Distributed energy generation power prediction method and system based on federated learning

Abstract

The invention discloses a distributed energy generation power prediction method and system based on federated learning, and the method comprises the steps: deploying a generation power prediction global model of each distributed energy node in a distributed energy network and a corresponding generation power prediction local model in each node cell based on a federated learning framework; initializing global model parameters and local model parameters; performing a first round of iteration, and updating local model parameters according to historical data of the distributed energy nodes; aggregating the updated local model parameters, and updating global model parameters according to an aggregation result; and judging whether parameters of the global model converge or not, if not, performing next iteration, and if yes, predicting the generated power of the distributed energy nodes within a certain time in the future by using the generated power prediction global model at the moment. According to the method, the generated power prediction global model and the generated power prediction local model are respectively deployed, the local model parameters of the data cells with different privacy levels are pointedly aggregated, and the data privacy security of the different cells is considered while the prediction effect of the global model is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, and specifically relates to a distributed energy generation power prediction method and system based on federated learning. Background Art

[0002] In recent years, distributed energy has become an important part of the power system. Distributed energy can use natural energy to provide considerable power supply, reduce carbon emissions from power grid power generation, and help the power grid transform to a low-carbon and green one. Among them, the power generation information of distributed energy has important reference value for the planning, scheduling, and maintenance of the power grid. By predicting the power generation of each distributed energy node in advance, it can assist managers in better formulating power grid scheduling plans. Therefore, the prediction of distributed energy power generation is of great significance.

[0003] With the development of artificial intelligence, the use of deep neural network models for time series prediction has become the main means of distributed energy power generation prediction. However, the defects of this type of model are that its training process usually requires processing a large amount of data, which requires high computing power, and there are also data privacy and security issues when collecting data. Summary of the invention

[0004] The purpose of the present invention is to provide a distributed energy generation power prediction method and system based on federated learning in order to address the above-mentioned problems existing in the prior art.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention proposes a distributed energy generation power prediction method based on federated learning, comprising:

[0007] S1. Based on the federated learning framework, deploy the global power prediction model of each distributed energy node in the distributed energy network, as well as the corresponding local power prediction model in each distributed energy node cell, and initialize the global model parameters and local model parameters;

[0008] S2, perform the first round of iterations, and update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes;

[0009] S3, aggregating the updated local model parameters of each power generation prediction local model, and updating the global model parameters of the power generation prediction global model according to the aggregation result;

[0010] S4. Determine whether the global model parameters have converged. If not, return to step S2 for the next round of iteration. If converged, use the current power generation prediction global model to predict the power generation of distributed energy nodes within a certain period of time in the future.

[0011] The distributed energy node cells include low data privacy level cells and high data privacy level cells;

[0012] In S2, for cells with low data privacy levels, the updating formula of the model parameters is:

[0013]

[0014] In the above formula, k represents a cell with low data privacy level, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，k is the nth data sample of the kth cell;

[0015] For cells with high data privacy levels, a random noise matrix is ​​generated in each cell to perturb the update of its model parameters. The update formula is:

[0016]

[0017] In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，j is the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

[0018] The S3 includes:

[0019] S31. For cells with low data privacy levels, the following formula is used to aggregate the local model parameters of each cell:

[0020]

[0021] In the above formula, K is the total number of cells with low data privacy level, w t+1，Kis the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration;

[0022] For cells with high data privacy levels, the following formula is used to aggregate the local model parameters of each cell:

[0023]

[0024] In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D J is the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed;

[0025] S32. According to the aggregation results of the local model parameters of the cells with low data privacy level and the cells with high data privacy level, the global model parameters of the global model for power generation prediction are updated using the following formula:

[0026]

[0027] In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

[0028] In a second aspect, the present invention proposes a distributed energy generation power prediction system based on federated learning, including a model deployment module, a local model parameter update module, a global model parameter update module, and an iterative judgment module;

[0029] The model deployment module is used to deploy a global power generation prediction model of each distributed energy node in the distributed energy network and a corresponding local power generation prediction model in each distributed energy node cell based on a federated learning framework, and initialize global model parameters and local model parameters;

[0030] The local model parameter updating module is used to perform a first round of iterations and update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes;

[0031] The global model parameter updating module is used to aggregate the updated local model parameters of each power generation prediction local model, and update the global model parameters of the power generation prediction global model according to the aggregation result;

[0032] The iterative judgment module is used to judge whether the global model parameters converge. If not, it returns to the local model parameter update module for the next round of iteration; if converged, the current power generation prediction global model is used to predict the power generation of distributed energy nodes within a certain period of time in the future.

[0033] The local model parameter updating module includes a low data privacy level cell model parameter updating unit and a high data privacy level cell model parameter updating unit;

[0034] The low data privacy level cell model parameter updating unit is used to update the model parameters of the low data privacy level cell using the following formula:

[0035]

[0036] In the above formula, k represents a cell with low data privacy level, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，k is the nth data sample of the kth cell;

[0037] The high data privacy level cell model parameter updating unit is used to generate a random noise matrix in each cell, and update the model parameters of the high data privacy level cell using the following formula:

[0038]

[0039] In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，j is the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

[0040] The global model parameter updating module includes a low data privacy level cell parameter aggregation unit, a high data privacy level cell parameter aggregation unit, and a global model parameter updating unit;

[0041] The low data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the low data privacy level cell using the following formula:

[0042]

[0043] In the above formula, K is the total number of cells with low data privacy level, w t+1，K is the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration;

[0044] The high data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the high data privacy level cell using the following formula:

[0045]

[0046] In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D J is the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed;

[0047] The global model parameter updating unit is used to update the global model parameters of the global model for power generation prediction according to the aggregation results of the local model parameters of the cells with low data privacy level and the cells with high data privacy level by using the following formula:

[0048]

[0049] In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

[0050] In a third aspect, the present invention proposes a distributed energy generation power prediction device based on federated learning, comprising a processor and a memory;

[0051] The memory is used to store computer program code and transmit the computer program code to the processor;

[0052] The processor is used to execute the aforementioned distributed energy generation power prediction method based on federated learning according to the instructions in the computer program code.

[0053] In a fourth aspect, the present invention provides a computer storage medium having a computer program stored thereon;

[0054] When the computer program is executed by the processor, the steps of the aforementioned method for predicting distributed energy generation power based on federated learning are implemented.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. The present invention proposes a distributed energy power generation prediction method and system based on federated learning. The method is based on the federated learning framework, deploys the global power generation prediction model of each distributed energy node in the distributed energy network, and the corresponding local power generation prediction model in each distributed energy node cell, and initializes the global model parameters and local model parameters; then the first round of iteration is performed, and the local model parameters of each local power generation prediction model are updated according to the historical data of the distributed energy node; and the updated local model parameters of each local power generation prediction model are aggregated, and the global model parameters of the global power generation prediction model are updated according to the aggregation results; finally, it is judged whether the global model parameters converge, if not, the next round of iteration is performed, and the local model parameters of each local power generation prediction model are updated again; if converged, the global power generation prediction model at this time is used to predict the power generation of the distributed energy node within a certain period of time in the future. Based on the federated learning framework, the method deploys the global power generation prediction model and local model of each distributed energy node in the distributed energy network respectively, and updates the global model parameters by aggregating the updated local model parameters, making full use of the characteristic data of each distributed energy node cell to make a more accurate power generation prediction.

[0057] 2. The present invention proposes a distributed energy power generation prediction method and system based on federated learning. When deploying the corresponding local power generation prediction model in each distributed energy node cell, this method takes into account the privacy level requirements of data in different cells in the distributed energy power generation network, and aggregates local model parameters with different privacy levels in a targeted manner. While ensuring the global model prediction effect, it also takes into account the data privacy requirements of different cells, thereby protecting the data privacy security in different cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The figure is an overall flow chart of the method of the present invention.

[0059] Figure 2 The structure diagram of the system of the present invention is shown in FIG.

[0060] Figure 3 It is a structural diagram of the device described in the present invention. DETAILED DESCRIPTION

[0061] The present invention is further described in detail below in conjunction with specific implementations and drawings.

[0062] The present invention proposes a distributed energy power generation prediction method and system based on federated learning. Based on the federated learning framework, the distributed energy power generation prediction model is deployed in a targeted manner taking into account the data privacy levels of different cells in the distributed energy power generation network. The cells with low data privacy levels directly upload the model parameters to the central server, and the cells with high data privacy levels upload the local model parameters to the central server after perturbing them. The central server receives the local model parameters of all cells and aggregates them to update the global model, thereby ensuring the prediction effect of the global model while taking into account the data privacy requirements of different cells.

[0063] Embodiment 1:

[0064] like Figure 1 As shown, a distributed energy generation power prediction method based on federated learning is performed in the following steps:

[0065] 1. Based on the federated learning framework, deploy the global power prediction model of each distributed energy node in the distributed energy network, as well as the corresponding local power prediction model in each distributed energy node cell, and initialize the global model parameters and local model parameters;

[0066] The global model for distributed energy generation power prediction is a deep neural network model for sequence prediction. The input of the model is the historical power generation data and historical environmental data of each distributed energy node, and the output is the power generation of the distributed energy node in a certain period of time in the future. The global model parameters include hyperparameters such as the number of network layers in the model and the width of the fully connected layer.

[0067] According to the geographical location and node type of all distributed energy nodes in the distributed energy network, such as photovoltaic nodes, wind turbine nodes and other factors, the distributed energy nodes are divided into several cells G, and considering the privacy level requirements of different distributed energy node cell data, the distributed energy node cells are divided into low data privacy level cells and high data privacy level cells, where the low data privacy level cells are denoted as k=1,...,K, K is the total number of low data privacy level cells, and the high data privacy level cells are denoted as j=1,...,J, J is the total number of high data privacy level cells. A local model for power generation prediction of distributed energy nodes is deployed in each distributed energy node cell, and initial local model parameters of the local model for power generation prediction of each cell are obtained from the server. The initial local model parameters are the same as the initialized global model parameters.

[0068] 2. Carry out the first round of iterations, update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes, and upload the updated local model parameters of each cell to the central server;

[0069] For cells with low data privacy level, in the kth cell, the update formula of its model parameters is:

[0070]

[0071] In the above formula, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，k is the nth data sample of the kth cell;

[0072] For cells with high data privacy level, a random noise matrix n is generated in each cell j. j The update of the perturbation model parameters, the noise matrix obeys Gaussian distribution, and the update formula is:

[0073]

[0074] In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，jis the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

[0075] 3. The central server receives local model parameters from each cell, aggregates the updated local model parameters of each power generation prediction local model, and updates the global model parameters of the power generation prediction global model according to the aggregation results;

[0076] For cells with low data privacy levels, the local model parameters of each cell are directly uploaded to the central server, and the central server directly uses such cell model parameters for aggregation:

[0077]

[0078] In the above formula, K is the total number of cells with low data privacy level, w t+1，K is the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration;

[0079] For cells with high data privacy level, the local model parameters of each cell after adding disturbance are uploaded to the central server for parameter aggregation:

[0080]

[0081] In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D J is the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed;

[0082] The central server aggregates the results of local model parameters for cells with low data privacy levels and cells with high data privacy levels, and updates the global model parameters of the global model for power generation prediction:

[0083]

[0084] In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

[0085] 4. Determine whether the global model parameters have converged. If not, return to step 2 for the next round of iteration. If converged, use the current power generation prediction global model to predict the power generation of distributed energy nodes within a certain period of time in the future.

[0086] Embodiment 2:

[0087] like Figure 2 As shown, a distributed energy generation power prediction system based on federated learning includes a model deployment module, a local model parameter update module, a global model parameter update module, and an iterative judgment module;

[0088] The model deployment module is used to deploy a global power generation prediction model of each distributed energy node in the distributed energy network and a corresponding local power generation prediction model in each distributed energy node cell based on a federated learning framework, and initialize global model parameters and local model parameters;

[0089] The local model parameter updating module is used to perform a first round of iterations and update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes;

[0090] The global model parameter updating module is used to aggregate the updated local model parameters of each power generation prediction local model, and update the global model parameters of the power generation prediction global model according to the aggregation result;

[0091] The iterative judgment module is used to judge whether the global model parameters converge. If not, it returns to the local model parameter update module for the next round of iteration; if converged, the current power generation prediction global model is used to predict the power generation of distributed energy nodes within a certain period of time in the future.

[0092] The local model parameter updating module includes a low data privacy level cell model parameter updating unit and a high data privacy level cell model parameter updating unit;

[0093] The low data privacy level cell model parameter updating unit is used to update the model parameters of the low data privacy level cell using the following formula:

[0094]

[0095] In the above formula, k represents a cell with low data privacy level, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，kis the nth data sample of the kth cell;

[0096] The high data privacy level cell model parameter updating unit is used to generate a random noise matrix in each cell, and update the model parameters of the high data privacy level cell using the following formula:

[0097]

[0098] In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，j is the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

[0099] The global model parameter updating module includes a low data privacy level cell parameter aggregation unit, a high data privacy level cell parameter aggregation unit, and a global model parameter updating unit;

[0100] The low data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the low data privacy level cell using the following formula:

[0101]

[0102] In the above formula, K is the total number of cells with low data privacy level, w t+1，K is the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration;

[0103] The high data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the high data privacy level cell using the following formula:

[0104]

[0105] In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D Jis the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed;

[0106] The global model parameter updating unit is used to update the global model parameters of the global model for power generation prediction according to the aggregation results of the local model parameters of the cells with low data privacy level and the cells with high data privacy level by using the following formula:

[0107]

[0108] In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

[0109] Embodiment 3:

[0110] like Figure 3 As shown, a distributed energy generation power prediction device based on federated learning includes a processor and a memory;

[0111] The memory is used to store computer program code and transmit the computer program code to the processor;

[0112] The processor is used to execute a distributed energy generation power prediction method based on federated learning described in Example 1 according to the instructions in the computer program code.

[0113] Embodiment 4:

[0114] A computer storage medium having a computer program stored thereon;

[0115] When the computer program is executed by the processor, the steps of a distributed energy generation power prediction method based on federated learning described in this solution are implemented.

Claims

1. A distributed energy generation power prediction method based on federated learning, characterized in that: The method comprises: S1. Based on the federated learning framework, deploy the global power prediction model of each distributed energy node in the distributed energy network, as well as the corresponding local power prediction model in each distributed energy node cell, and initialize the global model parameters and local model parameters; S2, perform the first round of iterations, and update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes; S3, aggregating the updated local model parameters of each power generation prediction local model, and updating the global model parameters of the power generation prediction global model according to the aggregation result; S4. Determine whether the global model parameters have converged. If not, return to step S2 for the next round of iteration. If converged, use the current power generation prediction global model to predict the power generation of distributed energy nodes within a certain period of time in the future.

2. According to claim 1, a distributed energy generation power prediction method based on federated learning is characterized in that: The distributed energy node cells include low data privacy level cells and high data privacy level cells; In S2, for cells with low data privacy levels, the updating formula of the model parameters is: In the above formula, k represents a cell with low data privacy level, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，k is the nth data sample of the kth cell; For cells with high data privacy levels, a random noise matrix is ​​generated in each cell to perturb the update of its model parameters. The update formula is: In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，j is the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

3. According to a method for predicting distributed energy generation power based on federated learning according to claim 1, it is characterized in that: The S3 includes: S31. For cells with low data privacy levels, the following formula is used to aggregate the local model parameters of each cell: In the above formula, K is the total number of cells with low data privacy level, w t+1，K is the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration; For cells with high data privacy levels, the following formula is used to aggregate the local model parameters of each cell: In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D J is the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed; S32. According to the aggregation results of the local model parameters of the cells with low data privacy level and the cells with high data privacy level, the global model parameters of the global model for power generation prediction are updated using the following formula: In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

4. A distributed energy generation power prediction system based on federated learning, characterized in that: The system includes a model deployment module, a local model parameter update module, a global model parameter update module, and an iteration judgment module; The model deployment module is used to deploy a global power generation prediction model of each distributed energy node in the distributed energy network and a corresponding local power generation prediction model in each distributed energy node cell based on a federated learning framework, and initialize global model parameters and local model parameters; The local model parameter updating module is used to perform a first round of iterations and update the local model parameters of each power generation prediction local model according to the historical data of the distributed energy nodes; The global model parameter updating module is used to aggregate the updated local model parameters of each power generation prediction local model, and update the global model parameters of the power generation prediction global model according to the aggregation result; The iterative judgment module is used to judge whether the global model parameters converge. If not, it returns to the local model parameter update module for the next round of iteration; if converged, the current power generation prediction global model is used to predict the power generation of distributed energy nodes within a certain period of time in the future.

5. A distributed energy generation power prediction system based on federated learning according to claim 4, characterized in that: The local model parameter updating module includes a low data privacy level cell model parameter updating unit and a high data privacy level cell model parameter updating unit; The low data privacy level cell model parameter updating unit is used to update the model parameters of the low data privacy level cell using the following formula: In the above formula, k represents a cell with low data privacy level, w t+1，k is the model parameter of the kth cell in the t+1th round of iteration, η k is the parameter update step size of the kth cell, D k is the data sample size in the kth cell, is the gradient of the loss function of the k-th cell model, d n，k is the nth data sample of the kth cell; The high data privacy level cell model parameter updating unit is used to generate a random noise matrix in each cell, and update the model parameters of the high data privacy level cell using the following formula: In the above formula, j represents a cell with a high data privacy level. is the model parameter in the t+1th iteration after the jth cell is disturbed, η j is the parameter update step size of the kth cell, D j is the data sample size in the jth cell, is the gradient of the loss function of the j-th cell model, d m，j is the mth data sample of the jth cell, w t，j is the model parameter of the jth cell in the tth iteration, n j is a random noise matrix.

6. A distributed energy generation power prediction system based on federated learning according to claim 4, characterized in that: The global model parameter updating module includes a low data privacy level cell parameter aggregation unit, a high data privacy level cell parameter aggregation unit, and a global model parameter updating unit; The low data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the low data privacy level cell using the following formula: In the above formula, K is the total number of cells with low data privacy level, w t+1，K is the model parameter aggregated by all cells with low data privacy level after the t+1th iteration, D K is the data sample size of all cells with low data privacy level, D k is the data sample size in the kth cell, w t+1，k is the model parameter of the k-th cell in the t+1th iteration; The high data privacy level cell parameter aggregation unit is used to aggregate the local model parameters of the high data privacy level cell using the following formula: In the above formula, J is the total number of cells with high data privacy level, w t+1，J is the model parameter aggregated by all cells with high data privacy level after the t+1th iteration, D J is the data sample size of all cells with high data privacy level, D j is the data sample size in the jth cell, The model parameters in the t+1th iteration after the jth cell is disturbed; The global model parameter updating unit is used to update the global model parameters of the global model for power generation prediction according to the aggregation results of the local model parameters of the cells with low data privacy level and the cells with high data privacy level by using the following formula: In the above formula, G represents all the cells divided by distributed energy nodes, w t+1，G are the model parameters of all distributed energy node cells after the t+1th round of iteration.

7. A distributed energy generation power prediction device based on federated learning, characterized in that: including a processor and a memory; The memory is used to store computer program code and transmit the computer program code to the processor; The processor is used to execute a distributed energy generation power prediction method based on federated learning as described in any one of claims 1-3 according to the instructions in the computer program code.

8. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a distributed energy generation power prediction method based on federated learning described in any one of claims 1-3 are implemented.

Images