Method and system for evaluating balance capability of balance responsibility area
By constructing a balanced responsibility area load and new energy output model based on recurrent neural networks and implicit quantile networks, the problem of difficulty in effectively evaluating the balanced responsibility area of the power system in the existing technology is solved, and a more accurate and efficient modeling effect is achieved.
Patent Information
- Application Number
- CN202411783939.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-06
- Publication Date
- 2025-03-14
AI Technical Summary
It is difficult to effectively utilize historical data in the prior art and it is difficult to accurately evaluate the balance capability of the power system's balance responsibility area, especially when new energy is connected to the power system.
Using a method based on recurrent neural network and implicit quantile network, a balanced responsibility area load and new energy output model is constructed. By training the model and predicting data, combining the quantile loss function and gradient descent method, the model parameters are optimized to improve the modeling effect.
It improves the utilization rate of historical data, can more accurately model the trend of load and new energy output, and provides a more reliable balanced responsibility area balance capability assessment, which is significantly improved compared with traditional statistical methods.
Smart Images
Figure CN119623865A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of power system regulation and control, and in particular to a method and system for evaluating balancing capacity of a balancing responsibility area. Background Art
[0002] At present, my country's new energy industry is developing rapidly, and the installed capacity of new energy is constantly increasing. However. New energy itself is highly uncertain and intermittent. The high proportion of new energy access to the power system has made the active power balance problem of the power system increasingly prominent. Constructing a balancing responsibility area in a power grid with a lower voltage level is an effective means to regulate the balance of the power system and ensure the stable operation of the power system. When dividing the balancing responsibility area, factors such as the load of the power grid in each region, the output of new energy, and the energy storage capacity should be comprehensively considered to evaluate the current balancing capacity of the power grid. This requires an effective load and new energy output modeling method to cover as many changes in load and new energy output as possible, so as to analyze the balancing capacity of the balancing responsibility area.
[0003] The classic modeling methods for power grid load and renewable energy output are based on autoregressive integrated moving average models or Markov models. These traditional statistical methods are difficult to use a large amount of historical data and cannot achieve good modeling results. Given the advantages of machine learning methods in time series modeling, how to effectively use machine learning methods and balancing responsibility area data to evaluate balancing capacity needs to be studied urgently. Summary of the invention
[0004] The purpose of the present invention is to provide a method and system for evaluating the balancing capacity of a balancing responsibility area, which can effectively utilize power balancing market data to evaluate the balancing capacity of a balancing responsibility area.
[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions: A method for evaluating the balancing ability of a balancing responsibility area comprises the following steps: S1: Obtain historical load and new energy output data of the balancing responsibility area and construct a data set; S2: Construct a model for balancing the load in the responsible area and the output of new energy sources based on a recurrent neural network; S3: Based on the data set constructed in S1, train the parameters to be trained in the load and new energy output model described in S2; S4: Input the historical load of the balancing responsibility area and the output data of new energy into the model trained in S3, and output the predicted load and output data of the balancing responsibility area; S5: Based on the forecast results obtained in S4, find the data with the largest supply-demand gap and use its confidence interval as the balancing capacity indicator of the balancing responsibility area.
[0006] Further, step S2 includes the steps of: The load and new energy output model of the balanced responsibility area obtains the forecast data based on the historical load and new energy output data of the balanced responsibility area, and the historical data includes the regional load at time t. and new energy , the forecast data is the load forecast data and new energy forecast data , and then according to the predicted data and predict and ,in, T Indicates the total amount of historical data; The model for balancing the load in the responsible area and the output of new energy sources is based on the regional labels of the data. , predict the load and renewable energy output in different balancing responsibility areas; The calculation process of the balanced responsibility area load and new energy output model includes the following steps: S201: Recurrent Neural Network Computation , in are the parameters to be trained, Represents the output of the recurrent neural network; S202: Random Sampling , using an implicit quantile network to predict and .
[0007] Further, the implicit quantile network prediction includes the steps; Compute using implicit quantile network ,in , are the parameters to be trained, Represents the implicit quantile network output, n represents the network order, and is calculated using a feedforward neural network ,in are the parameters to be trained.
[0008] Further, step S3 includes: S301: According to steps S201-S202, obtain prediction data and , calculate the quantile loss function L; S302: Update parameters based on gradient descent method , is the update step size, are the recurrent neural network and implicit quantile network parameters, Gradient operator; S303: Select different t , repeat S301.
[0009] Furthermore, the quantile loss function L is calculated as: ; in, For calculation and The random number sampled at and is the corresponding true value.
[0010] Further, step S4 includes: S401: From the earliest known moment of historical data t start; S402: If , Known, then , ; Otherwise, use the predicted data , ; S403: Recurrent Neural Network Computation ,in are the parameters to be trained, Represents the output of the recurrent neural network; S404: Random Sampling ,calculate: , represents the implicit quantile network output, and n represents the network order; S405: Feedforward Neural Network Calculation ,in For the parameters to be trained, the predicted load and new energy output data are obtained; S406: If End time T Then proceed to the next step; otherwise, , return to S402; S407: Repeat S401 to sample different , and obtain various prediction results and , and calculate the corresponding confidence interval.
[0011] A balancing capacity assessment system for a balancing responsibility area is used to implement the above-mentioned balancing capacity assessment method for a balancing responsibility area, including a balancing responsibility area load and new energy output model of a recurrent neural network. The balancing responsibility area load and new energy output model of the recurrent neural network includes a recurrent neural network, an implicit quantile network and a quantile loss function module. The input of the recurrent neural network is the historical load and new energy output data of the balancing responsibility area, and the input end of the implicit quantile network is connected to the output end of the recurrent neural network.
[0012] The advantages of the present invention are: It can model the load and new energy output change trends based on the regional historical load and new energy output data. The load and new energy output model constructed can output a variety of possible load and new energy output changes through quantile sampling, providing a reliable reference for the assessment of balancing responsibility and balancing capacity. Compared with traditional statistical modeling methods, it improves the utilization rate of historical data.
[0013] Furthermore, the constructed load and renewable energy output model combines a recurrent neural network and an implicit quantile network. The recurrent neural network can effectively utilize historical information and improve computational efficiency when the network parameters are small; the implicit quantile network can model the probability distribution of load and renewable energy output to obtain more comprehensive prediction results. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is a flow chart of a method for evaluating the balancing ability of a balancing responsibility area according to an embodiment of the present invention; Figure 2 Schematic diagram of the load and new energy output model structure of an embodiment of the present invention. DETAILED DESCRIPTION
[0015] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0016] This embodiment proposes a method for evaluating the balancing ability of a balancing responsibility area. Figure 1 , including the following steps: S1: Obtain historical load and renewable energy output data in the balancing responsibility area and construct a data set.
[0017] Load data should include accurate time and the regional load at that time , new energy output data should include accurate time and the new energy of the moment It should be noted that the time of the above two data should correspond one to one. T Indicates the total amount of historical data.
[0018] Optionally, the data set may include loads and renewable energy outputs in different balancing responsibility areas, and include a discrete variable area label a in the corresponding data, so that the established model can be adapted to different balancing responsibility areas.
[0019] S2: Construct a model for balancing the load in the responsible area and the output of new energy sources based on a recurrent neural network.
[0020] The load balancing responsibility area and new energy output model should be based on the historical data of regional load at time t. and historical data of new energy at time t Obtaining forecast data and . Further, according to the predicted and Predictable and .
[0021] Optionally, the load and renewable energy output model for the balanced responsibility area can also predict the load and renewable energy output conditions of different balanced responsibility areas according to the regional label a of the data.
[0022] S201: Recurrent Neural Network Computation , in are the parameters to be trained.
[0023] S202: Random Sampling , using an implicit quantile network to predict and .
[0024] Specifically, the calculation steps of the implicit quantile network are: calculation ,in , are the parameters to be trained, Represents the implicit quantile network output, and n represents the network order. Calculated using a feedforward neural network ,in are the parameters to be trained.
[0025] The above steps can be done by Figure 2 The system implementation shown includes a balanced responsibility area load and new energy output model of a recurrent neural network, wherein the balanced responsibility area load and new energy output model of the recurrent neural network includes a recurrent neural network, an implicit quantile network and a quantile loss function module, wherein the input of the recurrent neural network is the historical load and new energy output data of the balanced responsibility area, and the input end of the implicit quantile network is connected to the output end of the recurrent neural network.
[0026] S3: Based on the data set constructed in S1, train the parameters to be trained in the load and new energy output model described in S2.
[0027] S301: Follow the calculation process of S201-S203 to get the prediction and , calculate the quantile loss function .
[0028] The quantile loss function is in, For calculation and The random number sampled at and is the corresponding true value.
[0029] S302: Update parameters based on gradient descent method , is the update step size, are the recurrent neural network and implicit quantile network parameters, Gradient operator; S303: Select different t , repeat S301.
[0030] S4: Input the historical load of the balancing responsibility area and the output data of new energy into the model trained in S3, and output the predicted load and output data of the balancing responsibility area.
[0031] S401: From the earliest known moment of historical data t start; S402: If , Known, then , ; Otherwise, use the predicted data , ; S403: Recurrent Neural Network Computation ,in are the parameters to be trained, Represents the output of the recurrent neural network; S404: Random Sampling ,calculate: , represents the implicit quantile network output, and n represents the network order; S405: Feedforward Neural Network Calculation ,in For the parameters to be trained, the predicted load and new energy output data are obtained; S406: If End time T Then proceed to the next step; otherwise, , return to S402; S407: Repeat S401 to sample different , and obtain various prediction results and , and calculate the corresponding confidence interval.
[0032] S5: Based on the forecast results obtained in S4, find the data with the largest supply-demand gap and use its confidence interval as the balancing capacity indicator of the balancing responsibility area.
[0033] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for evaluating the balancing ability of a balancing responsibility area, characterized in that: The following steps are involved: S1: Obtain historical load and new energy output data of the balancing responsibility area and construct a data set; S2: Construct a model for balancing the load in the responsible area and the output of new energy sources based on a recurrent neural network; S3: Based on the data set constructed in S1, train the parameters to be trained in the load and new energy output model described in S2; S4: Input the historical load of the balancing responsibility area and the output data of new energy into the model trained in S3, and output the predicted load and output data of the balancing responsibility area; S5: Based on the forecast results obtained in S4, find the data with the largest supply-demand gap and use its confidence interval as the balancing capacity indicator of the balancing responsibility area.
2. The method for evaluating the balancing ability of a balancing responsibility area according to claim 1, characterized in that: The step S2 includes the following steps: The load and new energy output model of the balanced responsibility area obtains the forecast data based on the historical load and new energy output data of the balanced responsibility area, and the historical data includes the regional load at time t. and new energy , the forecast data is the load forecast data and new energy forecast data , and then according to the predicted data and predict and ,in, T Indicates the total amount of historical data; The model for balancing the load in the responsible area and the output of new energy sources is based on the regional labels of the data. , predict the load and renewable energy output in different balancing responsibility areas; The calculation process of the balanced responsibility area load and new energy output model includes the following steps: S201: Recurrent Neural Network Computation , in are the parameters to be trained, Represents the output of the recurrent neural network; S202: Random Sampling , using an implicit quantile network to predict and .
3. The method for evaluating the balancing ability of a balancing responsibility area according to claim 2, characterized in that: The implicit quantile network prediction comprises the steps of: Compute using implicit quantile network ,in , are the parameters to be trained, Represents the implicit quantile network output, n represents the network order, and is calculated using a feedforward neural network ,in are the parameters to be trained.
4. The method for evaluating the balancing ability of a balancing responsibility area according to claim 3, characterized in that: The step S3 comprises: S301: According to steps S201-S202, obtain prediction data and , calculate the quantile loss function L; S302: Update parameters based on gradient descent method , is the update step size, are the recurrent neural network and implicit quantile network parameters, Gradient operator; S303: Select different t , repeat S301.
5. The method for evaluating the balancing ability of a balancing responsibility area according to claim 4, characterized in that: The quantile loss function L is calculated as follows: ； in, For calculation and The random number sampled at and is the corresponding true value.
6. The method for evaluating the balancing ability of a balancing responsibility area according to claim 1, characterized in that: The step S4 comprises: S401: From the earliest known moment of historical data t start; S402: If , Known, then , ; Otherwise, use the predicted data , ; S403: Recurrent Neural Network Computation ,in are the parameters to be trained, Represents the output of the recurrent neural network; S404: Random Sampling ,calculate: , represents the implicit quantile network output, and n represents the network order; S405: Feedforward Neural Network Calculation ,in For the parameters to be trained, the predicted load and new energy output data are obtained; S406: If End time T Then proceed to the next step; otherwise, , return to S402; S407: Repeat S401 to sample different , and obtain various prediction results and , and calculate the corresponding confidence interval.
7. A balancing capacity assessment system for a balancing responsibility area, used to implement the balancing capacity assessment method for a balancing responsibility area according to any one of claims 1 to 6, characterized in that: The invention comprises a balanced responsibility area load and new energy output model of a recurrent neural network, wherein the balanced responsibility area load and new energy output model of the recurrent neural network comprises a recurrent neural network, an implicit quantile network and a quantile loss function module, wherein the input of the recurrent neural network is the historical load and new energy output data of the balanced responsibility area, and the input end of the implicit quantile network is connected to the output end of the recurrent neural network.