CN116611859A - Power transaction spot price difference trend prediction method based on gradient lifting decision tree - Google Patents
Power transaction spot price difference trend prediction method based on gradient lifting decision tree Download PDFInfo
- Publication number
- CN116611859A CN116611859A CN202310896464.4A CN202310896464A CN116611859A CN 116611859 A CN116611859 A CN 116611859A CN 202310896464 A CN202310896464 A CN 202310896464A CN 116611859 A CN116611859 A CN 116611859A
- Authority
- CN
- China
- Prior art keywords
- power generation
- quotation
- data
- day
- enterprise
- 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
Links
- 238000003066 decision tree Methods 0.000 title claims abstract description 42
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000010248 power generation Methods 0.000 claims abstract description 315
- 238000011156 evaluation Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 6
- 239000000843 powder Substances 0.000 claims description 2
- 230000005611 electricity Effects 0.000 abstract description 8
- 239000000446 fuel Substances 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
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
- G06Q30/0206—Price or cost determination based on market factors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/27—Regression, e.g. linear or logistic regression
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Strategic Management (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Finance (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Accounting & Taxation (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Human Resources & Organizations (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Technology Law (AREA)
- Primary Health Care (AREA)
- Game Theory and Decision Science (AREA)
- Tourism & Hospitality (AREA)
- Probability & Statistics with Applications (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of electricity price trend prediction, and provides a power transaction spot price difference trend prediction method based on a gradient lifting decision tree, which comprises the following steps: collecting historical power generation data, power generation type related data, historical quotation data and daily quotation data; according to the historical power generation data and the power generation type related data, acquiring daily power generation influence factors of each power generation type in each period, and acquiring quotation convolution weights of each period of each power generation enterprise by combining the historical quotation data; according to the quotation convolution weight and the historical quotation data, obtaining ideal reporting value of each period and forecast quotation data of each power generation enterprise; and obtaining the quotation evaluation value of each power generation enterprise according to the historical quotation data, the predicted quotation data and the ideal reporting value, and finishing the prediction of the current price difference trend of the power transaction and the selection of the power generation enterprises. The invention aims to solve the problem that the algorithm efficiency and the prediction accuracy are affected due to the fact that the initial weak classifier of the traditional gradient lifting decision tree is rough.
Description
Technical Field
The invention relates to the technical field of electricity price trend prediction, in particular to a gradient-lifting decision tree-based power transaction spot price difference trend prediction method.
Background
The electric power spot transaction is in the electric power transaction market, and can purchase or sell electric quantity in a bargaining mode in a specific electric power transaction time; the electric power spot transaction market mainly extends around auxiliary service transactions such as medium-and-long-term, day-ahead, real-time electric energy transaction, standby, frequency modulation and the like; there are two types of price mechanisms for power spot transactions: one is to settle the price per quote of each market subject, the other is a unified price mechanism to settle the price per marginal price.
The power spot transaction market is based on real-time supply and demand relation and local market price difference to adjust the electricity price, and has the problems of monopoly and information asymmetry, so that price fluctuation is large; when a user enterprise selects a power generation enterprise to cooperate, plans a future cooperation scheme and adjusts a cooperation direction, the future market cannot be accurately predicted according to the current market, so that the user enterprise loses initiative, any power generation enterprise defines a transaction rule, and unnecessary cost waste is caused; therefore, the change trend of the price difference of the electric power spot transaction market is predicted through the gradient lifting decision tree, and meanwhile, the ideal reporting value of the initial weak classifier is corrected by combining the quotations of all power generation enterprises, so that the algorithm efficiency is improved, the accuracy of a prediction result is ensured, and the electricity cost of the user enterprises is saved.
Disclosure of Invention
The invention provides a power transaction spot price difference trend prediction method based on a gradient lifting decision tree, which aims to solve the problem that the algorithm efficiency and the prediction accuracy are affected due to the fact that the initial weak classifier of the traditional gradient lifting decision tree is rough, and adopts the following technical scheme:
one embodiment of the invention provides a method for predicting the current price difference trend of electric power transaction based on a gradient lifting decision tree, which comprises the following steps:
collecting historical power generation data, historical quotation data and daily quotation data of a plurality of power generation enterprises in the power market, and recording power generation type related data of each period;
acquiring quotation convolution weight of each period of each power generation enterprise according to daily power generation influence factors and historical quotation data of each period of each power generation type;
according to ideal report value and historical quotation data, constructing and training a gradient lifting decision tree, inputting daily power generation influence factors and daily quotation data, and outputting predicted quotation data of each power generation enterprise;
and obtaining the quotation evaluation value of each power generation enterprise according to the historical quotation data, the predicted quotation data and the ideal reporting value, and finishing the prediction of the current price difference trend of the power transaction and the selection of the power generation enterprises.
Further, the method for obtaining the quotation convolution weight of each period of each power generation enterprise specifically comprises the following steps:
acquiring daily power generation influence factors of each power generation type in each period according to historical power generation data and power generation type related data, and acquiring daily power generation amount of each power generation enterprise in each day according to the historical power generation data, namelyThe power generation type->The calculation method of the quotation convolution weight of any day of each power generation enterprise comprises the following steps:
wherein ,indicate->The power generation type->Quotation reference coefficient of any day of each power generation enterprise, < + >>Indicate->Daily power generation influence factor of each power generation type on the day, < > for each power generation type>Represents the sum of daily power generation amounts of the day, +.>Indicate->The power generation type->Daily power generation amount of each power generation enterprise in the day, +.>Indicate->The number of power generation enterprises in the power generation types, +.>Indicate->The power generation type->Price quotation data of each power generation enterprise on the same day, +.>Indicate->Average value of quotation data of all power generation enterprises in each power generation type in the day, < >>Price quotation data mean value of day of all power generation enterprises, < ->Represents the number of types of power generation,indicate->The power generation type->Price quotation data of each power generation enterprise on the same day, +.>To avoid hyper-parameters with denominator 0, < ->Representing absolute value;
acquiring quotation reference coefficients of each power generation enterprise in the day, normalizing all the quotation reference coefficients, and marking the obtained result as the quotation convolution weight of each power generation enterprise in the day;
and acquiring the quotation convolution weight of each day of each power generation enterprise, wherein each period corresponds to one day, and acquiring the quotation convolution weight of each period of each power generation enterprise.
Further, the method for obtaining the daily power generation influence factor of each power generation type in each period according to the historical power generation data and the power generation type related data comprises the following specific steps:
acquiring daily power generation amount of each day of all power generation enterprises of each power generation type, and carrying out big powder on power generation type related data in a corresponding period of the power generation type according to the total daily power generation amount of the same day of the same power generation type to obtain the score of each power generation type related data;
according to the relevant data of the power generation types, the score is used as a label to train the classification network, the relevant data of the power generation types in each period of each power generation type is input into the classification network after training, and daily power generation influence factors in each period of each power generation type are output and obtained.
Further, the gradient promotion decision tree constructed and trained according to the ideal report value and the historical quotation data comprises the following specific methods:
the ideal report value of each day is obtained according to the quotation convolution weight and the historical quotation data, and the specific expression of the loss function loss for predicting and outputting residual errors is obtained by combining the historical quotation data of the power generation enterprises:
wherein ,representing days in historical quotation data, +.>Indicate->The power generation enterprises are->The quotation data of the day,indicate->Ideal reporting value of the day;
will beThe set formed by the results corresponding to different days is recorded as residual error, and for the decision tree of the regression problem, the residual error of the current model is fitted by the model in the next iteration to obtain a regression tree;
adding a new regression tree into the current model, and recalculating a predicted value, wherein the predicted value of each sample point is equal to the sum of the predicted results of all the previous weak classifiers, and adding the product obtained by multiplying the predicted result of the current decision tree by a learning rate; and repeatedly updating the weak classifier according to the residual error of the previous iteration until the objective function is converged to the minimum, and completing the training process of the predictor to obtain a trained gradient lifting decision tree.
Further, the method for obtaining the ideal report value of each day according to the quotation convolution weight and the historical quotation data comprises the following specific steps:
acquiring quotation data of all power generation enterprises on any day in the historical quotation data, weighting and summing the quotation data of each power generation enterprise on the day according to the quotation convolution weight of each power generation enterprise on the day, and marking the obtained result as an ideal report value of the day; and obtaining ideal report value of each day.
Further, the method for obtaining the quotation evaluation value of each power generation enterprise comprises the following specific steps:
acquiring quotation deviation values of each day of each power generation enterprise according to the ideal report value and the historical quotation data; first, thePrice assessment value of individual power generation enterprises>The calculation method of (1) is as follows:
wherein ,minimum value representing predicted quotation data of all power generation enterprises,/->Indicate->Predictive quotation data for individual power generation enterprises, +.>Indicate->Variance of all quotation deviation values of individual power generation enterprises, < >>An exponential function that is based on a natural constant; and acquiring a quotation evaluation value of each power generation enterprise.
Further, the method for obtaining the price bias value of each day of each power generation enterprise according to the ideal report value and the historical price bias data comprises the following specific steps:
acquisition of the firstThe absolute value of the difference between the quotation data of each day and the corresponding ideal report value in the historical quotation data of each power generation enterprise is recorded as +.>Price bias value of each day of each power generation enterprise; and acquiring the price quotation deviation value of each power generation enterprise on each day.
The beneficial effects of the invention are as follows: the method improves the problems of low algorithm efficiency and inaccurate prediction results by optimizing a gradient lifting decision tree algorithm, wherein the problems of larger regression tree depth and slow algorithm convergence are caused by the rough initial weak classifier of the algorithm, daily power generation influence factors of power generation types are obtained through daily power generation capacity and related data of the power generation types according to the power generation types of different power generation enterprises, the power generation contribution rate of the power generation types and the variance and difference of historical quotation data are combined to obtain quotation convolution weights of each power generation enterprise, and the historical quotation data are convolved to obtain an ideal quotation value function which accords with the daily power generation conditions and quotation value emphasis of actual histories and is an ideal reporting value under actual conditions; the method is used as an initial weak classifier, so that residual quantity increased by partial enterprise false reports, monopoly and abnormal quotations disturbing markets can be effectively reduced, the depth of a regression tree is shortened, the algorithm converges faster, the prediction is more accurate, and then the power generation enterprises which cooperate in the next period are preferentially selected according to the prediction result; the method greatly improves the algorithm efficiency and the prediction accuracy of the gradient lifting decision tree, and simultaneously maximally saves the electricity cost of a user enterprise.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the invention, and that other drawings can be obtained according to these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a flow chart of a method for predicting current price difference trend of electric power transaction based on a gradient-lifting decision tree according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, a flow chart of a method for predicting current price difference trend of electric power transaction based on a gradient-lifting decision tree according to an embodiment of the invention is shown, the method comprises the following steps:
and S001, collecting historical power generation data, power generation type related data and historical quotation data of a plurality of power generation enterprises in the power market, and obtaining daily quotation data.
The purpose of the embodiment is to predict the current price difference trend of the electric power transaction, and obtain the current day predicted quotation through a gradient lifting decision tree (GBDT algorithm) so as to avoid the problems of monopoly and information asymmetry, thereby providing a proper power generation enterprise for the user enterprise to cooperate and saving the electricity cost of the user enterprise; therefore, firstly, related data of the power market of the region where the user enterprise is located needs to be collected, in this embodiment, quotation data of a plurality of power generation enterprises in the power market is collected, including historical quotation data and daily quotation data, and the historical quotation data is formed by quotation data of each day in the last five years; collecting the generated energy of each day in the last five years of each power generation enterprise to form historical power generation data; meanwhile, the power generation type of each power generation enterprise, namely the power generation mode, is different, so that the related data of the power generation type corresponding to each enterprise needs to be obtained.
Thus, historical power generation data, power generation type related data and historical quotation data of a plurality of power generation enterprises in the power market are obtained, and meanwhile, the daily quotation data of each power generation enterprise is obtained.
Step S002, according to the historical power generation data and the related data of the power generation types, acquiring daily power generation influence factors of each power generation type in each period, and acquiring the quotation convolution weight of each period of each power generation enterprise by combining the historical quotation data.
It should be noted that, in the same region, the power spot price quotations of different power generation enterprises in different periods are different, so that the power spot transaction price difference trend prediction is not linear prediction according to the average value of the market price quotations, but the price difference change rule of different power generation enterprises needs to be analyzed, the power generation modes of each power generation enterprise are different, such as fire power, wind power, water power, photovoltaic power generation and the like, and the influences on each period are also different, so that the prediction mode is too coarse only according to the conventional average value fitting, and obvious over-fitting and under-fitting problems exist.
It should be further noted that the GBDT algorithm of the existing gradient boosting decision tree can be applied to predictive regression, so that the above problem can be effectively avoided, and a forward step algorithm is adopted, and each step trains a new weak classifier to fit the residual errors of all the previous classifiers; wherein each weak classifier is decision tree based, each split minimizing the delta on the loss function, thereby generating an optimal partitioning structure; finally, carrying out weighted summation on the results of all the trees to obtain a final prediction result; in the regression problem, the decision tree becomes a regression tree, and the final prediction result is obtained by the sum of regression tree layers classified by the weak classifier each time, so that the shorter the ideal regression tree layer depth is, the faster the prediction speed is, and the smaller the error is.
It should be further noted that, the price of the power spot transaction is continuously changed in time sequence, so that the historical quotation data of each power generation enterprise can be fitted into a continuous curve in time sequence; in the gradient lifting decision tree algorithm, fitting residual errors need to be updated, and because of different power generation modes and influence factors, the initial weak classifier is inaccurate in acquiring the residual errors, so that the GBDT iteration time is longer, the regression tree depth is longer, and the accuracy of a prediction result is influenced; the initial baseline, i.e. the initial weak classifier process, should be convolution calculation to improve the performance of the initial weak classifier and the accuracy of the residual error output by the initial weak classifier, so that the subsequent reclassification decision of the residual error is shorter and more accurate.
Specifically, firstly, daily power generation influence factors of each power generation enterprise in each period are acquired by analyzing the daily power generation amount and corresponding power generation type related data of each enterprise; taking thermal power generation as an example, acquiring historical power generation data of all thermal power generation enterprises, and acquiring daily power generation amount in the historical power generation data of each thermal power generation enterprise, namely power generation amount of each day, acquiring power generation type related data corresponding to each thermal power generation enterprise, namely fuel price data of each day, wherein the daily power generation amounts of all thermal power generation enterprises form a statistical set, and meanwhile, all the daily power generation amounts in each day correspond to one fuel price data, and evaluating the influence of the power generation type related data on the power generation amount in a manner of manually scoring all the fuel price data according to the corresponding daily power generation total amount, wherein the score is between 0 and 1, the larger the influence on the power generation efficiency is, the higher the score is, and conversely, the lower the daily power generation total amount is the sum of the daily power generation amounts of the thermal power generation enterprises in the same day; scoring all the power generation type related data according to the corresponding daily power generation amount to obtain a score corresponding to each power generation type related data, forming all the power generation type related data into an external influence characteristic data set, constructing a classification network, setting a loss function as a cross entropy loss function by adopting an Encoder-FC (fiber channel) in a network structure, and setting the external influence characteristic data set as 7:3, dividing the proportion into a training set and a verification set, inputting each element in the external influence characteristic data set into a classification network in the training process, taking the score corresponding to each element as a label, training by a gradient descent method until the loss function converges, and finishing training of the classification network; for any power generation type, inputting power generation type related data of any day into a classification network after training is completed, and recording an output result as a daily power generation influence factor of the power generation type in the day, wherein the nature of an output value is probability, and the value range is 0-1; according to the method, the daily power generation influence factors of each power generation type are obtained, and the power generation type related data of the same day of the power generation enterprises of the same type are the same, wherein each day is each day in the last five years.
Further, according to the daily power generation influence factor obtained through the power generation type related data, the daily power generation amount and the historical quotation data are combined, and the power generation contribution ratio and the variance contribution ratio of the power generation enterprises of the same type in each period are calculated, so that each period of each power generation enterprise is obtainedQuotation convolution weights to the firstThe power generation type->For example, the calculation method of the quotation convolution weight of any period of the power generation enterprises comprises the following steps:
wherein ,indicate->The power generation type->Quotation reference coefficient of any period (any day) of each power generation enterprise, +.>Indicate->Daily power generation influence factor of each power generation type on the day, < > for each power generation type>Represents the sum of the daily power generation amounts of the day,indicate->The power generation type->Daily power generation amount of each power generation enterprise in the day, +.>Indicate->The number of power generation enterprises in the power generation types, +.>Indicate->The power generation type->Quotation data of each power generation enterprise on the day, namely quotation data of each power generation enterprise on the day in historical quotation data, < >>Indicate->Average value of quotation data of all power generation enterprises in each power generation type in the day, < >>Price quotation data mean value of day of all power generation enterprises, < ->Representing the number of power generation types, in this embodiment +.>Calculation is performed (i.e. a->Indicate->The power generation type->Price quotation data of each power generation enterprise on the same day, +.>To avoid hyper-parameters with denominator 0, this embodiment uses +.>Calculation is performed (i.e. a->Representing absolute value; acquiring quotation reference coefficients of each power generation enterprise in the day according to the method, and carrying out softmax normalization on all the quotation reference coefficients, wherein the acquired result is recorded as the quotation convolution weight of each power generation enterprise in the day; according to the method, the quotation convolution weight of each day, that is, the quotation convolution weight of each period, of each power generation enterprise is obtained, and it should be noted that, in this embodiment, the convolution weight analysis is performed according to the daily power generation amount and the quotation data of each day, and then each period corresponds to each day.
At the moment, comprehensively quantifying to obtain a quotation reference coefficient through a daily power generation influence factor, a power generation contribution rate, a quotation variance contribution rate and an enterprise quotation difference ratio, wherein the larger the daily power generation influence factor is, the power generation contribution rate is obtained from the sum of the same type of daily power generation amount and the sum of the daily power generation amount, and the larger the power generation contribution rate is, the larger the reference coefficient for the power generation type is; the quotation variance contribution rate is to quantize the quotation data variances of different power generation types, and the smaller the quotation data variance is, the more stable the quotation of the power generation type is, the smaller the corresponding quotation variance contribution rate is, and the larger the quotation reference coefficient of the power generation type is; the enterprise quotation difference ratio is obtained by comparing the difference between the quotation data of the power generation enterprise and the average value of the daily quotation data with the average value, and meanwhile, the smaller the difference is, the larger the quotation reference coefficient is, so that the reciprocal of the ratio is used for participating in calculation.
Thus, the quotation convolution weight of each period of each power generation enterprise is obtained.
And step S003, optimizing an initial weak classifier of the GBDT algorithm according to the quotation convolution weight and the historical quotation data to obtain ideal reporting value of each period and predicted quotation data of each power generation enterprise.
It should be noted that, generally, the GBDT regression algorithm adopts a mean square error function, but because the initial weak classifier obtained according to the median and the mean is rough, the iterative process is longer, and the algorithm efficiency is lower, so the initial weak classifier is optimized according to the convolution weight and is brought into the loss function to obtain the residual error.
Specifically, firstly, for the quotation data of all power generation enterprises in any day in the historical quotation data, weighting and summing the quotation data of each power generation enterprise in the day according to the quotation convolution weight of each power generation enterprise, and marking the obtained result as an ideal reporting value of the day, namely, convolving the quotation data through the quotation convolution weight to obtain the ideal reporting value; according to the method, the ideal report value of each day is obtained, an ideal report value function is used as an initial weak classifier in a gradient lifting decision tree, the future ideal report value can be coarsely predicted according to the weak classifier, then a residual error is output, the residual error is fitted continuously, the classifier is updated after each fitting until the finally output residual error is minimum, and all regression tree layers are obtained.
Further, according to the quotation data of each day and the corresponding ideal quotation value function of each power generation enterprise, a loss function loss for predicting and outputting residual errors can be obtained, and the loss has the following specific expression:
wherein ,representing days in historical quotation data, +.>Indicate->The power generation enterprises are->The quotation data of the day,indicate->Ideal reporting value of the day; the loss function is obtained by the mean square error of the actual quotation of the power generation enterprise on the same day and the report value of the ideal report value prediction on the same day; but->The set of results corresponding to different days is the residual error, and for the decision tree of the regression problem, the model only needs to fit the residual error of the current model in the next iteration to learn a regression tree.
Further, adding a new regression tree into the current model, and recalculating a predicted value, wherein the predicted value of each sample point is equal to the sum of the predicted results of all the previous weak classifiers, and adding a product obtained by multiplying the predicted result of the current decision tree by a learning rate, wherein the learning rate is described by 0.05 in the embodiment, and is used for controlling the weight of the new classifier in each iteration; repeatedly updating the weak classifier according to the residual error of the previous iteration until reaching the minimum convergence of the objective function, so as to finish the training process of the predictor and obtain a gradient lifting decision tree after the training is finished; for the current day quotation data of each enterprise, acquiring current day power generation influence factors of each power generation type according to a classification network, inputting the current day quotation data and the current day power generation influence factors of the power generation type corresponding to the power generation enterprises into a gradient lifting decision tree after training is completed, and predicting quotation of the next period of each power generation enterprise, so that quotation data of the next period of each enterprise is obtained and recorded as predicted quotation data; it should be noted that, the training process is iterated by taking daily quotations of each day as a unit, that is, each day is taken as a period, the period length is not limited, and the corresponding predicted quotation data can be obtained by adjusting the quotation statistics period, such as one week, half month, one quarter, and the like, and adjusting the statistics period for the corresponding historical power generation data and the related data of the power generation type.
Thus, the construction and training of the gradient lifting decision tree are completed, and the predicted quotation data of each power generation enterprise are obtained.
And step S004, according to the historical quotation data, the predicted quotation data and the ideal reporting value, obtaining quotation evaluation values of each power generation enterprise, and finishing the prediction of the current price difference trend of the power transaction and the selection of the power generation enterprises.
It should be noted that, after the predicted quotation data is obtained, the quotation of the power generation enterprise can be evaluated according to the historical quotation data, the predicted quotation data and the ideal reporting value, and the smaller the variance obtained by the difference between the historical quotation data and the corresponding ideal reporting value of each day is, the smaller the predicted quotation data is, which indicates that the quotation of the power generation enterprise is more stable and the price is lower, and the quotation evaluation value is higher.
Specifically, by the firstFor example, the power generation enterprises can obtain +.>The absolute value of the difference between the quotation data of each day and the corresponding ideal report value in the historical quotation data of each power generation enterprise is recorded as +.>Price bias value of each day of each power generation enterprise, then the firstPrice assessment value of individual power generation enterprises>The calculation method of (1) is as follows:
wherein ,minimum value representing predicted quotation data of all power generation enterprises,/->Indicate->Predictive quotation data for individual power generation enterprises, +.>Indicate->Variance of all quotation deviation values of individual power generation enterprises, < >>Representing an exponential function based on natural constants, this embodiment by +.>The inverse proportion relation and normalization processing are presented, and an implementer can set an inverse proportion function and a normalization function according to actual conditions; according to the method, the quotation evaluation value of each power generation enterprise is obtained, and the power generation enterprise with the largest quotation evaluation value is selected for cooperation for the user enterprise, so that the electricity cost of the user enterprise can be saved to the greatest extent.
The prediction of the current price difference trend of the power transaction is completed through the gradient lifting decision tree, predicted quotation data are obtained for each power generation enterprise, and quotation evaluation values of the power generation enterprises are obtained by combining the historical quotation data of the power generation enterprises, so that the electricity cost of the user enterprises is saved to the greatest extent.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (7)
1. The power transaction spot price difference trend prediction method based on the gradient lifting decision tree is characterized by comprising the following steps of:
collecting historical power generation data, historical quotation data and daily quotation data of a plurality of power generation enterprises in the power market, and recording power generation type related data of each period;
acquiring quotation convolution weight of each period of each power generation enterprise according to daily power generation influence factors and historical quotation data of each period of each power generation type;
according to ideal report value and historical quotation data, constructing and training a gradient lifting decision tree, inputting daily power generation influence factors and daily quotation data, and outputting predicted quotation data of each power generation enterprise;
and obtaining the quotation evaluation value of each power generation enterprise according to the historical quotation data, the predicted quotation data and the ideal reporting value, and finishing the prediction of the current price difference trend of the power transaction and the selection of the power generation enterprises.
2. The method for predicting the current price difference trend of electric power transaction based on the gradient-lifting decision tree according to claim 1, wherein the method for obtaining the quotation convolution weight of each period of each power generation enterprise is as follows:
acquiring daily power generation influence factors of each power generation type in each period according to historical power generation data and power generation type related data, and acquiring daily power generation amount of each power generation enterprise in each day according to the historical power generation data, namelyThe power generation type->The calculation method of the quotation convolution weight of any day of each power generation enterprise comprises the following steps:
wherein ,indicate->The power generation type->Quotation reference coefficient of any day of each power generation enterprise, < + >>Indicate->Daily power generation influence factor of each power generation type on the day, < > for each power generation type>Represents the sum of daily power generation amounts of the day, +.>Indicate->The power generation type->Daily power generation amount of each power generation enterprise in the day, +.>Indicate->The number of power generation enterprises in the power generation types, +.>Indicate->The power generation type->Price quotation data of each power generation enterprise on the same day, +.>Indicate->Average value of quotation data of all power generation enterprises in each power generation type in the day, < >>Price quotation data mean value of day of all power generation enterprises, < ->Indicating the number of power generation types>Indicate->The power generation type->Price quotation data of each power generation enterprise on the same day, +.>To avoid hyper-parameters with denominator 0, < ->Representing absolute value;
acquiring quotation reference coefficients of each power generation enterprise in the day, normalizing all the quotation reference coefficients, and marking the obtained result as the quotation convolution weight of each power generation enterprise in the day;
and acquiring the quotation convolution weight of each day of each power generation enterprise, wherein each period corresponds to one day, and acquiring the quotation convolution weight of each period of each power generation enterprise.
3. The method for predicting the current price difference trend of electric power transaction based on the gradient boost decision tree according to claim 2, wherein the method for obtaining the daily power generation influence factor of each power generation type in each period according to the historical power generation data and the power generation type related data comprises the following specific steps:
acquiring daily power generation amount of each day of all power generation enterprises of each power generation type, and carrying out big powder on power generation type related data in a corresponding period of the power generation type according to the total daily power generation amount of the same day of the same power generation type to obtain the score of each power generation type related data;
according to the relevant data of the power generation types, the score is used as a label to train the classification network, the relevant data of the power generation types in each period of each power generation type is input into the classification network after training, and daily power generation influence factors in each period of each power generation type are output and obtained.
4. The method for predicting the current price difference trend of electric power transaction based on the gradient-boost decision tree according to claim 1, wherein the gradient-boost decision tree constructed and trained according to ideal report values and historical quotation data comprises the following specific steps:
the ideal report value of each day is obtained according to the quotation convolution weight and the historical quotation data, and the specific expression of the loss function loss for predicting and outputting residual errors is obtained by combining the historical quotation data of the power generation enterprises:
wherein ,representing days in historical quotation data, +.>Indicate->The power generation enterprises are->Quotation of heavenThe data set is used to determine, based on the data,indicate->Ideal reporting value of the day;
will beThe set formed by the results corresponding to different days is recorded as residual error, and for the decision tree of the regression problem, the residual error of the current model is fitted by the model in the next iteration to obtain a regression tree;
adding a new regression tree into the current model, and recalculating a predicted value, wherein the predicted value of each sample point is equal to the sum of the predicted results of all the previous weak classifiers, and adding the product obtained by multiplying the predicted result of the current decision tree by a learning rate; and repeatedly updating the weak classifier according to the residual error of the previous iteration until the objective function is converged to the minimum, and completing the training process of the predictor to obtain a trained gradient lifting decision tree.
5. The method for predicting the current price difference trend of the electric power transaction based on the gradient boost decision tree according to claim 4, wherein the obtaining the ideal report value of each day according to the quotation convolution weight and the historical quotation data comprises the following specific steps:
acquiring quotation data of all power generation enterprises on any day in the historical quotation data, weighting and summing the quotation data of each power generation enterprise on the day according to the quotation convolution weight of each power generation enterprise on the day, and marking the obtained result as an ideal report value of the day; and obtaining ideal report value of each day.
6. The method for predicting the current price difference trend of electric power transaction based on the gradient-lifting decision tree according to claim 1, wherein the method for obtaining the price evaluation value of each power generation enterprise comprises the following specific steps:
acquiring quotation deviation values of each day of each power generation enterprise according to the ideal report value and the historical quotation data; first, thePrice assessment value of individual power generation enterprises>The calculation method of (1) is as follows:
wherein ,minimum value representing predicted quotation data of all power generation enterprises,/->Indicate->Predictive quotation data for individual power generation enterprises, +.>Indicate->Variance of all quotation deviation values of individual power generation enterprises, < >>An exponential function that is based on a natural constant; and acquiring a quotation evaluation value of each power generation enterprise.
7. The method for predicting the current price difference trend of electric power transaction based on the gradient-lifting decision tree according to claim 6, wherein the acquiring the price deviation value of each day of each power generation enterprise according to the ideal report value and the historical price data comprises the following specific steps:
acquisition of the firstThe absolute value of the difference between the quotation data of each day and the corresponding ideal report value in the historical quotation data of each power generation enterprise is recorded as +.>Price bias value of each day of each power generation enterprise; and acquiring the price quotation deviation value of each power generation enterprise on each day.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310896464.4A CN116611859A (en) | 2023-07-21 | 2023-07-21 | Power transaction spot price difference trend prediction method based on gradient lifting decision tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310896464.4A CN116611859A (en) | 2023-07-21 | 2023-07-21 | Power transaction spot price difference trend prediction method based on gradient lifting decision tree |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116611859A true CN116611859A (en) | 2023-08-18 |
Family
ID=87684052
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310896464.4A Pending CN116611859A (en) | 2023-07-21 | 2023-07-21 | Power transaction spot price difference trend prediction method based on gradient lifting decision tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116611859A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117151766A (en) * | 2023-10-27 | 2023-12-01 | 国网山东综合能源服务有限公司 | Stock market clearing method based on supply and demand combined big data analysis |
| CN118297640A (en) * | 2024-06-06 | 2024-07-05 | 南京信息工程大学 | Product marketing management system and method based on big data |
-
2023
- 2023-07-21 CN CN202310896464.4A patent/CN116611859A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117151766A (en) * | 2023-10-27 | 2023-12-01 | 国网山东综合能源服务有限公司 | Stock market clearing method based on supply and demand combined big data analysis |
| CN118297640A (en) * | 2024-06-06 | 2024-07-05 | 南京信息工程大学 | Product marketing management system and method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116611859A (en) | Power transaction spot price difference trend prediction method based on gradient lifting decision tree | |
| CN105678404B (en) | Based on online shopping electricity and dynamically associate the micro-grid load forecasting system and method for the factor | |
| CN110705743A (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
| CN107730054B (en) | Gas load combined prediction method based on support vector regression | |
| CN112288164B (en) | Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast | |
| CN110443417A (en) | Multi-model integrated load prediction method based on wavelet transformation | |
| CN108805743A (en) | A kind of power grid enterprises' sale of electricity company operation Benefit Evaluation Method | |
| CN115099511A (en) | Photovoltaic power probability estimation method and system based on optimized copula | |
| CN111626473A (en) | Two-stage photovoltaic power prediction method considering error correction | |
| CN117575663A (en) | Fitment cost estimation method and system based on deep learning | |
| CN112418476A (en) | Ultra-short-term power load prediction method | |
| CN113762387B (en) | Multi-element load prediction method for data center station based on hybrid model prediction | |
| CN117277372A (en) | Multi-time-scale joint scheduling method and system for optical storage station and electronic equipment | |
| CN115809719A (en) | Short-term load prediction correction method based on morphological clustering | |
| CN115640874A (en) | Transformer state prediction method based on improved grey model theory | |
| CN116826737A (en) | Photovoltaic power prediction method, device, storage medium and equipment | |
| CN115238948A (en) | Method and device for predicting power generation capacity of small hydropower station | |
| CN110956304A (en) | Distributed photovoltaic power generation capacity short-term prediction method based on GA-RBM | |
| CN113222234A (en) | Gas demand prediction method and system based on integrated modal decomposition | |
| CN117290673A (en) | Ship energy consumption high-precision prediction system based on multi-model fusion | |
| CN116404637A (en) | Short-term load prediction method and device for electric power system | |
| CN115809927A (en) | Risk assessment method based on digital economy | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN113191069B (en) | Air conditioner load estimation method and system based on double-branch deep learning model | |
| CN112446550B (en) | Short-term building load probability density prediction method |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230818 |