault warning unit calls the fault diagnosis model of the digital twin model module according to the time series data of the power resource status in the real-time entity data to form a diagnosis report.

[0049] The response coordination unit is responsible for receiving the diagnostic report from the fault warning unit and dispatching the maintenance team to perform on-site or remote maintenance based on the diagnostic report. After the maintenance task is completed, feedback information is collected to optimize the fault diagnosis model.

[0050] The three-dimensional visualization module presents the digital twin model, power dispatching decision and fault warning in the form of three-dimensional animation, showing the power generation equipment and energy storage equipment involved in power dispatching and dynamically displaying the path planning, while identifying potential fault points of power resources based on fault warnings.

[0051] The beneficial results achieved by the present invention using the above scheme are as follows:

[0052] (1) The present invention comprehensively monitors power resource data and establishes a digital twin model based on the collected data to achieve real-time mapping between physical entities and digital twins. At the same time, it uses machine learning algorithms to achieve load demand prediction and fault diagnosis, and then realizes power dispatch and fault warning functions. Finally, it uses three-dimensional visualization technology to dynamically display the entire power system, realizing efficient and intelligent management and control of power resources.

[0053] (2) In view of the problem that the grid connection of renewable energy and the continuous increase in electricity demand make it difficult to predict the power load demand, the present invention establishes a load demand prediction model based on a hybrid RNN. Specifically, a neural network model that is a hybrid of RNN and a fully connected network is constructed to learn historical time series, time features, historical statistical features and similarity features, thereby improving the model's ability to capture the dependency of the load demand time series.

[0054] (3) In order to solve the problem that the high complexity and strong coupling of the power system make it difficult to achieve power dispatch with multiple optimization objectives, the present invention formulates power dispatch decisions by using a multi-objective function optimization algorithm based on an improved particle swarm algorithm, and improves the particle swarm algorithm by adding a mutation strategy for high-fitness particles. This solves the problem that the particle swarm algorithm is prone to fall into local optimal solutions in practice, thereby achieving optimal power dispatch with economic, environmental and safety objectives. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a module structure diagram of the intelligent control system of power resources based on digital twins;

[0056] Figure 2 Flowchart of the multi-objective function optimization algorithm based on improved particle swarm.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] Example 1, see Figure 1 , Figure 1 : is a module structure diagram of an intelligent control system for electric power resources based on digital twins. In this embodiment, the intelligent control system for electric power resources based on digital twins includes a data acquisition module, a digital twin model module, a power dispatching module, a fault warning module, and a three-dimensional visualization module. The output end of the data acquisition module is connected to the input end of the digital twin model module, the output end of the digital twin model module is connected to the input end of the power dispatching module, the fault warning module, and the three-dimensional visualization module, respectively, the output end of the power dispatching module is connected to the input end of the three-dimensional visualization module, and the output end of the fault warning module is connected to the input end of the three-dimensional visualization module.

[0060] The data acquisition module monitors power resources in real time and collects entity data.

[0061] The digital twin model module builds and updates the digital twin model of power resources in real time, and predicts the load demand and fault diagnosis of the power system through machine learning algorithms.

[0062] The power dispatching module is responsible for calling the digital twin model module to make power dispatching decisions and converting the power dispatching decisions into control commands.

[0063] The fault warning module is responsible for calling the digital twin model module, monitoring the operation of each power resource in real time, issuing fault warnings for power resources, and coordinating the response actions of various maintenance teams.

[0064] The 3D visualization module uses 3D modeling technology and graphics rendering technology to display the digital twin model, power dispatching decision and fault warning in 3D form, providing a visual user interface.

[0065] Embodiment 2: Based on the above embodiment, this embodiment realizes data collection of power resources, specifically:

[0066] The data acquisition module is composed of a large number of distributed sensors, intelligent instruments and detection equipment, which are installed at various key nodes of the power system, including power generation equipment, energy storage equipment, transformer facilities, transmission lines, distribution equipment, and load centers. Distributed sensors are used to monitor and record key parameters of power resources in real time, including current, voltage, power, and energy consumption; intelligent instruments are used to measure the performance indicators of the power system and environmental parameters related to temperature and humidity; detection equipment is used to diagnose the health status of power equipment, including vibration analyzers, infrared thermal imagers, current transformers, voltage transformers, and power quality analyzers.

[0067] The data acquisition module uses a high-precision analog-to-digital converter to convert the analog signals obtained by distributed sensors, intelligent instruments and detection equipment into digital signals, remove environmental noise and electrical interference from the digital signals through filters, and enhance the strength of the digital signals through amplifiers.

[0068] During the digital signal transmission process, the AES data encryption algorithm is used for encryption. The length of the key and data block is 128 bits. The PKCS5Padding method is used to fill the data blocks less than 128 bits. Then, each data block is subjected to byte replacement, row shift, column confusion, and round key addition operations in turn. These four steps are repeated ten times, and the column confusion operation is not performed for the last time.

[0069] Embodiment 3: Based on the above embodiments, this embodiment realizes the construction of a digital twin model of power resources, specifically:

[0070] After receiving the entity data, the digital twin model module saves the data to the cloud database and calls it through the cloud API instructions.

[0071] The digital twin model module includes a model building unit, a state update unit, and a behavior prediction unit:

[0072] The model building unit builds a digital twin model for the power system, creates a three-dimensional model of the power resource through Creo parametric and Unity3D software, and changes and adjusts the geometric structure, texture details, texture characteristics, and mechanism characteristics of the object according to needs when building the physical model, and maps the corresponding entity data in the digital twin model. After the data twin model is built, it is saved in the database in the cloud.

[0073] The status update unit calls the database in the cloud to obtain the entity data transmitted in real time by the data acquisition module, and updates the parameters of the digital twin model according to the entity data, including the operating efficiency of power resources, load capacity, power consumption, status parameters, environmental parameters, and topological structure.

[0074] The behavior prediction unit first uses feature engineering technology to preprocess the entity data to obtain preprocessed data, and then establishes a load demand prediction model based on a hybrid RNN model and a fault diagnosis model based on an LSTM model.

[0075] Use feature engineering technology to preprocess entity data. The specific steps are as follows:

[0076] Step D1: Remove duplicates in entity data: Compare records in the dataset, identify and remove identical records to avoid bias in the model training process;

[0077] Step D2: Convert each feature in the entity data into a unified unit and format, including converting temperature into Celsius and date and time formats into ISO 8601 format;

[0078] Step D3: Use clustering algorithm to detect and process distorted data: Use k-means algorithm to establish clusters with different data densities for each feature and perform detection, and delete distorted data;

[0079] Step D4: Use linear interpolation and polynomial interpolation algorithms to fill in the missing values ​​of each feature in the entity data: Calculate the missing rate of the entity data, use polynomial interpolation to estimate the missing values ​​for entity data with a missing rate less than 90%, and use simple linear interpolation to estimate the middle value between two known points using a straight line segment for data with a missing rate greater than 90%;

[0080] A load demand forecasting model is established based on the hybrid RNN model. The specific steps are as follows:

[0081] Step K1: Select four types of features from the preprocessed data, including historical time series, time features, historical statistical features and similarity features. The historical time series selects the load demand time series data of the past week; the time features include the hour information, holiday mark and week information of the time series data; the historical statistical features include the maximum load, minimum load and average load in the past week; the similarity features find similar load behaviors by comparing the load patterns in different time periods;

[0082] Step K2: Construct a hybrid RNN model: The hybrid RNN model consists of an RNN network, a fully connected network, and an output module. The RNN is responsible for processing time series data, including historical time series. The fully connected neural network processes non-time series features, including time features, historical statistical features, and similarity features. The activation function of the RNN and fully connected network is tanh. The output module includes 1 series layer and 1 fully connected layer. The series layer concatenates the outputs of the RNN and the fully connected network. The overall formula is as follows:

[0083] ;

[0084] in represents the load demand forecast value, represents a fully connected layer, RNN represents an RNN network, and FCNN represents a fully connected network. represents the concatenation operation, represents the historical time series, Represents time characteristics, Represents historical statistical characteristics, Indicates similarity characteristics;

[0085] Step K3: Training the hybrid RNN model: Input historical time series, time features, historical statistical features and similarity feature data to train the hybrid RNN model. The model outputs the forecast results for load demand. The forecast results are compared with the actual results. The model parameters are optimized through gradient regularization and back propagation to improve the generalization ability and overfitting phenomenon. The mean square error is used as the loss function, and a regularization term is added to the loss function. The formula of the regularization term is as follows:

[0086] ;

[0087] in represents the hybrid RNN model parameters, express The corresponding regularization term is, represents the regularization coefficient.

[0088] The formula of the mean square error loss function after the introduction of the regularization term is as follows:

[0089] ;

[0090] in represents the hybrid RNN model parameters, express The corresponding regularization term is, represents the calculation result of the mean square error loss function, represents the number of samples, represents the load demand forecast value, Indicates the actual value of load demand.

[0091] The back-propagation algorithm is used to calculate the gradient of the loss function for each parameter, and the hybrid RNN model parameters are updated according to the gradient and learning rate, and iterate to the preset maximum number of iterations.

[0092] A fault diagnosis model is established based on the LSTM model. Specifically, the LSTM model is trained using the power resource status time series data and maintenance records in the preprocessed entity data. The types of equipment faults are classified according to the maintenance records, and the current, voltage, temperature, power, and timestamp data of the equipment are integrated to form a time series. After normalization, they are fed into the LSTM model. The model outputs the prediction of the health status of the equipment and the type of equipment fault, and compares them with the actual health status of the equipment and the type of equipment fault. The LSTM model uses accuracy, recall rate, and F1 value as performance evaluation indicators, and uses cross entropy as the loss function. The LSTM model structure and hyperparameters are adjusted according to the evaluation results, and the fault diagnosis model is verified by the cross-validation algorithm.

[0093] Example 4, see Figure 2 Based on the above embodiments, this embodiment formulates and implements power dispatching decisions, specifically:

[0094] The power dispatch module calls the digital twin model module to make power dispatch decisions, converts the power dispatch decisions into control commands, and implements the control commands through automated control technology. The module includes a decision support unit and an execution unit:

[0095] The decision support unit obtains the operating data of power resources from the cloud database, calls the load demand forecasting model of the digital twin model module, inputs the entity data into the load demand forecasting model, obtains the load demand forecasting results, and then makes power dispatching decisions through the multi-objective function optimization algorithm based on the improved particle swarm.

[0096] The steps of the multi-objective function optimization algorithm based on improved particle swarm are as follows:

[0097] S1: Define the objective function: including minimizing economic cost, gas emission and voltage deviation, to achieve the economic, environmental and safety goals of the power system respectively. The economic cost includes fuel cost, operation cost and maintenance cost, gas emission is the sum of gas emission of each network node, and power deviation is the sum of voltage deviation of each network node.

[0098] S2: Define constraints: including total power generation, generator output, network node voltage and line current carrying capacity. The total power generation needs to be greater than or equal to the load demand forecast result, and the generator output power, network node voltage and line current carrying capacity need to be between the upper and lower limits of the corresponding physical quantities;

[0099] S3: Initialize particle swarm: randomly generate a group of particle swarms, each particle represents a solution for power dispatch, randomly initialize the position and speed of each particle, set the number of swarms to 30, the number of iterations to 400, and initialize the iteration counter to 0;

[0100] S4: Calculate the fitness of each particle according to the objective function, and compare the new fitness with the current individual optimal solution and the global optimal solution, and replace the current individual optimal solution and the global optimal solution with the higher fitness;

[0101] S5: Update the speed and position of each particle. The formula for calculating the particle update speed is as follows:

[0102] ;

[0103] in represents the velocity of the individual particle at the tth iteration, represents the inertia weight, and is the learning factor, and is a random number between 0 and 1, is the individual optimal solution, is the global optimal solution, is the position of the individual particle at the tth iteration.

[0104] Sort the particles from small to large according to their fitness. The formula for updating the positions of the individuals in the half of the population with low fitness is as follows:

[0105] ;

[0106] in is the position of the individual particle at the tth iteration, represents the velocity at the position at the t+1th iteration.

[0107] For half of the population with high fitness, a mutation strategy is adopted, and the formula for updating the position of individual particles is as follows:

[0108] ;

[0109] in is the position of the individual particle at the tth iteration, and is a random number between 0 and 1, is the global optimal solution, and are the positions of two randomly selected particle individuals in the half of the population with high fitness.

[0110] S6: Repeat S4-S5 until the preset maximum number of iterations 400 is reached. The global optimal solution at this time is the power dispatch decision.

[0111] The execution unit is responsible for receiving the power dispatching decisions provided by the decision support unit, converting the power dispatching decisions into specific control instructions, and sending the instructions to each power system control center. The control center adjusts the relevant power generation equipment and energy storage equipment, monitors the execution of the control instructions, and compares the real-time entity data of the relevant power generation equipment and energy storage equipment with the expected results, so as to promptly discover deviations or faults in the execution process.

[0112] Embodiment 5: Based on the above embodiments, this embodiment implements and responds to fault warnings, specifically:

[0113] The fault warning module calls the digital twin model module to monitor the operation status of each power resource in real time, issue fault warnings for power resources, and coordinate the response actions of various maintenance teams, including the fault warning unit and the response coordination unit.

[0114] The fault warning unit calls the fault diagnosis model of the digital twin model module based on the real-time physical data of the power resource status, predicts and identifies potential faults in the power system, and forms a diagnosis report.

[0115] The response coordination unit receives the diagnostic report and dispatches the maintenance team to perform on-site or remote maintenance based on the diagnostic report to carry out fault repair, equipment calibration, and system updates. After the maintenance task is completed, feedback information is collected, including maintenance results, measures taken, and causes of failures, and the fault diagnosis model is further optimized based on the new equipment fault information data.

[0116] Embodiment 6: Based on the above embodiments, this embodiment realizes the three-dimensional visualization of the digital twin model, specifically:

[0117] The 3D visualization module presents the digital twin model, power dispatching decision and fault warning in the form of 3D animation: the digital twin model is presented in 3D form, and the real-time entity data of the power resource mapping can be viewed; when viewing the power dispatching decision, the power generation equipment and energy storage equipment involved in the power dispatching are marked with prominent colors, and when planning the path of power transmission and distribution, the path of power flow is dynamically displayed, including the layout of the transmission circuit and the direction of power flow; when the fault warning unit detects a potential fault, the warning icon is used to mark the fault point on the potential faulty equipment so that the operator can quickly identify and respond; in addition, the operator can view the digital twin model from different angles by zooming, rotating and panning the view.

Claims

1. The intelligent control system of power resources based on digital twins is characterized by: It includes a data acquisition module, a digital twin model module, a power dispatching module, a fault warning module, and a three-dimensional visualization module; the output end of the data acquisition module is connected to the input end of the digital twin model module, the output end of the digital twin model module is respectively connected to the input ends of the power dispatching module, the fault warning module and the three-dimensional visualization module, the output end of the power dispatching module is connected to the input end of the three-dimensional visualization module, and the output end of the fault warning module is connected to the input end of the three-dimensional visualization module; The data acquisition module monitors power resources in real time and collects entity data; The digital twin model module constructs and updates the digital twin model of power resources in real time, and predicts the load demand and fault diagnosis of the power system through machine learning algorithms; The power dispatch module makes power dispatch decisions and converts the power dispatch decisions into control commands; The fault warning module provides fault warning for electric power resources; The three-dimensional visualization module presents the digital twin model, power dispatching decision and fault warning in three dimensions.

2. The digital twin-based intelligent control system for electric power resources according to claim 1 is characterized in that: The digital twin model module includes a model building unit, a state updating unit and a behavior prediction unit: The model building unit builds a digital twin model for the power system, specifically creating a three-dimensional model of the power resource according to the physical characteristics, geometric properties, material and texture characteristics of the power resource, and mapping the entity data in the digital twin model; The state updating unit updates the digital twin model in real time; The behavior prediction unit first pre-processes the entity data using feature engineering technology to obtain pre-processed data, then establishes a load demand prediction model based on a hybrid RNN model, and establishes a fault diagnosis model based on an LSTM model; The load demand forecasting model training steps are as follows: Step K1: Select the historical time series, time features, historical statistical features and similarity features in the preprocessed data as the input of the hybrid RNN model; Step K2: Construct a hybrid RNN model: The hybrid RNN model consists of an RNN network, a fully connected network, and an output module. The output module includes 1 series layer and 1 fully connected layer. Step K3: Train the hybrid RNN model: use gradient regularization and back-propagation to optimize parameters, and use mean square error as the loss function.

3. The digital twin-based intelligent control system for electric power resources according to claim 1 or 2, characterized in that: The power dispatch module includes a decision support unit and an execution unit: The decision support unit calls the load demand forecasting model to obtain the load demand forecasting result, and then makes a power dispatch decision through the multi-objective function optimization algorithm based on the improved particle swarm. The specific steps of the multi-objective function optimization algorithm based on the improved particle swarm are as follows: S1: define the objective function, and take minimizing economic cost, gas emission and voltage deviation as the objective function; S2: define constraints, including total power generation, generator output power, network node voltage, and line current carrying capacity; S3: Initialize the particle swarm, set the maximum number of iterations, and randomly initialize the position and velocity of each particle; S4: Calculate the fitness of each particle according to the objective function, and update the individual optimal solution and the global optimal solution of each particle according to the fitness; S5: Update the speed and position of each particle, sort the particles according to their fitness from small to large, and adopt a mutation strategy for the position update of half of the population with high fitness; S6: Repeat S4-S5 until the maximum number of iterations is reached; The execution unit converts the power dispatch decision into a specific control instruction and implements it, while monitoring the execution of the control instruction.

4. The digital twin-based intelligent control system for electric power resources according to claim 1 or 2, characterized in that: The fault warning module includes a fault warning unit and a response coordination unit: The fault warning unit calls the fault diagnosis model to form a diagnosis report; The response coordination unit dispatches maintenance personnel to perform maintenance on the power resources according to the diagnosis report.

Images