Cross-regional power dispatching method and system based on digital twinning

Abstract

The invention discloses a cross-regional power dispatching method and system based on digital twinning, and the method comprises the steps: obtaining power equipment parameters, power dispatching parameters and environment parameters in each power supply region, and building a digital twinning model of each power supply region based on the power equipment parameters, the power dispatching parameters and the environment parameters; different types of cross-regional power dispatching tasks are configured in the digital twin model; pre-training an MMOE model of the multi-subtask network, and predicting parameter values of different types of parameters in the digital twin model by using the trained MMOE model; obtaining differential data of parameters of the same type in the power supply area, obtaining power equipment parameters, power dispatching parameters and environment parameters in the current corresponding power supply area, and inputting the differential data, the power equipment parameters, the power dispatching parameters and the environment parameters into the trained MMOE model, and outputting a cross-regional power dispatching task through the task network of the trained MMOE model.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power dispatching, and in particular to a cross-regional electric power dispatching method and system based on digital twins. Background Art

[0002] At present, the traditional power dispatching method usually adopts a centralized control mode for power dispatching. For example, various power-related equipment and sensors collect relevant power data and upload the power data to the central control system. The central control system uses preset dispatching rules, which are generally preset dispatching rules to minimize the power generation cost, and performs cross-regional power dispatching based on the power data. The above-mentioned existing power dispatching methods mainly have the following technical problems: 1. The relatively fixed power dispatching rules make it difficult to handle dynamic and complex collaborative tasks. 2. Power dispatching is a non-convex, high-dimensional and non-linear load distribution problem. Therefore, the traditional power dispatching scheme of the preset rule algorithm has the problem of poor scene adaptability. 3. The existing relatively fixed power dispatching rules do not take into account the impact of the environment on power dispatching, resulting in poor prediction results. Summary of the invention

[0003] One of the inventive purposes of the present invention is to provide a cross-regional power dispatching method and system based on digital twins. The method and system construct a digital twin model, and use the digital twin model to collect in real time parameters including but not limited to energy storage and distribution equipment parameters, power transmission parameters, monitoring and protection equipment parameters, environmental sensor parameters, communication equipment parameters and related material parameters in different regions, and construct a dynamic digital twin model based on the above-mentioned different types of parameters, which is used to dynamically perceive all parameters in the power grid that may affect power dispatching, and dynamically update the digital twin model based on the parameters, wherein the digital twin model provides a cross-regional power dispatching strategy based on the above-mentioned different parameters, and updates the cross-regional power dispatching strategy based on the parameters updated by the digital twin model, thereby achieving accurate and efficient dispatching under complex power dispatching tasks.

[0004] Another invention object of the present invention is to provide a cross-regional power dispatching method and system based on digital twins, wherein the method and system provide an MMOE model (multi-task learning model) as the AI ​​reasoning prediction model of the digital twin model, and use the MMOE model to pre-train different types of subtask networks including but not limited to power equipment subtask networks, environmental data subtask networks, historical data subtask networks, and material performance subtask networks. After the pre-training of the MMOE model is completed, the energy storage and distribution equipment parameters corresponding to physical sensing in different regions, power transmission parameters, monitoring and protection equipment parameters, environmental sensor parameters, communication equipment parameters, and related material parameters are collected and input into the gated network of the trained MMOE model for routing processing of corresponding type parameters, and the subtask reasoning prediction results of the corresponding type parameters are output through the matching subtask network. Different subtask networks select and output weighted results of reasoning prediction corresponding to multiple subtask types through the task type of the MMOE model. In the present invention, the parameters of the digital twin model are updated in real time through the reasoning prediction results of the different types of subtask networks, and the task results of the corresponding MMOE model are executed.

[0005] Another inventive object of the present invention is to provide a cross-regional power dispatching method and system based on digital twins. After obtaining the digital twin model parameters in different regions, the method and system further construct differential data corresponding to the same type of parameters of the digital twin models in different regions, and use the differential data to input into the MMOE model for routing processing. After analyzing the differential data according to different subtask networks, the predicted value of the corresponding type of data in the next time series is output, and the feature fusion processing is performed through the subtask network output weights assigned by the specific task network of the MMOE model to obtain the corresponding task output. The MMOE model can be used to achieve efficient and accurate dispatch of electricity between different regions, and the digital twin model has better scenario adaptability.

[0006] In order to achieve at least one of the above-mentioned invention objects, the present invention further provides a cross-regional power dispatching method based on digital twins, the method comprising: Obtaining power equipment parameters, power dispatching parameters and environmental parameters in each power supply area, and establishing a digital twin model including each power supply area based on the power equipment parameters, power dispatching parameters and environmental parameters; configuring different types of cross-regional power dispatching tasks in the digital twin model; Pre-training an MMOE model of a multi-task network based on historical data including power equipment parameters, power dispatch parameters, environmental parameters, and power dispatch tasks, and using the trained MMOE model to predict parameter values ​​of different types of parameters in the digital twin model; Obtain differential data of the same type of parameters in the power supply area, and obtain the power equipment parameters, power dispatching parameters and environmental parameters in the current corresponding power supply area, and input the differential data, power equipment parameters, power dispatching parameters and environmental parameters into the trained MMOE model. The task network output of the trained MMOE model includes cross-regional power dispatching tasks.

[0007] According to one of the preferred embodiments of the present invention, the method for constructing a digital twin model of a power supply area includes: constructing a data-driven formula of the digital twin model according to the power equipment parameters, power dispatch parameters, and environmental parameters, and configuring a state update formula of the digital twin model, wherein the data-driven formula is: , where f i represents the i-th subtask model function of the MMOE model, y i represents the output result of the i-th subtask model, x t is the input data at time t, θ is the model parameter corresponding to the MMOE model, the state update function of the digital twin model is defined as h(), the state variable at the current time t is s(t), and the state variable is updated according to the following formula: s(t+1)=h(s(t); d t ) ; where s(t+1) is the state variable at time t+1, d t is the updated data at time t.

[0008] According to another preferred embodiment of the present invention, the method includes using the digital twin model to optimize the model according to the task type, and the optimized model formula is: \left [ {min} \right ]J(u)=\int ^{t}_{{t}_{0}} {L(s(t),u(t))}dt , where min represents the minimum constraint of the loss function of the digital twin model according to the task type; J corresponds to the model optimization result of the task type, u represents the control variable, s(t) represents the state variable at time t, L is the loss function, dt represents the differential at time t, and is defined as the corresponding input data at time t x The MMOE model has m subtask models with output results of corresponding tasks, then , Represents the weight of the corresponding subtask i, where the loss function type is matched and set according to the task type.

[0009] According to another preferred embodiment of the present invention, the MMOE model training method includes: performing feature conversion, normalization processing and embedding processing on multi-type historical parameters including the power equipment parameters, power dispatch parameters and environmental parameters, so that the high-dimensional multi-type parameters are mapped into a standard low-dimensional feature vector, and the low-dimensional feature vector is input into the MMOE model for training, a corresponding gating network is configured for each subtask model, the gating network is used to calculate the matching degree between the low-dimensional feature vector and the corresponding subtask model, and the routing weight of the subtask network is dynamically allocated according to the matching degree, and the corresponding type of subtask model is activated according to the routing weight, wherein each subtask model is configured with an independent loss function for calculating the independent task result of each subtask model.

[0010] According to another preferred embodiment of the present invention, the MMOE model training method includes: using an independent feedforward neural network DNN as the neural network of each subtask model, wherein the control formula of the gate control network is: ,in is the output weight of the gating network, k corresponds to the kth subtask, and are the weights and biases of the gating networks respectively, and after all subtasks corresponding to the outputs of all gating networks are weighted for all tasks, the outputs of all subtasks for each task are weighted summed: ,in, represents the weight of the i-th subtask of the k-th gating network, y i Output for the i-th subtask and confirm.

[0011] According to another preferred embodiment of the present invention, the method for calculating the differential data includes: defining the acquired parameter as x j , where j represents the type of parameter and the current detection time is t n , the next adjacent detection time is t n+1 , the differential data corresponding to the type parameter is: ,in It represents the difference of corresponding type parameters at different detection times, and inputs the differential data of corresponding type parameters in different regions into the MMOE model for gating processing and corresponding subtask network processing. The differential data is used to adjust the coordinated processing of power dispatching tasks including intra-regional and cross-regional power dispatching tasks in the time dimension.

[0012] According to another preferred embodiment of the present invention, a dynamic extended vocabulary of different types of parameters is constructed, corresponding label data is pre-constructed based on differential data of different types of parameters, and the label data is input into the dynamic extended vocabulary after feature conversion, and is indexed into a dynamic feature vector using the dynamic extended vocabulary, and the dynamic feature vector of the differential data is input into the MMOE model, and the network weights of all sub-tasks outputted are calculated through the gating network of the MMOE model for the collaborative task of dynamically outputting power scheduling.

[0013] According to another preferred embodiment of the present invention, the method includes: after obtaining differential data of multiple corresponding type parameters, further using K-Means clustering method to map the differential data labels of the same similar categories into the same word embedding, and using the feature labels of the center cluster as feature words in the dynamic expansion vocabulary to reduce the complexity of feature vector processing, so that the MMOE model has dynamic adaptability to new labels.

[0014] In order to achieve at least one of the above-mentioned purposes of the invention, the present invention further provides a cross-regional power dispatching system based on digital twins, which executes the above-mentioned cross-regional power dispatching method based on digital twins.

[0015] The present invention further provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned cross-regional power dispatching method based on digital twins. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Shown is a flow chart of a cross-regional power dispatching method based on digital twins in the present invention. DETAILED DESCRIPTION

[0017] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0018] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0019] Please combine Figure 1, the present invention discloses a cross-regional power dispatching method and system based on digital twins, the method comprising: first, the present invention needs to construct a digital twin model related to power dispatching, especially a digital twin model for different cross-regional power dispatching. It should be noted that the digital twin model in the present invention needs to collect historical data including but not limited to power equipment parameters, power dispatching parameters and environmental parameters, and establish a relational digital twin model based on the parameter type of the above-mentioned historical data collected. The digital twin model can be implemented using system modeling tools including but not limited to MATLAB / Simulink, and the MATLAB / Simulink is used to establish a digital twin model including but not limited to hierarchical classification, and the digital twin model has power dispatching related parameters, such as including but not limited to the rated voltage, rated current, rated capacity, transformation ratio, temperature rise data, efficiency value, no-load current, loss and impedance voltage of the transformer. The equipment types in the power equipment parameters in the present invention include but are not limited to transformers, power dispatching control chips, energy storage batteries, mutual inductors, relays, generators and circuit breakers, etc., wherein each of the above power equipment has its own related power equipment parameters, and the present invention will not be described in detail.

[0020] Specifically, the present invention establishes digital twin models of their respective physical relationships based on the performance parameters of the above-mentioned different devices. For example, a model including SOC and SOH can be established according to the type and performance of the energy storage battery, or a transformer dynamic performance model can be established according to the power transmission performance of the transformer, or a safety model including but not limited to power transmission can be established according to the electrical isolation performance of the relay, etc. The models of the above-mentioned different electrical equipment are all related to the performance of cross-regional power dispatching, and the digital twin models of the above-mentioned power equipment all conform to the power characteristics of the corresponding equipment, so they are universal and not specific. Therefore, the present invention can integrate the relevant models of different power equipment to obtain digital twin models including but not limited to power transmission safety models, power transmission efficiency models, power transmission balance models, and power transmission stability models. It should be noted that due to the different tasks performed by the digital twin model, the present invention does not specifically limit how to construct a digital twin model for cross-regional power dispatching. The present invention pre-trains an MMOE model (multi-task model) based on the digital twin model for cross-regional power dispatching. The MMOE model includes multiple subtask models, each of which is used to solve different categories of technical problems in the digital twin model. For example, the task types of the subtask model include but are not limited to regression prediction tasks, classification tasks for similarity detection, etc. According to the power dispatching task, n subtask types are preset, and n subtask models of the MMOE model are constructed according to the n subtask types, and each subtask model is configured with an independent gating network. The present invention uses the historical data of the power equipment parameters, power dispatching parameters and environmental parameters to train each subtask model of the MMOE model separately, and applies an independent loss function to train each subtask model, so that each subtask model has the ability to independently solve specific power dispatching technical problems. For example, the subtask model includes a power safety transmission task, wherein the power safety transmission includes transformer safety-related parameters, circuit breaker safety parameters, and energy storage battery safety parameters, wherein the above safety parameters include specific temperature thresholds, current thresholds or voltage thresholds, etc. Therefore, after obtaining the safety parameters of power transmission of specific temperature data, overvoltage data and overcurrent data, the above safety parameters can be input into the MMOE model, routed through the gating network of the MMOE model, and input into subtask models including but not limited to the transformer safe operation subtask model and the energy storage battery safe operation subtask model for processing, and the weighted processing results of the corresponding subtask models are output according to the gating network of the MMOE model.

[0021] It is worth mentioning that the present invention utilizes the subtask model of the MMOE model to construct the data-driven model of the digital twin model, and configures the state update formula of the digital twin model, wherein the data-driven formula is: , where fi represents the ith subtask model function of the MMOE model, and the subtask model is not the only model, and multiple subtask types are configured according to the task type and quantity, which will not be described in detail in the present invention. In the above data-driven formula, y i represents the output result of the ith subtask model, which can be the predicted value of the corresponding type parameter or the power dispatch strategy or safety protection strategy output by the corresponding subtask model, etc. t is the input data at time t, θ is the model parameter corresponding to the MMOE model, the state update function of the digital twin model is defined as h(), the state variable at the current time t is s(t), and the state variable is updated according to the following formula: s(t+1)=h(s(t); d t ) ; where s(t+1) is the state variable at time t+1, d t It is the updated data at time t. It should be noted that the state variable can be data after function mapping according to the state parameters of different electrical devices. For example, the state variable can be a safety state, a stability state, an energy-saving state, etc. Different state variables need to call the state parameters of different devices and be obtained through data fusion calculation, wherein the data fusion calculation is an algorithm that can be referred to by those skilled in the art, and the present invention will not be described in detail.

[0022] Furthermore, the training method of the subtask model of the MMOE model of the present invention includes: performing feature conversion, normalization processing and embedding processing on multi-type historical parameters including the power equipment parameters, power dispatching parameters and environmental parameters, so that the high-dimensional multi-type parameters are mapped to standard low-dimensional feature vectors, wherein the environmental parameters include temperature, weather conditions and altitude in different power supply areas, and the environmental parameters can be obtained by obtaining relevant sensors or directly calling real-time weather network data in the local area, wherein the environmental parameters are generally label data of specific vocabulary, so it is necessary to perform unified data preprocessing on the labels including the above-mentioned feature conversion, normalization processing and embedding processing, and input the low-dimensional feature vector into the MMOE model for training, configure a corresponding gating network for each subtask model, use the gating network to calculate the matching degree between the low-dimensional feature vector and the corresponding subtask model, and dynamically allocate the routing weight of the subtask network according to the matching degree, and activate the corresponding type of subtask model according to the routing weight, wherein each subtask model is respectively configured with an independent loss function for calculating the independent task result of each subtask model. Therefore, when the input data is related parameters of different devices, standard feature data can be obtained through feature conversion, normalization processing and embedding processing, and then input into the MMOE model for routing processing. Therefore, the present invention can take into account the diversity of data input content, and at the same time solve the targeted output of data content, thereby reducing the output of irrelevant content and increasing the logic and relevance of the output content.

[0023] It is worth mentioning that the construction method of the gated network includes: using an independent feedforward neural network DNN as the neural network of each subtask model. Since the DNN model has the function of layer-by-layer feature transformation processing, the DNN model can process high-dimensional data and low-dimensional data at the same time, especially for image data, audio data and text data. It has a good processing effect. Therefore, the present invention can use the DNN network to realize the processing of multimodal data and integrate and output the corresponding task results. The control formula of the gated network is ,in is the output weight of the gating network, k corresponds to the kth subtask, and are the weight matrix and bias of the gating network respectively, and after all subtasks corresponding to the outputs of all gating networks are weighted for all tasks, the outputs of all subtasks for each task are weighted summed: ,in, represents the weight of the i-th subtask of the k-th gating network, y i is the output of the ith subtask. It should be noted that each gating network is connected to a subtask network, and the weight and bias of the corresponding task k in the corresponding subtask network are and To determine the output weight of the corresponding subtask network. In the present invention, it is necessary to set the lower limit threshold G of the subtask output weight s , calculate the gated output weights in all subtasks ≥G s The subtask model outputs the result y i (x), and filter out the corresponding subtask model output result y i (x) Perform weighted summation.

[0024] It is worth mentioning that in order to improve the collaborative processing effect of the multi-subtask model, the present invention also uses the digital twin model to optimize the model according to the task type, and the optimized model formula is: \left [ {min} \right ]J(u)=\int ^{t}_{{t}_{0}} {L(s(t),u(t))}dt , where min represents the minimum constraint of the loss function of the digital twin model according to the task type; J corresponds to the model optimization result of the task type, u represents the control variable, s(t) represents the state variable at time t, L is the loss function, dt represents the differential at time t, and is defined as the corresponding input data at time t x The MMOE model has m subtask models with output results of corresponding tasks, then , Represents the weight of the corresponding subtask i, wherein the loss function class is matched and set according to the task type. For example, for a regression prediction task, a loss function that minimizes the mean square error can be set, and the present invention does not limit this in detail.

[0025] The method for calculating the differential data in the present invention includes: defining the acquired parameter as x j , where j represents the type of parameter and the current detection time is t n , the next adjacent detection time is t n+1 , the differential data corresponding to the type parameter is: ,in It represents the difference of the corresponding type of parameters at different detection times, and the differential data of the corresponding type of parameters in different regions are input into the MMOE model for gating processing and corresponding subtask network processing. The differential data is used to adjust the coordinated processing of power dispatch tasks including intra-regional and cross-regional power dispatch tasks in the time dimension. It should be noted that the differential data is used to calculate the change rate of the corresponding parameter type. In a specific subtask, the change rate reflects the stability of the parameter characteristics, which is conducive to the dynamic adjustment of the corresponding digital twin model.

[0026] In order to adapt to the corresponding dynamically changing parameters related to power dispatching, the present invention needs to construct a dynamic expansion vocabulary of different types of parameters, pre-construct corresponding label data based on the differential data of different types of parameters, and input the label data into the dynamic expansion vocabulary after feature conversion. For example, the label data can be labels such as overcurrent, overvoltage, high temperature, etc., and the mapping relationship of the labels is established through the relevant threshold range. When the numerical value of the corresponding type of parameter is collected, the corresponding parameter can be compared within a specific threshold range to extract the corresponding label parameter. The present invention further uses the dynamic expansion vocabulary to index the dynamic feature vector, inputs the dynamic feature vector of the differential data into the MMOE model, and calculates the output network weights of all subtasks through the gated network of the MMOE model, which is used to dynamically output the collaborative task of power dispatching.

[0027] After obtaining differential data of multiple corresponding type parameters, the K-Means clustering method is further used to map the differential data labels of the same similar category into the same word embedding, and the feature labels of the center cluster are used as the feature words in the dynamic expansion vocabulary to reduce the complexity of feature vector processing, so that the MMOE model has dynamic adaptability of new labels. In other preferred embodiments of the present invention, other clustering algorithms can also be used to obtain the feature words of the center cluster for the acquisition of the dynamic vocabulary. The present invention will not be described in detail.

[0028] The embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical cable, RF, etc., or any suitable combination of the foregoing.

[0029] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0030] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be deformed or modified in any way.

Claims

1. A cross-regional power dispatching method based on digital twins, characterized in that: The method comprises: Obtaining power equipment parameters, power dispatching parameters and environmental parameters in each power supply area, and establishing a digital twin model including each power supply area based on the power equipment parameters, power dispatching parameters and environmental parameters; configuring different types of cross-regional power dispatching tasks in the digital twin model; Pre-training an MMOE model of a multi-task network based on historical data including power equipment parameters, power dispatch parameters, environmental parameters, and power dispatch tasks, and using the trained MMOE model to predict parameter values ​​of different types of parameters in the digital twin model; Obtain differential data of the same type of parameters in the power supply area, and obtain the power equipment parameters, power dispatching parameters and environmental parameters in the current corresponding power supply area, and input the differential data, power equipment parameters, power dispatching parameters and environmental parameters into the trained MMOE model. The task network output of the trained MMOE model includes cross-regional power dispatching tasks.

2. According to a cross-regional power dispatching method based on digital twins according to claim 1, it is characterized in that: The method for constructing a digital twin model of a power supply area includes: constructing a data-driven formula of the digital twin model according to the power equipment parameters, power dispatch parameters, and environmental parameters, and configuring a state update formula of the digital twin model, wherein the data-driven formula is: , where f i represents the i-th subtask model function of the MMOE model, y i represents the output result of the i-th subtask model, x t is the input data at time t, θ is the model parameter corresponding to the MMOE model, the state update function of the digital twin model is defined as h(), the state variable at the current time t is s(t), and the state variable is updated according to the following formula: s(t+1)=h(s(t); d t ) ; where s(t+1) is the state variable at time t+1, d t is the updated data at time t.

3. According to a cross-regional power dispatching method based on digital twins according to claim 2, it is characterized in that: The method includes using the digital twin model to perform model optimization according to the task type, and the optimized model formula is: , where min represents the minimum constraint of the loss function of the digital twin model according to the task type; J corresponds to the model optimization result of the task type, u represents the control variable, s(t) represents the state variable at time t, L is the loss function, dt represents the differential at time t, and is defined as the corresponding input data at time t x The MMOE model has m subtask models with output results of corresponding tasks, then , Represents the weight of the corresponding subtask i, where the loss function type is matched and set according to the task type.

4. According to a cross-regional power dispatching method based on digital twins according to claim 1, it is characterized in that: The MMOE model training method includes: performing normalization processing and embedding processing on multi-type historical parameters including the power equipment parameters, power dispatch parameters, and environmental parameters, so that the high-dimensional multi-type parameters are mapped into a standard low-dimensional feature vector, and the low-dimensional feature vector is input into the MMOE model for training, a corresponding gating network is configured for each subtask model, the gating network is used to calculate the matching degree between the low-dimensional feature vector and the corresponding subtask model, and the routing weight of the subtask network is dynamically allocated according to the matching degree, and the corresponding type of subtask model is activated according to the routing weight, wherein each subtask model is configured with an independent loss function for calculating the independent task result of each subtask model.

5. According to a cross-regional power dispatching method based on digital twins according to claim 4, it is characterized in that: The MMOE model training method includes: using an independent feedforward neural network DNN as the neural network of each subtask model, wherein the control formula of the gate control network is: ,in is the output weight of the gating network, k corresponds to the kth task, and are the weights and biases of the gating networks respectively, and after all subtasks corresponding to the outputs of all gating networks are weighted for all tasks, the outputs of all subtasks for each task are weighted summed: ,in, represents the weight of the i-th subtask of the k-th gating network, y i Output of the ith subtask; set the lower limit threshold G of the subtask output weight s , calculate the gated output weights in all subtasks ≥G s The subtask model outputs the result y i (x), and filter out the corresponding subtask model output result y i (x) Perform weighted summation.

6. According to a cross-regional power dispatching method based on digital twins according to claim 1, it is characterized in that: The method for calculating the differential data includes: defining the acquired parameter as x j , where j represents the type of parameter and the current detection time is t n , the next adjacent detection time is t n+1 , the differential data corresponding to the type parameter is: ,in It represents the difference of corresponding type parameters at different detection times, and inputs the differential data of corresponding type parameters in different regions into the MMOE model for gating processing and corresponding subtask network processing. The differential data is used to adjust the collaborative processing of power dispatching tasks including intra-regional and cross-regional power dispatching tasks in the time dimension.

7. The cross-regional power dispatching method based on digital twin according to claim 1 is characterized in that: Construct a dynamic extended vocabulary of different types of parameters, pre-construct corresponding label data based on differential data of different types of parameters, and input the label data into the dynamic extended vocabulary after feature conversion, and use the dynamic extended vocabulary to index into a dynamic feature vector, input the dynamic feature vector of the differential data into the MMOE model, and calculate the output network weights of all subtasks through the gating network of the MMOE model, which are used for collaborative tasks of dynamic output power scheduling.

8. The cross-regional power dispatching method based on digital twin according to claim 4 is characterized in that: The method includes: after obtaining differential data of multiple corresponding type parameters, further using K-Means clustering method to map differential data labels of the same similar category into the same word embedding, and using the feature labels of the center cluster as feature words in the dynamic expansion vocabulary to reduce the complexity of feature vector processing, so that the MMOE model has dynamic adaptability to new labels.

9. A cross-regional power dispatching system based on digital twins, characterized in that: The system executes a cross-regional power dispatching method based on digital twins as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a cross-regional power dispatching method based on digital twins as described in any one of claims 1 to 8.

Images