CN117436927A - Virtual power plant peak shaving market price prediction method and system - Google Patents

Virtual power plant peak shaving market price prediction method and system Download PDF

Info

Publication number
CN117436927A
CN117436927A CN202311764691.8A CN202311764691A CN117436927A CN 117436927 A CN117436927 A CN 117436927A CN 202311764691 A CN202311764691 A CN 202311764691A CN 117436927 A CN117436927 A CN 117436927A
Authority
CN
China
Prior art keywords
model
prediction
characteristic
analysis
peak
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311764691.8A
Other languages
Chinese (zh)
Inventor
汪敏
周立明
严妍
项伟
马文婧
时舜
杨春宇
宋夏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Cape Cloud Information Technology Co ltd
Original Assignee
Cape Cloud Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Cape Cloud Information Technology Co ltd filed Critical Cape Cloud Information Technology Co ltd
Priority to CN202311764691.8A priority Critical patent/CN117436927A/en
Publication of CN117436927A publication Critical patent/CN117436927A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0206—Price or cost determination based on market factors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Theoretical Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Evolutionary Computation (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Tourism & Hospitality (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application discloses a virtual power plant peak shaving market price prediction method and system. Firstly, analyzing characteristic contribution degree and screening input data; then predicting peak shaving market price by using XGBoost model, periodically updating model super-parameters, and evaluating prediction accuracy by MAPE indexes; and finally, predicting the peak-to-peak market price by using the trained XGBoost model. According to the invention, the peak shaving market price prediction model is constructed, screened prediction data is taken as input, a precondition is provided for real-time peak shaving quotation, and the virtual power plant quotation is ensured to meet the peak shaving dispatching electricity price requirement.

Description

Virtual power plant peak shaving market price prediction method and system
Technical Field
The invention relates to the field of power peak shaving, in particular to a virtual power plant peak shaving market price prediction method and system.
Background
Currently, in the power market, commonly used price prediction models include time series based analysis, statistical based models, machine learning based algorithms, and the like. The prediction model based on time sequence analysis is a traditional method, the model predicts based on historical data and time sequence rules, and can predict the trend of the price of the electric power market, but other factors and variables in the market are not considered, and the prediction result has certain limitation. The prediction model based on the statistical model can model a plurality of variables, well considers other factors in the market, can provide more accurate and practical prediction results, but has higher data requirements, and needs more historical data and stronger statistical analysis capability. The prediction model based on the machine learning algorithm is a newer prediction method, and price prediction can be realized under the condition of considering a large number of market variables through data mining and machine learning technology, so that the accuracy is higher, but the requirements on the data processing capacity and algorithm selection are higher. In addition to the above methods, some studies utilize multiple models and methods to perform hybrid prediction, improving the stability and accuracy of the prediction, and the method is more robust than a single method, and can provide reliable prediction results under different conditions. However, the problems with price prediction are not negligible. For example: the model is complex due to excessive characteristics, the model precision is greatly influenced by super parameters, the data is easier to be outdated, the market adaptability to continuous changes is poor, and the like. In summary, the problems of the prior art mainly include the following aspects:
(1) The price factors affecting the peak shaving market are more, the influence of different factors is greatly different, the characteristics are selected according to the manual experience in the prior art, contribution degree analysis is not carried out on the characteristics, the model is complicated due to the fact that all characteristic data are input, and the efficiency of the model is reduced.
(2) The complexity of the partial peak shaving market price prediction model is higher, and the training process is longer. In addition, the model hyper-parameters can have a great influence on the predicted performance and effect, a certain experience and skill are required for optimizing the hyper-parameters, the hyper-parameters are not modified after the existing model is trained, and the predicted effect is reduced due to the fact that the hyper-parameters are fixed over time.
(3) With the gradual development of peak shaving markets, the market environment and competition patterns are continuously changed, the number of users participating in peak shaving is increased, and the peak shaving electric quantity is also changed. The historical data collected during model training cannot reflect the characteristics of the real-time change market, and cannot accurately identify the market development.
Disclosure of Invention
Based on the above, the embodiment of the application provides a virtual power plant peak shaving market price prediction method and a virtual power plant peak shaving market price prediction system, which utilize an extreme gradient lifting tree technology to construct a peak shaving market price prediction model, take day-ahead prediction data as input, provide preconditions for real-time peak shaving quotation in the day of VPP (Virtual Power Plant, VPP), and ensure that the VPP quotation meets the peak shaving dispatching electricity price requirement.
In a first aspect, a virtual power plant peak shaving market price prediction method is provided, the method comprising:
determining each feature for model input through contribution analysis, and acquiring historical data corresponding to each feature; the contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
constructing an XGBoost model based on the extreme gradient lifting tree, and training the constructed XGBoost model by utilizing the acquired historical data; the XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine predicted price;
and carrying out online prediction on the peak-to-peak market price by using the trained XGBoost model.
Optionally, determining the respective features for performing the model input by model analysis includes:
quantifying the contribution degree of each feature to be determined by using an XGBoost model;
sequencing the contribution degree of each feature to be determined from high to low;
and selecting a preset number of high-contribution to-be-determined features in the contribution sequence as the features for model input.
Optionally, constructing the XGBoost model based on the extreme gradient lift tree includes passing through the formula:
determining a model predictive value of the XGBoost model, wherein,the model predicted value of the ith sample, K is the number of trees; f is the collection space of the tree; x is x i A feature vector representing an i-th sample; f (f) k Corresponding to the leaf weight of the kth independent tree structure.
Optionally, the loss function L of the XGBoost model includes two parts, namely:
wherein the first partFor predictive value +.>And target true value y i Training errors between; second partIs the sum of the complexity of the tree, is a regularization term used to control the complexity of the model, i.e
；
And->And respectively representing penalty coefficients of the model, wherein T is a super parameter representing the maximum depth of the decision tree, and w is a variable representing a weight parameter in the model.
Optionally, the sum of the tree complexity is added to the delta function f for each round in the sequence minimization process t (x i ) The loss function is reduced to the greatest extent as far as possible, and the objective function of the t-th round is specifically as follows:
where n is the actual number of samples used to characterize the total training round and i is the sample number.
Optionally, obtaining peak shaving price gains through greedy methods of different peak shaving state tree structures and performing feature evaluation to determine a predicted price, including:
the sub-tree is partitioned by a greedy algorithm and the feasible partitioning points are enumerated, i.e. new partitioning is added to the existing leaves each time and the maximum gain thus obtained is calculated.
Optionally, training the constructed XGBoost model by using the acquired historical data, further includes:
and periodically updating the model hyper-parameters of the XGBoost model, and evaluating the model prediction accuracy through root mean square error and average absolute percentage error.
Optionally, the estimating the model prediction accuracy by the root mean square error and the average absolute percentage error includes:
according to the formula
The root mean square error RMSE and the mean absolute percentage error MAPE are determined, wherein,is the model predictive value of the ith sample, y i Is the actual value and n is the actual number of samples.
Optionally, the on-line prediction of the peak-to-peak market price using the trained XGBoost model includes:
sampling the latest data in the prediction process, and arranging the latest data into a characteristic data format required by a model;
updating the characteristic data regularly;
performing market analysis by using time-series analysis; wherein, the method specifically comprises trend analysis and periodicity analysis of market price;
and inputting the latest characteristic data into the model, and obtaining a prediction result through a prediction function of the model.
In a second aspect, a virtual power plant peak shaver market price prediction system is provided, the system comprising:
the analysis module is used for determining each characteristic used for model input through contribution degree analysis and obtaining historical data corresponding to each characteristic; the contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
the construction module is used for constructing an XGBoost model based on the extreme gradient lifting tree and training the constructed XGBoost model by utilizing the acquired historical data; the XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine predicted price;
and the prediction module is used for carrying out online prediction on the peak-to-peak market price by utilizing the XGBoost model after training.
The beneficial effects that technical scheme that this application embodiment provided include at least:
(1) The method for analyzing the characteristic contribution degree is used for removing redundant characteristics with weak correlation with the peak shaving market price, extracting characteristics with strong correlation with the price, and using the characteristics in model training, so that the prediction accuracy of the model is improved, and the influence of excessive characteristics on the model is reduced.
(2) The model prediction efficiency is guaranteed by using a simple and efficient machine learning algorithm and an automatic parameter adjustment technology, the super parameters of the peak shaving market price prediction model are optimized, the manual parameter adjustment time is reduced, the model accuracy is improved, the super parameters are updated in a rolling mode, the market environment change can be adapted, and the precision reduction caused by untimely adjustment of the super parameters is avoided.
(3) Compared with the traditional data acquisition and prediction, for peak shaving markets with larger variation, a real-time prediction method is established, the model is updated in time, the change of the peak shaving market is adapted, and the practicability of the model is enhanced. The latest market information and data can be timely acquired by online data acquisition, so that timeliness of a prediction model is guaranteed, data outdated caused by prediction by adopting historical data can be avoided, and prediction accuracy is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It will be apparent to those skilled in the art from this disclosure that the drawings described below are merely exemplary and that other embodiments may be derived from the drawings provided without undue effort.
FIG. 1 is a flowchart of a method for predicting peak shaving market price of a virtual power plant according to an embodiment of the present application;
fig. 2 is a block diagram of a virtual power plant peak shaving market price prediction system according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
In the description of the present invention, the terms "comprises," "comprising," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements but may include other steps or elements not expressly listed but inherent to such process, method, article, or apparatus or steps or elements added based on further optimization of the inventive concept.
According to the method, a unique feature screening method is adopted, feature contribution degree analysis is carried out, a series of feature parameters are screened and obtained at first to serve as an input feature set for model training, data are prevented from being fitted excessively, then features before importance ranking are extracted and input into a peak shaving price prediction model for training, a price prediction value is obtained, a prediction error evaluation index is built to integrate a peak shaving market price actual value into the model for cyclic training, meanwhile, feature data screening adopts a method of periodically updating super parameters to improve prediction accuracy of the model, and the model is more suitable for market environment changes. Specifically, please refer to fig. 1, which illustrates a flowchart of a virtual power plant peak shaving market price prediction method according to an embodiment of the present application, the method may include the following steps:
and step 101, determining each feature for model input through contribution analysis, and acquiring historical data corresponding to each feature.
The contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
in the step, the characteristic contribution degree is analyzed, and the input data is screened. Specifically:
in the process of realizing prediction, firstly, contribution degree analysis is carried out on the characteristics of the selected data, and input data screening is carried out, so that the performance of a machine learning model is improved. In actual operation, feature engineering is performed by combining manual experience and model analysis. The artificial experience is based on domain knowledge and professional experience, and features which are significant to the problem are selected and constructed through understanding data and grasping domain characteristics. The model analysis is based on the training process of a machine learning model, and features are further screened and optimized by evaluating and analyzing the correlation and the influence degree between the features and the target variables.
In the peak shaving market, the characteristics of the photovoltaic and charging pile equipment are selected through manual experience. Specifically, the following features may be selected by human experience: photovoltaic installed capacity, radiation, temperature, number of charging piles, type of charging piles, etc. Data processing of different forms is carried out on various characteristics respectively, as shown in table 1.
Table 1 characteristic introduction table
Contribution analysis refers to a method of evaluating and analyzing the degree of contribution of data features in a machine learning model. The influence degree of each feature on the model prediction result can be known through contribution degree analysis, so that the importance and the value of the feature are judged, the importance and the contribution degree of the feature can be known, the selection and the construction of the feature are optimized, and the performance of the machine learning model is improved.
In an embodiment of the present application, determining, through model analysis, respective features for performing model input includes: quantifying the contribution degree of each feature to be determined by using an XGBoost model; sequencing the contribution degree of each feature to be determined from high to low; and selecting a preset number of high-contribution to-be-determined features in the contribution sequence as the features for model input. Specifically:
the XGBoost model is utilized to quantify the contribution degree of each feature, a quantized result sample graph generated by encoding all the features can be selected from the quantized result sample graph, the features which are relatively strong in relation to electricity price are selected, and input features are selected according to feature importance ranking so as to reduce the complexity of the model. The characteristic input of 60% before characteristic contribution degree sequencing is generally selected, and the characteristic which has weak relevance or redundancy with electricity price is removed. The characteristic contribution degree analysis is used for carrying out characteristic processing on the data, so that extraction of key core influence indexes is enhanced, meanwhile, excessively redundant variable influence in the model is reduced, and the prediction capability of the model is optimized and improved.
And 102, constructing an XGBoost model based on the extreme gradient lifting tree, and training the constructed XGBoost model by using the acquired historical data.
The XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine the predicted price.
In an alternative embodiment of the present application, the method further includes periodically updating model hyper-parameters of the XGBoost model, and evaluating model prediction accuracy through root mean square error and average absolute percentage error.
In the step, the XGBoost model is used for predicting peak shaving market price, the model super-parameters are updated regularly, and MAPE indexes evaluate prediction accuracy.
For peak shaving market price prediction, a XGBoost (eXtreme Gradient Boosting) model is adopted for prediction, XGBoost is a machine learning algorithm based on a gradient lifting tree, and the model is improved and expanded from the gradient lifting tree model and is widely applied to various prediction problems including market price prediction. In the scheme, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out. The specific mode is as follows:
（1）
determining a model predictive value of the XGBoost model by the formula (1), wherein,the model predicted value of the ith sample, K is the number of trees; f is the collection space of the tree; x is x i A feature vector representing an i-th sample; f (f) k Corresponding to the leaf weight of the kth independent tree structure.
The XGBoost model loss function L consists of two parts:
（2）
wherein the first part is a predicted valueAnd target true value y i Training errors between; the second part is the sum of the complexity of the tree, which is a regularization term for controlling the complexity of the model, namely:
（3）
in the method, in the process of the invention,and->And representing a penalty coefficient for the model, wherein T is a super-parameter representing the maximum depth of the decision tree, and w is a variable representing a weight parameter in the model.
Equation (3) delta function added per round during sequence minimizationMinimizing the loss function as much as possible. The objective function for the t-th round can be written as:
（4）
the second order taylor expansion is used for equation (4) to approximate the objective function. Defining the sample set in each leaf of the j-th tree as. Wherein (1)>，/>The first and second derivatives of the loss function, respectively. This can be achieved by:
（5）
definition of the definitionAccumulated value representing the first derivative of the loss function, < +.>And representing the accumulated value of the second derivative of the loss function, G and H being used to represent approximations of the first and second derivatives of the delta function of the j-th tree. The method can obtain:
（6）
the T parameter, the I parameter and the variable w are selected, the optimal parameter is selected through a random search mode, and values are randomly selected in a defined variable range to search the optimal variable setting. The random search can more efficiently search for the best value than the grid search. The bias derivative of the variable w of the weight or parameter in the model is available:
（7）
the weight is brought into an objective function to obtain:
（8）
the smaller the loss function, the better the representation model, the partitioning of the subtree by greedy algorithm is performed and the feasible partitioning points are enumerated, i.e. each time a new partitioning is added to the existing leaf, and the maximum gain thus obtained is calculated. Gain ofThe calculation mode of (2) is as follows:
（9）
wherein, items 1 and 2 respectively represent gains generated after splitting of left and right subtrees; item 3 is the gain without subtree splitting.
In the process of using XGBoost model to realize peak shaving market price prediction, in order to evaluate accuracy of the prediction model, some indexes are needed to measure difference between the prediction result and the actual result. Common prediction error assessment indicators include Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), etc. Wherein Root Mean Square Error (RMSE) is a common indicator representing the difference between the predicted and actual results, it is the root mean square value of the predicted error that is calculated as follows:
（10）
in the method, in the process of the invention,is the model predictive value of the ith sample, y i Is the actual value and n is the actual number of samples.
The Mean Absolute Percentage Error (MAPE) is an indicator that represents the percentage difference between the predicted outcome and the actual outcome. It calculates the average percentage of the absolute value of the prediction error to the actual value as follows:
（11）
the smaller these indices, the smaller the difference between the predicted and actual results, the better the predictive power of the model. During model selection and parameter tuning, the accuracy of the model can be evaluated according to the performance of the indexes, and the model with the best performance can be selected or the super-parameters of the model can be adjusted to improve the prediction capability.
XGBoost has many adjustable hyper-parameters including learning rate, number of trees, depth of tree, etc. In order to find the optimal super-parameter combination, grid searching, random searching and other methods can be used for parameter tuning. The super parameters are updated regularly, so that the time and energy of manual parameter adjustment are reduced, the parameter adjustment efficiency and model accuracy are improved, and the prediction capability of the model is always consistent with market change. The market price is affected by various factors, such as seasonal changes, economic factors and the like, so that the prediction accuracy of the model can be improved by timely updating the super parameters.
And step 103, online predicting the peak-to-peak market price by using the trained XGBoost model.
The online prediction process is specifically implemented in this step, specifically:
on-line prediction refers to predicting the peak-to-peak market price by using a trained XGBoost model under real-time or near real-time conditions. In order to perform online prediction, the following steps are required:
(1) Online sampling data: in order to ensure the accuracy of the prediction, the latest data needs to be sampled at the time of prediction. Market price data is collected periodically and is organized into a characteristic data format required by the model.
(2) Updating the characteristic data: over time, market prices may change due to a variety of factors. Therefore, the feature data needs to be updated periodically to reflect the latest change in the market. The characteristic data may include historical price data, market demand data, seasonal factors, and the like. Updating the feature data helps to improve the predictive power of the model.
(3) Analysis of market changes: analysis of market changes is required before on-line predictions can be made. The scheme is used for knowing the trend and the periodical change of the market price by using a time sequence analysis method such as trend analysis, periodical analysis and the like. These analysis results can be used as inputs to a predictive model to help the model better capture market changes.
(4) On-line prediction: and inputting the latest characteristic data into the model, and obtaining a prediction result through a prediction function of the model.
The frequency of online predictions can be adjusted as needed, and the application can set predictions to be made hourly, daily or weekly. Updating the feature data and retraining the model on a regular basis is also an important step in ensuring prediction accuracy. By continuously updating the data and the model, the prediction result can be more accurate, and a decision maker can be helped to make better decisions.
As shown in fig. 2, the embodiment of the application also provides a virtual power plant peak shaving market price prediction system. The system comprises:
the analysis module is used for determining each characteristic used for model input through contribution degree analysis and obtaining historical data corresponding to each characteristic; the contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
the construction module is used for constructing an XGBoost model based on the extreme gradient lifting tree and training the constructed XGBoost model by utilizing the acquired historical data; the XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine predicted price;
and the prediction module is used for carrying out online prediction on the peak-to-peak market price by utilizing the XGBoost model after training.
The virtual power plant peak shaving market price prediction system provided in the embodiment of the present application is used to implement the virtual power plant peak shaving market price prediction method, and specific limitation regarding the virtual power plant peak shaving market price prediction system may be referred to the limitation regarding the virtual power plant peak shaving market price prediction method hereinabove, which is not described herein. The various parts of the virtual power plant peak shaving market price prediction system described above may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or independent of a processor in the device, or may be stored in software in a memory in the device, so that the processor may call and execute operations corresponding to the above modules.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the claims. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.

Claims (10)

1. A method for predicting peak shaver market prices of a virtual power plant, the method comprising:
determining each feature for model input through contribution analysis, and acquiring historical data corresponding to each feature; the contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
constructing an XGBoost model based on the extreme gradient lifting tree, and training the constructed XGBoost model by utilizing the acquired historical data; the XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine predicted price;
and carrying out online prediction on the peak-to-peak market price by using the trained XGBoost model.
2. The method of claim 1, wherein determining individual features for model input by model analysis comprises:
quantifying the contribution degree of each feature to be determined by using an XGBoost model;
sequencing the contribution degree of each feature to be determined from high to low;
and selecting a preset number of high-contribution to-be-determined features in the contribution sequence as the features for model input.
3. The method of claim 1, wherein constructing the XGBoost model based on the extreme gradient lifted tree comprises:
determining a model predictive value of an XGBoost model, wherein +.>The model predicted value of the ith sample, K is the number of trees; f is the collection space of the tree; x is x i A feature vector representing an i-th sample; f (f) k Corresponding to the leaf weight of the kth independent tree structure.
4. A method according to claim 3, characterized in that the loss function L of the XGBoost model comprises two parts, namely:
wherein the first part->For predictive value +.>And target true value y i Training errors between; second part->Is the sum of the complexity of the tree, is a regularization term used to control the complexity of the model, i.e
；
And->And respectively representing penalty coefficients of the model, wherein T is a super parameter representing the maximum depth of the decision tree, and w is a variable representing a weight parameter in the model.
5. The method of claim 4, wherein the sum of the tree complexity is a delta function f added for each round during the sequence minimization process t (x i ) The loss function is reduced to the greatest extent as far as possible, and the objective function of the t-th round is specifically as follows:
where n is the actual number of samples used to characterize the total training round and i is the sample number.
6. The method of claim 1, wherein determining peak shaver price gains by greedy methods of different peak shaver state tree structures and performing feature evaluation to determine a predicted price comprises:
the sub-tree is partitioned by a greedy algorithm and the feasible partitioning points are enumerated, i.e. new partitioning is added to the existing leaves each time and the maximum gain thus obtained is calculated.
7. The method of claim 1, wherein training the constructed XGBoost model using the acquired historical data further comprises:
and periodically updating the model hyper-parameters of the XGBoost model, and evaluating the model prediction accuracy through root mean square error and average absolute percentage error.
8. The method of claim 7, wherein said estimating model prediction accuracy by root mean square error and mean absolute percentage error comprises:
according to the formula
Determining a root mean square error RMSE and a mean absolute percentage error MAPE, wherein +.>Is the model predictive value of the ith sample, y i Is the actual value and n is the actual number of samples.
9. The method of claim 1, wherein online predicting the on-peak market price using the trained XGBoost model comprises:
sampling the latest data in the prediction process, and arranging the latest data into a characteristic data format required by a model;
updating the characteristic data regularly;
performing market analysis by using time-series analysis; wherein, the method specifically comprises trend analysis and periodicity analysis of market price;
and inputting the latest characteristic data into the model, and obtaining a prediction result through a prediction function of the model.
10. A virtual power plant peak shaver market price prediction system, the system comprising:
the analysis module is used for determining each characteristic used for model input through contribution degree analysis and obtaining historical data corresponding to each characteristic; the contribution degree analysis comprises manual experience analysis and model analysis, and each characteristic at least comprises a photovoltaic characteristic, a charging pile characteristic, a weather characteristic, an electricity price characteristic and corresponding time;
the construction module is used for constructing an XGBoost model based on the extreme gradient lifting tree and training the constructed XGBoost model by utilizing the acquired historical data; the XGBoost model is used for predicting peak shaving market price, peak shaving price gain is obtained through greedy methods of different peak shaving state tree structures, and feature evaluation is carried out to determine predicted price;
and the prediction module is used for carrying out online prediction on the peak-to-peak market price by utilizing the XGBoost model after training.
CN202311764691.8A 2023-12-21 2023-12-21 Virtual power plant peak shaving market price prediction method and system Pending CN117436927A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311764691.8A CN117436927A (en) 2023-12-21 2023-12-21 Virtual power plant peak shaving market price prediction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311764691.8A CN117436927A (en) 2023-12-21 2023-12-21 Virtual power plant peak shaving market price prediction method and system

Publications (1)

Publication Number Publication Date
CN117436927A true CN117436927A (en) 2024-01-23

Family

ID=89548406

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311764691.8A Pending CN117436927A (en) 2023-12-21 2023-12-21 Virtual power plant peak shaving market price prediction method and system

Country Status (1)

Country Link
CN (1) CN117436927A (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112308335A (en) * 2020-11-12 2021-02-02 南方电网能源发展研究院有限责任公司 Short-term electricity price prediction method and device based on xgboost algorithm

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112308335A (en) * 2020-11-12 2021-02-02 南方电网能源发展研究院有限责任公司 Short-term electricity price prediction method and device based on xgboost algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
史佳琪 等: "基于串行–并行集成学习的高峰负荷预测方法", 《中国电机工程学报》, vol. 40, no. 14, 20 July 2020 (2020-07-20), pages 4464 - 4472 *

Similar Documents

Publication Publication Date Title
CN110969290B (en) Runoff probability prediction method and system based on deep learning
CN112016734A (en) Stack type self-coding multi-model load prediction method and system based on LSTM
CN110163429B (en) Short-term load prediction method based on similarity day optimization screening
CN108694470B (en) Data prediction method and device based on artificial intelligence
CN110689190A (en) Power grid load prediction method and device and related equipment
CN110751318A (en) IPSO-LSTM-based ultra-short-term power load prediction method
CN111932024B (en) Energy load prediction method and device, computer equipment and storage medium
CN106600037B (en) Multi-parameter auxiliary load prediction method based on principal component analysis
CN112633604A (en) Short-term power consumption prediction method based on I-LSTM
CN111525587B (en) Reactive load situation-based power grid reactive voltage control method and system
CN111027629A (en) Power distribution network fault outage rate prediction method and system based on improved random forest
CN113901977A (en) Deep learning-based power consumer electricity stealing identification method and system
CN111079989A (en) Water supply company water supply amount prediction device based on DWT-PCA-LSTM
CN116316599A (en) Intelligent electricity load prediction method
CN111738477A (en) Deep feature combination-based power grid new energy consumption capability prediction method
CN114169434A (en) Load prediction method
CN115495991A (en) Rainfall interval prediction method based on time convolution network
CN115222138A (en) Photovoltaic short-term power interval prediction method based on EEMD-LSTM microgrid
CN112949207A (en) Short-term load prediction method based on improved least square support vector machine
CN112184412A (en) Modeling method, device, medium and electronic equipment of credit rating card model
CN116245019A (en) Load prediction method, system, device and storage medium based on Bagging sampling and improved random forest algorithm
CN111222669A (en) Overhead line cost prediction method and device
CN114118401A (en) Neural network-based power distribution network flow prediction method, system, device and storage medium
CN112348236B (en) Abnormal daily load demand prediction system and method for intelligent electricity consumption monitoring terminal
CN117113159A (en) Deep learning-based power consumer side load classification method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination