Data-driven distributed power generation power uncertainty aggregation quantification method

Abstract

The invention discloses a data-driven distributed power generation power uncertainty aggregation quantification method. The method comprises the following steps: collecting interval predicted values of each distributed power generation power in each region in a power distribution network; constructing a global uncertainty aggregation quantification model of the generated power of the distributed power supply at a cloud end, and initializing model parameters; the cloud sends the global uncertainty aggregation quantification model parameters to edge computing devices of different areas of the jurisdiction power distribution network; each edge computing device constructs a local uncertainty aggregation quantitative model training database by using the interval predicted value of the generated power of each distributed power supply in the region, and trains a local uncertainty aggregation quantitative model of the edge device by using the database; and after training is completed, each edge computing device uploads local model parameters to the cloud for updating and the like. According to the method, the accuracy of uncertainty aggregation can be improved on the premise of fully extracting the spatio-temporal characteristics of the data and not leaking the privacy of the user in the region.

Description

Technical Field

[0001] The invention relates to a data-driven distributed power generation power uncertainty aggregation quantification method, belonging to the field of power systems. Background Art

[0002] In order to mitigate climate change and reduce carbon emissions, countries around the world have begun to vigorously develop renewable energy, which has led to the access of a large number of distributed power sources to the power grid, especially in the low-voltage distribution network side. These large-scale access to new energy sources can play an auxiliary supporting role for the power grid, but distributed power sources often have a minimum allowable capacity (generally 1-10MW) when participating in auxiliary support services. In addition, due to the dispersed geographical locations of distributed power sources, the differences in scale, environment and characteristics of distributed power sources in different geographical locations can produce complementary power outputs. In addition, due to the influence of weather factors, there is a large uncertainty in the output of each distributed power source, and when studying the active power aggregation of distributed power sources, this uncertainty aggregation also needs to be characterized. Existing work has ignored the topological connection relationship between distributed power sources, which is unreasonable in actual situations. In view of this, the present invention needs to take into account the topological connection relationship of distributed power sources while aggregating the uncertainty of the generated power of multiple distributed power sources to better provide auxiliary support services for the power grid.

[0003] In addition, for the purpose of user privacy protection, some owners of distributed power sources (such as independent operators) are unwilling to share their own data, which will cause data island problems or incomplete data problems. How to aggregate the uncertainty of distributed power generation in this case is also one of the research points. Therefore, the present invention proposes a data-driven distributed power generation uncertainty aggregation quantification method. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a data-driven distributed power generation power generation uncertainty aggregation and quantification method, which can take into account the topological connection relationship between each distributed power source and realize the aggregation of the uncertainty of distributed power generation power while protecting user privacy.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solution, a data-driven distributed power generation power uncertainty aggregation quantification method, which specifically includes the following steps: (1) Collect the interval prediction value of the power generation of each distributed power source in each area of ​​the distribution network, the topological connection relationship and line admittance between each distributed power source; (2) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud and initialize the model parameters. The cloud sends the initialized global uncertainty aggregation quantification model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantification models. (3) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database, and uses this database to train the local uncertainty aggregation quantization model of the edge device; (4) After the training is completed, each edge computing device uploads the local model parameters to the cloud for updating. The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model. The training of the uncertainty aggregation quantization model is achieved through this iteration. (5) Each edge device uses the trained local uncertainty aggregation and quantification model to realize the uncertainty aggregation and quantification of the power generation of each distributed power source in the region.

[0006] Furthermore, the step (1) comprises: Step (1) is as follows: Collect the interval prediction value of the power generated by each distributed power source in the distribution network , No. The topological connection relationship between each distributed power source in a region It is expressed as: (1) (2) in For the The first distributed power source and the The admittance of the connecting line between the distributed power sources, Indicates the number of distributed generation in the area.

[0007] Furthermore, the step (2) comprises: (2.1) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud, including constructing a spatial characteristic module, a temporal characteristic module, and an uncertainty aggregation quantification module, and initializing the model parameters: (2.1.1) The spatial feature module is specifically expressed as: (3) in , is a weighted adjacency matrix, represents the unit diagonal matrix, Representation Matrix The degree matrix of Indicates the space module The input of the network layer, Indicates the space module +1 layer of network input, Indicates The weight matrix of the layer neural network, is a nonlinear activation function, (2.1.2) The time characteristic module is specifically expressed as: (4) in, is a nonlinear activation function, For the Gate of Forgetfulness, is the input gate, is the output gate, , , and is the weight matrix of the temporal feature module, , , and is the corresponding bias, for The output of the hidden layer at time , for Input at the moment, Representatives will and Concatenate into a vector, is the state of the memory unit at the current moment, yes Input information at the time, for The output of the hidden layer at time t, In order to better learn the temporal characteristics of uncertain data, the input of the temporal feature module is forward data and backward data, which can be expressed as follows: (5) in and yes The forward and backward data at a given moment are output after being processed by the time feature module. concat represents the concatenation operation. Expressed as formula (4), (2.1.3) Uncertainty aggregation quantification module: In order to quantify the uncertainty after the aggregation of each distributed power source, the objective of the uncertainty aggregation quantification model is expressed as: (6) (7) (8) in, represents the loss of the uncertainty aggregation quantization model, Indicated at the quantile level The uncertainty aggregation quantifies the loss of the model under and Respectively represent Samples at the quantile level Predicted value and true value, is the quantile level, is an arbitrarily small positive number, The index aggregation for the dataset, is the index set of quantiles, represents the cardinality function, (2.2) The cloud sends the initialized global uncertainty aggregation quantization model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantization models, which are specifically expressed as: (9) in, Aggregate quantized model parameters for the initial global uncertainty in the cloud, For the The weight parameters of the local initial uncertainty aggregation quantization model of each edge computing device.

[0008] Furthermore, the step (3) comprises: (3.1) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database. Will The data at this moment is the same as before The data at each moment are combined to form the input data of a single sample. The aggregated output at each moment is used as the output data of a single sample. The input data and output data are combined into sample data to form a database. All samples are shuffled, and 70% of the samples are randomly selected as training data, and the remaining samples are used as test data. (3.2) Each edge computing device uses the local database to train the model. The training goal is to minimize the loss function (6) and update the parameters using the gradient descent algorithm, which is expressed as: (10) in, The first step of the local uncertainty aggregation quantification model weight parameters, is the learning rate.

[0009] Furthermore, the step (4) comprises: (4.1) Each edge computing device uploads the local model parameters after local training to the cloud for update, which is specifically expressed as: (11) in For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of edge devices, For the After iterations, the global uncertainty in the cloud aggregates the quantitative model parameters. is the number of data sets possessed by all edge computing devices, For the The number of training datasets owned by each edge computing device; (4.2) The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model, which is expressed as: (12) In the formula For the After iterations, the global uncertainty in the cloud aggregates the quantitative model parameters. For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of each edge device.

[0010] (4.3) Based on (4.1) and (4.2), the cloud and edge computing servers iteratively train the uncertainty aggregation quantization model with the goal of minimizing the sum of the loss functions in each edge computing device until convergence.

[0011] Furthermore, the step (5) comprises: Each edge computing device uses the trained local uncertainty aggregation and quantification model to aggregate and quantify the interval data of each distributed power source, thereby realizing the characterization of the uncertainty of the aggregated distributed power source.

[0012] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the data-driven distributed power generation power generation uncertainty aggregation quantification method is implemented.

[0013] A computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the data-driven distributed power generation power uncertainty aggregation quantification method.

[0014] Compared with the prior art, the data-driven distributed power generation power uncertainty aggregation quantification method provided by the embodiment of the present invention has the following beneficial effects: (1) The distributed power generation uncertainty aggregation quantification method proposed in the present invention takes into account the spatiotemporal correlation characteristics of distributed power generation uncertainty data. By using the time characteristic module and the space characteristic module, it can fully explore the spatiotemporal correlation characteristics of distributed power generation uncertainty data and improve the accuracy of power generation aggregation.

[0015] (2) The distributed power uncertainty aggregation and quantification method proposed in the present invention enables each edge computing device to achieve uncertainty aggregation and quantification without sharing local uncertainty data. It can also achieve the aggregation of the uncertainty of distributed power generation in each area of ​​the distribution network while protecting data privacy.

[0016] (3) The distributed power generation uncertainty aggregation quantification method proposed in the present invention can achieve the uncertainty quantification characterization of the aggregated power generation without assuming the probability distribution of the aggregated uncertainty, and provide the aggregated power results under different confidence intervals. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a data-driven distributed power generation power uncertainty aggregation quantification method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0019] like Figure 1 As shown, this embodiment provides a data-driven distributed power generation power uncertainty aggregation quantification method, including: (1) Collect the interval prediction value of the power generation of each distributed power source in each area of ​​the distribution network, the topological connection relationship and line admittance between each distributed power source; (2) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud and initialize the model parameters. The cloud sends the initialized global uncertainty aggregation quantification model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantification models. (3) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database, and uses this database to train the local uncertainty aggregation quantization model of the edge device; (4) After the training is completed, each edge computing device uploads the local model parameters to the cloud for updating. The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model. The training of the uncertainty aggregation quantization model is achieved through this iteration. (5) Each edge device uses the trained local uncertainty aggregation and quantification model to realize the uncertainty aggregation and quantification of the power generation of each distributed power source in the region.

[0020] The specific steps include: Step 1: Collect the interval prediction values ​​of the power generation of each distributed generation in the distribution network , No. The topological connection relationship between each distributed power source in a region It is expressed as: (1) (2) in For the The first distributed power source and the The admittance of the connecting line between the distributed power sources, Indicates the number of distributed generation in the area.

[0021] Step 2: Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud and initialize the model parameters. The cloud sends the initialized global uncertainty aggregation quantification model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantification models. (2.1) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud, including constructing a spatial characteristic module, a temporal characteristic module, and an uncertainty aggregation quantification module, and initializing the model parameters: (2.1.1) The spatial feature module is specifically expressed as: (3) in , is a weighted adjacency matrix, represents the unit diagonal matrix, Representation Matrix The degree matrix of Indicates the space module The input of the network layer, Indicates the space module +1 layer of network input, Indicates The weight matrix of the layer neural network, is a nonlinear activation function, (2.1.2) The time characteristic module is specifically expressed as: (4) in, is a nonlinear activation function, For the Gate of Forgetfulness, is the input gate, is the output gate, , , and is the weight matrix of the temporal feature module, , , and is the corresponding bias, for The output of the hidden layer at time , for Input at the moment, Representatives will and Concatenate into a vector, is the state of the memory unit at the current moment, yes Input information at the time, for The output of the hidden layer at time t, In order to better learn the temporal characteristics of uncertain data, the input of the temporal feature module is forward data and backward data, which can be expressed as follows: (5) in and yes The forward and backward data at a given moment are output after being processed by the time feature module. concat represents the concatenation operation. Expressed as formula (4), (2.1.3) Uncertainty aggregation quantification module: In order to quantify the uncertainty after the aggregation of each distributed power source, the objective of the uncertainty aggregation quantification model is expressed as: (6) (7) (8) in, represents the loss of the uncertainty aggregation quantization model, Indicated at the quantile level The uncertainty aggregation quantifies the loss of the model under and Respectively represent Samples at the quantile level Predicted value and true value, is the quantile level, is an arbitrarily small positive number, The index aggregation for the dataset, is the index set of quantiles, represents the cardinality function, (2.2) The cloud sends the initialized global uncertainty aggregation quantization model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantization models, which are specifically expressed as: (9) in, Aggregate quantized model parameters for the initial global uncertainty in the cloud, For the The weight parameters of the local initial uncertainty aggregation quantization model of each edge computing device.

[0022] Step 3: Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database, and uses the database to train the local uncertainty aggregation quantization model of the edge device.

[0023] (3.1) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database. Will The data at this moment is the same as before The data at each moment are combined to form the input data of a single sample. The aggregated output at each moment is used as the output data of a single sample. The input data and output data are combined into sample data to form a database. All samples are shuffled, and 70% of the samples are randomly selected as training data, and the remaining samples are used as test data. (3.2) Each edge computing device uses the local database to train the model. The training goal is to minimize the loss function (6) and update the parameters using the gradient descent algorithm, which is expressed as: (10) in, The first step of the local uncertainty aggregation quantification model weight parameters, is the learning rate.

[0024] Step 4: After the training is completed, each edge computing device uploads the local model parameters to the cloud for updating. The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model. The training of the uncertainty aggregation quantization model is achieved through this iteration. (4.1) Each edge computing device uploads the local model parameters after local training to the cloud for update, which is specifically expressed as: (11) in For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of edge devices, For the After iterations, the global uncertainty in the cloud aggregates the quantitative model parameters. is the number of data sets possessed by all edge computing devices, For the The number of training datasets owned by each edge computing device; (4.2) The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model, which is expressed as: (12) In the formula For the After iterations, the global uncertainty in the cloud aggregates the quantitative model parameters. For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of each edge device.

[0025] (4.3) Based on (4.1) and (4.2), the cloud and edge computing servers iteratively train the uncertainty aggregation quantization model with the goal of minimizing the sum of the loss functions in each edge computing device until convergence.

[0026] Step 5: Each edge computing device uses the trained local uncertainty aggregation quantization model to aggregate and quantize the interval data of each distributed power source to achieve the characterization of the uncertainty of the aggregated distributed power source.

[0027] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A data-driven distributed power generation power uncertainty aggregation quantification method, characterized in that: The following steps are involved: (1) Collect the interval prediction value of the power generation of each distributed power source in each area of ​​the distribution network, the topological connection relationship and line admittance between each distributed power source; (2) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud and initialize the model parameters. The cloud sends the initialized global uncertainty aggregation quantification model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantification models. (3) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database, and uses this database to train the local uncertainty aggregation quantization model of the edge device; (4) After the training is completed, each edge computing device uploads the local model parameters to the cloud for updating. The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model. The training of the uncertainty aggregation quantization model is achieved through this iteration. (5) Each edge device uses the trained local uncertainty aggregation and quantification model to realize the uncertainty aggregation and quantification of the power generation of each distributed power source in the region.

2. The data-driven distributed power generation power uncertainty aggregation quantification method according to claim 1 is characterized in that: Step (1) is as follows: Collect the interval prediction value of the power generated by each distributed power source in the distribution network , No. The topological connection relationship between each distributed power source in a region It is expressed as: （1）， （2）， in For the The first distributed power source and the The admittance of the connecting line between the distributed power sources, Indicates the number of distributed generation in the area.

3. The data-driven distributed power generation power uncertainty aggregation quantification method according to claim 1 is characterized in that: Step (2) is as follows: (2.1) Construct a global uncertainty aggregation quantification model of distributed power generation in the cloud, including constructing a spatial characteristic module, a temporal characteristic module, and an uncertainty aggregation quantification module, and initializing the model parameters: (2.1.1) The spatial feature module is specifically expressed as: （3）， in , is a weighted adjacency matrix, represents the unit diagonal matrix, Representation Matrix The degree matrix of Indicates the space module The input of the network layer, Indicates the space module +1 layer of network input, Indicates The weight matrix of the layer neural network, is a nonlinear activation function, (2.1.2) The time characteristic module is specifically expressed as: （4）， in, is a nonlinear activation function, For the Gate of Forgetfulness, is the input gate, is the output gate, , , and is the weight matrix of the temporal feature module, , , and is the corresponding bias, for The output of the hidden layer at time , for Input at the moment, Representatives will and Concatenate into a vector, is the state of the memory unit at the current moment, yes Input information at the time, for The output of the hidden layer at time t, In order to better learn the temporal characteristics of uncertain data, the input of the temporal feature module is forward data and backward data, which can be expressed as follows: （5）， in and yes The forward and backward data at a given moment are output after being processed by the time feature module. concat represents the concatenation operation. Expressed as formula (4), (2.1.3) Uncertainty aggregation quantification module: In order to quantify the uncertainty after the aggregation of each distributed power source, the objective of the uncertainty aggregation quantification model is expressed as: （6）， （7）， （8）， in, represents the loss of the uncertainty aggregation quantization model, Indicated at the quantile level The uncertainty aggregation quantifies the loss of the model under and Respectively represent Samples at the quantile level Predicted value and true value, is the quantile level, is an arbitrarily small positive number, The index aggregation for the dataset, is the index set of quantiles, represents the cardinality function, (2.2) The cloud sends the initialized global uncertainty aggregation quantization model parameters to the edge computing devices that govern different areas of the distribution network as local uncertainty aggregation quantization models, which are specifically expressed as: （9）， in, Aggregate quantized model parameters for the initial global uncertainty in the cloud, For the The weight parameters of the local initial uncertainty aggregation quantization model of each edge computing device.

4. The data-driven distributed power generation power uncertainty aggregation quantification method according to claim 1 is characterized in that: Step (3) is as follows: (3.1) Each edge computing device uses the interval prediction value of the power generation of each distributed power source in the region to build a local uncertainty aggregation quantization model training database. Will The data at this moment is the same as before The data at each moment are combined to form the input data of a single sample. The aggregated output at each moment is used as the output data of a single sample. The input data and output data are combined into sample data to form a database. All samples are shuffled, and 70% of the samples are randomly selected as training data, and the remaining samples are used as test data. (3.2) Each edge computing device uses the local database to train the model. The training goal is to minimize the loss function (6) and update the parameters using the gradient descent algorithm, which is expressed as: （10）， in, The first step of the local uncertainty aggregation quantification model weight parameters, is the learning rate.

5. The data-driven distributed power generation power uncertainty aggregation quantification method according to claim 1 is characterized in that: Step (4) is as follows: (4.1) Each edge computing device uploads the local model parameters after local training to the cloud for update, which is specifically expressed as: （11）， in For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of edge devices, For the After iterations, the global uncertainty aggregation quantifies the model parameters in the cloud. is the number of data sets possessed by all edge computing devices, For the The number of training datasets owned by each edge computing device; (4.2) The cloud then sends the updated global uncertainty aggregation quantization model to each edge computing device to update the local model, which is expressed as: （12）， In the formula For the After iterations, the global uncertainty aggregation quantifies the model parameters in the cloud. For the In the iteration The weight parameters of the local uncertainty aggregation quantization model of each edge device, (4.3) Based on (4.1) and (4.2), the cloud and edge computing servers iteratively train the uncertainty aggregation quantization model with the goal of minimizing the sum of the loss functions in each edge computing device until convergence.

6. The data-driven distributed power generation power uncertainty aggregation quantification method according to claim 1 is characterized in that: Step (5) is as follows: Each edge computing device uses the trained local uncertainty aggregation and quantification model to aggregate and quantify the interval data of each distributed power source, thereby realizing the characterization of the uncertainty of the aggregated distributed power source.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the data-driven distributed power generation power uncertainty aggregation quantification method as described in any one of claims 1 to 6 above is implemented.

8. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, a data-driven distributed power generation power uncertainty aggregation quantification method as described in any one of claims 1-6 is implemented.

Images