Search

CN-121998722-A - Time-of-use electricity price generation method based on new energy time sequence output prediction

CN121998722ACN 121998722 ACN121998722 ACN 121998722ACN-121998722-A

Abstract

The invention discloses a time-sharing electricity price generation method based on new energy time sequence output prediction, which comprises the steps of integrating and preprocessing historical load data and new energy output data to obtain an original data set, sequentially carrying out local dimension reduction and global dimension reduction on the original data set to obtain a dimension reduction data set, clustering the dimension reduction data set to obtain a plurality of load categories, respectively training an LSTM (least squares) model for each load category, outputting a load prediction result of a future period, constructing a time-sharing electricity price period division model, solving the time-sharing electricity price period division model under a preset duration constraint based on the load prediction result to obtain a dynamic period division scheme, generating a correction coefficient according to real-time load data, the load prediction result and new energy output deviation, and determining a final time-sharing electricity price based on the dynamic period division scheme, the basic electricity price of each period and the correction coefficient. The invention realizes the dynamic electricity price generation which accurately reflects the supply and demand states of the power grid, and optimizes the power resource allocation.

Inventors

  • FU BIN
  • YANG MING
  • ZHANG ZHONG

Assignees

  • 重庆大学

Dates

Publication Date
20260508
Application Date
20260202

Claims (10)

  1. 1. A time-of-use electricity price generation method based on new energy time sequence output prediction is characterized by comprising the following steps: integrating and preprocessing historical load data and new energy output data to obtain an original data set; sequentially performing local dimension reduction and global dimension reduction on the original data set to obtain a dimension reduction data set; clustering the dimension reduction data sets to obtain a plurality of load categories; Respectively training an LSTM model aiming at each load category, and outputting a load prediction result of a future period; constructing a time-of-use electricity price time division model, and solving the time-of-use electricity price time division model under the constraint of preset duration based on the load prediction result to obtain a dynamic time division scheme; Generating a correction coefficient according to the real-time load data, the load prediction result and the new energy output deviation; And determining the final time-sharing electricity price based on the dynamic time-interval division scheme, the basic electricity price of each time interval and the correction coefficient.
  2. 2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The integrating and preprocessing of the historical load data and the new energy output data comprises the following steps: And collecting historical load data and new energy output data comprising weather, season and holiday characteristics, complementing the missing values, eliminating the abnormal values, and unifying the time dimension into fixed intervals to form an original data set.
  3. 3. The method of claim 1, wherein the step of determining the position of the substrate comprises, Sequentially performing local dimension reduction and global dimension reduction on the original data set to obtain a dimension reduction data set, wherein the dimension reduction data set comprises: The method comprises the steps of calculating sample point distances through a shortest path algorithm to replace traditional Euclidean distances in an LLE algorithm to obtain an improved LLE algorithm, adopting the improved LLE algorithm to carry out local dimension reduction on an original data set, mapping the data set subjected to the local dimension reduction to a high-dimensional space through a kernel function, carrying out decentralization on the mapped data, calculating a covariance matrix, solving characteristic values and characteristic vectors, sequencing the characteristic vectors from large to small according to the characteristic values, and outputting the data set subjected to the dimension reduction.
  4. 4. The method of claim 1, wherein the step of determining the position of the substrate comprises, Clustering the dimension-reduced dataset to obtain a plurality of load categories, including: And determining the optimal clustering number by using a contour coefficient method, and executing K-means clustering by using Euclidean distance as a measure, wherein data with similar characteristics are divided into the same load category.
  5. 5. The method of claim 1, wherein the step of determining the position of the substrate comprises, Respectively training an LSTM model aiming at each load category, outputting a load prediction result of a future period, and comprising the following steps: And respectively constructing an LSTM model for each load category, completing network parameter optimization of the LSTM model by using data of the corresponding category, respectively inputting clustered data into the corresponding optimized LSTM model, and outputting a load prediction result.
  6. 6. The method of claim 1, wherein the step of determining the position of the substrate comprises, Constructing a time-of-use electricity price time division model, solving the time-of-use electricity price time division model under a preset duration constraint based on the load prediction result, and obtaining a dynamic time division scheme, wherein the method comprises the following steps: The method comprises the steps of taking a load predicted value as a basis, calculating the peak membership and the valley membership of each moment through a semi-trapezoid membership function, setting the total time period duration, the minimum continuous time duration and the maximum continuous time duration as controllable constraints, constructing a time-sharing electricity price time period division model by taking the sum of membership distances in the time periods as a target, solving the time-sharing electricity price time division model, and obtaining a moment set corresponding to each time period to form a dynamic time period division scheme.
  7. 7. The method of claim 6, wherein the step of providing the first layer comprises, When solving the time-of-use electricity price time period division model, the method further comprises the following steps: And converting the nonlinear constraint into zero-integer programming by introducing zero-one auxiliary variable, and dynamically generating the auxiliary variable and constraint conditions by utilizing the global variable to realize model solving.
  8. 8. The method of claim 1, wherein the step of determining the position of the substrate comprises, The generation of the correction coefficient according to the real-time load data, the load prediction result and the new energy output deviation comprises the following steps: calculating the deviation ratio of the predicted output and the actual output of the new energy, determining a down-regulating correction coefficient according to the positive deviation ratio when the actual output is higher than the predicted output, determining an up-regulating correction coefficient according to the negative deviation ratio when the actual output is lower than the predicted output, and taking a reference value by the correction coefficient when the deviation is in a preset interval.
  9. 9. The method of claim 1, wherein the step of determining the position of the substrate comprises, The determining the final time-of-use electricity price comprises the following steps: And multiplying the predicted electricity price of the basic electricity price of each period by a correction coefficient to obtain the final time-sharing electricity price of the real-time supply and demand state.
  10. 10. A computer device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1-9.

Description

Time-of-use electricity price generation method based on new energy time sequence output prediction Technical Field The invention belongs to the technical field of power prediction, and particularly relates to a time-of-use electricity price generation method based on new energy time sequence output prediction. Background With the continuous improvement of the ratio of new energy sources (such as wind power and photovoltaic) in an electric power system, the intermittent and fluctuating characteristics of the output of the new energy sources bring serious challenges to the supply and demand balance of a power grid and the design of an electricity price mechanism. The traditional electricity price generation method is mostly dependent on a single data source, such as historical load data, lacks dynamic response to the time sequence output characteristics of new energy, and is difficult to adapt to the operation requirement of an electric power system under high-proportion new energy access. In the aspect of load prediction, the traditional model is often used for directly modeling high-dimensional load data, the influence of multidimensional features such as weather, seasons, holidays and the like is not considered, and the suitability of the traditional model for different load modes is poor, so that the prediction accuracy is insufficient, and reliable data support cannot be provided for electricity price generation. In the aspect of time-of-use electricity price time division, the existing method mostly adopts fixed time division rules, does not fully combine the dynamic characteristics of a load curve and the preference of a decision maker to the time duration of each time period, is difficult to realize the goal of peak clipping and valley filling, and cannot flexibly adapt to the load curve change caused by new energy output fluctuation. In addition, the deviation between the new energy output prediction and the actual output is common, and the traditional electricity price mechanism lacks an effective deviation correction mechanism, so that the electricity price signal is easy to be distorted. In view of the above problems in the prior art, it is needed to provide a time-of-use electricity price generation method based on new energy time-of-use output prediction. Disclosure of Invention In order to solve the technical problems, the invention provides a time-of-use electricity price generation method based on new energy time sequence output prediction. The invention provides a time-sharing electricity price generation method based on new energy time sequence output prediction, which comprises the following steps: integrating and preprocessing historical load data and new energy output data to obtain an original data set; sequentially performing local dimension reduction and global dimension reduction on the original data set to obtain a dimension reduction data set; clustering the dimension reduction data sets to obtain a plurality of load categories; Respectively training an LSTM model aiming at each load category, and outputting a load prediction result of a future period; constructing a time-of-use electricity price time division model, and solving the time-of-use electricity price time division model under the constraint of preset duration based on the load prediction result to obtain a dynamic time division scheme; Generating a correction coefficient according to the real-time load data, the load prediction result and the new energy output deviation; And determining the final time-sharing electricity price based on the dynamic time-interval division scheme, the basic electricity price of each time interval and the correction coefficient. Optionally, the integrating and preprocessing the historical load data and the new energy output data includes: And collecting historical load data and new energy output data comprising weather, season and holiday characteristics, complementing the missing values, eliminating the abnormal values, and unifying the time dimension into fixed intervals to form an original data set. Optionally, performing local dimension reduction and global dimension reduction on the original data set in sequence to obtain a dimension reduction data set, including: The method comprises the steps of calculating sample point distances through a shortest path algorithm to replace traditional Euclidean distances in an LLE algorithm to obtain an improved LLE algorithm, adopting the improved LLE algorithm to carry out local dimension reduction on an original data set, mapping the data set subjected to the local dimension reduction to a high-dimensional space through a kernel function, carrying out decentralization on the mapped data, calculating a covariance matrix, solving characteristic values and characteristic vectors, sequencing the characteristic vectors from large to small according to the characteristic values, and outputting the data set subjected to the dimension reduction. Optiona