A power price prediction method and system for a power transaction market

Abstract

The application discloses a power price prediction method and system for a power transaction market, collects a data set related to day-ahead clearing price prediction, adopts a mean filling method for granularity processing of day-ahead must-run and must-stop units, statutory holidays / weekends or not, and power coal prices, and adopts a downward filling method for granularity processing of weather type description and wind speed, so that the granularity is consistent with that of day-ahead clearing prices, a day-ahead clearing price prediction model is constructed, the day-ahead clearing price prediction model is trained, data required on a running day is collected, the data required on the running day is processed according to the method in step S2, the processed data is input into the trained model for day-ahead clearing price prediction, and the day-ahead clearing price prediction result is evaluated. Through the above data processing process and model structure design, the application can effectively improve the power price prediction accuracy, thereby providing an effective method and system for assisting power transaction decision-making.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power transaction, and particularly relates to a power price prediction method and system for a power transaction market. BACKGROUND

[0002] In the power transaction market, the power price fluctuates greatly due to the great volatility and uncertainty of the supply-side new energy power generation and the demand-side power consumption load. In the prior art, the schemes for predicting the power price include time series modeling, traditional similar day method and system simulation method. The time series method regards the fluctuation of the short-term power price as a random process, and determines a specific power price prediction formula through pattern recognition and parameter estimation. The method completely predicts the power price according to the historical data, and the fluctuation of the power price is affected by many uncertain factors, so that the prediction result of the power price is inaccurate and the accuracy is not high. The similar day method is to find historical days similar to the predicted daily load curve, and to obtain the predicted power price by weighting the power prices of the corresponding time periods of the historical days. The traditional similar day method in the prior art is simple in principle and easy to implement, but often only a single similarity algorithm is used to obtain similar days, and the power price prediction value is obtained by weighting the historical data, so the accuracy is still not high.

[0003] For a new energy power generation enterprise participating in the power market transaction, the high and low of the day-ahead clearing price directly affects its operating income. Therefore, how to accurately predict the day-ahead power price has great value and great significance for grasping the market law, formulating the power transaction strategy and improving the operating income of the new energy power plant. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the purpose of the present application is to provide a power price prediction method and system for a power transaction market, which solves the problem that in the prior art, the fluctuation of the power price is affected by many uncertain factors, so that the prediction result of the power price is inaccurate.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: A power price prediction method for a power transaction market, specifically comprising the following steps:

[0006] Step S1, collecting a data set related to the prediction of the day-ahead clearing price, the data set including the day-ahead clearing price, the new energy output prediction, the day-ahead interconnection line plan, the unified power consumption load, the day-ahead compulsory start and stop unit, the weather type description, the wind speed, whether it is a statutory holiday / weekend, the dynamic coal price;

[0007] In step S2, the granularity processing of the day-ahead must-on and must-off unit, the statutory holiday / weekend, and the granularity of the thermal coal price adopts the mean filling method and is consistent with the granularity of the day-ahead clearing price, the granularity processing of the weather type description and the wind speed adopts the downward filling method and is consistent with the granularity of the day-ahead clearing price, the granularity of the new energy output prediction, the day-ahead tie-line plan, and the unified power load is consistent with the granularity of the day-ahead clearing price;

[0008] In step S3, the day-ahead clearing price prediction model is constructed, and the day-ahead clearing price prediction model is trained.

[0009] In step S4, the data required on the operation day is collected, the data required on the operation day is processed according to the method in step S2, the processed data is input into the trained model for day-ahead clearing price prediction, and the day-ahead clearing price prediction result is evaluated.

[0010] Preferably, the step S2 specifically comprises the following steps:

[0011] In step S21, the capacities of the day-ahead must-on and must-off units are summed up to obtain the total capacity of the must-on units and the total capacity of the must-off units, respectively.

[0012] In step S22, the weather type description data of the text type is converted into the numerical type to obtain a weather index value W.

[0013] In step S23, the data of the statutory holiday / weekend is processed; for the statutory holiday / weekend, the value is 1, and for the non-statutory holiday / weekend, the value is 0, and the processed data of the statutory holiday / weekend is represented by H.

[0014] In step S24, the weather type description and the wind speed data are aggregated.

[0015] Preferably, the step S24 specifically comprises the following steps:

[0016] In step S241, the weather description data and the wind speed data of each county or district in the recent 30 days are obtained, the weather description data is processed according to the method in step S22 to obtain the corresponding weather index value W, the day-ahead clearing price P of the corresponding time node is obtained, thereby forming a data set X, the data set X includes the weather index W, the wind speed S, and the day-ahead clearing price P, and the specific expression is as follows:

[0017]

[0018] wherein a represents a certain county, a = 1,...,k, t represents a certain time, t = 1,...,n, w a,t represents the weather index value of the a county at the t time, s a,t represents the wind speed value of the a county at the t time; p tindicates the day-ahead clearing price at the t-th moment;

[0019] Step S242, respectively solving the Pearson correlation coefficient between the weather index value of each county and the day-ahead clearing price and the Pearson correlation coefficient between the wind speed value and the day-ahead clearing price The specific calculation formula is as follows:

[0020]

[0021] Among them, indicates the average value of the weather index value at each time point in a county; indicates the average value of the day-ahead clearing price at each time point; indicates the average value of the wind speed value at each time point in a county, p t indicates the day-ahead clearing price at the t-th moment;

[0022] Step S243, normalizing the Pearson correlation coefficient between the weather index value of each county and the day-ahead clearing price and the Pearson correlation coefficient between the wind speed value and the day-ahead clearing price to obtain the weather weight value of each county and the wind speed weight value The specific calculation formula is as follows:

[0023]

[0024] Step S244, adopting a weighted summation manner to combine the weather index value and the wind speed data of multiple counties at the t-th moment into a comprehensive weather index w t and a comprehensive wind speed value s t , so as to obtain a comprehensive weather index W' and a comprehensive wind speed value S'; the specific calculation manner is as follows:

[0025]

[0026] Step S245, data granularity unification processing; for the day-ahead must-on and must-off unit, the statutory holiday / weekend or not, and the processing of the steam coal price, the mean filling method is adopted, that is, the value at any moment within d days is the value collected in d days; for the granularity processing of the weather type description and the wind speed, the downward filling method is adopted, that is, within the range from t to t+1 moment, the value of all time points is the value corresponding to the t moment.

[0027] Preferably, the step S3 day-ahead clearing price prediction model comprises three units, which are respectively: an electricity price prediction algorithm unit based on a random forest algorithm, an electricity price prediction algorithm unit based on a polynomial regression, and a data fusion algorithm unit based on a linear regression.

[0028] Preferably, the input values of the electricity price prediction algorithm unit based on the random forest algorithm and the electricity price prediction algorithm unit based on the polynomial regression in the step S3 are the new energy output prediction E, the day-ahead tie-line plan T, the total capacity of the day-ahead must-start unit C, the total capacity of the day-ahead must-stop unit C, the comprehensive weather index W', the comprehensive wind speed S', whether it is a statutory holiday / weekend H, and the price of the power coal T, and the output value is the day-ahead clearing price prediction value LP. on , the total capacity of the day-ahead must-stop unit C off , the comprehensive weather index W', whether it is a statutory holiday / weekend H, and the price of the power coal T, and the output value is the day-ahead clearing price prediction value LP.

[0029] Preferably, the step S3 is specifically performed according to the following steps:

[0030] Step S31, divide the processed data into a training set and a validation set according to a ratio of 7:3; input the processed training set and validation set into the electricity price prediction algorithm unit based on the polynomial regression to determine the parameter order;

[0031] Step S32, train the processed training set in the electricity price prediction algorithm unit based on the random forest algorithm to obtain the prediction result of the training set.

[0032] Preferably, the parameter order is determined in a dynamic parameter selection manner, an order range k is set, wherein k∈{2,3,4}, and the value of the order k is sequentially traversed for model training, and the trained model is verified on the validation set, while the evaluation index MAE is used to evaluate the verification result, and then the order corresponding to the minimum MAE value is selected as the final order, wherein the MAE calculation formula is as follows:

[0033]

[0034] Wherein, p t represents the day-ahead clearing price at the t-th moment, and p t ' represents the day-ahead clearing price prediction value at the t-th moment.

[0035] Preferably, the step S3 inputs the prediction results of the electricity price prediction algorithm unit based on the random forest algorithm and the electricity price prediction algorithm unit based on the polynomial regression into the data fusion algorithm unit based on the linear regression for training; wherein the maximum number of iterations of the electricity price prediction algorithm unit based on the random forest algorithm is set to 100.

[0036] Preferably, the step S4 selects the mean absolute percentage error MAPE and the accuracy ACC as the day-ahead clearing price prediction result evaluation indexes, and the calculation formulas are as follows:

[0037]

[0038]

[0039] wherein, represents the average value of the day-ahead clearing price at n time points; p t represents the day-ahead clearing price prediction value at the t time point, and acc is a process variable.

[0040] A price prediction system for a power trading market comprises:

[0041] A data collection module is configured to collect a data set related to day-ahead clearing price prediction, the data set comprising day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, unified power consumption load, day-ahead must-run and must-stop unit, weather type description, wind speed, statutory holiday / weekend, and power coal price.

[0042] A data processing module is configured to process the data collected by the data collection module, and to process the granularity of the day-ahead must-run and must-stop unit, statutory holiday / weekend, and power coal price by using a mean filling method to make the granularity consistent with that of the day-ahead clearing price, to process the granularity of the weather type description and wind speed by using a downward filling method to make the granularity consistent with that of the day-ahead clearing price, and to make the granularity of the new energy output prediction, day-ahead tie-line plan, and unified power consumption load consistent with that of the day-ahead clearing price.

[0043] A model construction module is configured to construct a day-ahead clearing price prediction model and train the day-ahead clearing price prediction model.

[0044] A day-ahead clearing price prediction module is configured to collect data required on a running day, process the data required on the running day, and input the processed data into the trained model to predict the day-ahead clearing price.

[0045] Compared with the prior art, the price prediction method for the power trading market has the following beneficial effects: the price prediction method for the power trading market collects a data set related to day-ahead clearing price prediction, collects data related to price prediction, and makes the data more comprehensive; processes the data set, makes the granularity of the day-ahead must-run and must-stop unit, statutory holiday / weekend, power coal price, weather type description, and wind speed consistent with that of the day-ahead clearing price, fuses high-dimensional data into one comprehensive weather data, and thus solves the problem of high dimensionality of weather data; constructs a price prediction model, trains the model, and makes the prediction result more accurate; finally, visualizes the day-ahead clearing price prediction result and evaluates the prediction result, so that the data is easy to observe, and further provides data support for power traders of new energy power generation enterprises.

[0046] Further, the weather type description data of the text type is converted into the numerical type by using a corresponding relationship mapping method, so as to facilitate the processing in the subsequent steps.

[0047] Further, the aggregation of the weather type description and the wind speed data reduces the dimension of the data, reduces the feature dimension of the input model, and makes the training result of the model more accurate.

[0048] Further, the day-ahead clearing price prediction model combines the random forest algorithm-based price prediction algorithm unit, the polynomial regression-based price prediction algorithm unit, and the linear regression-based data fusion algorithm unit, so as to further utilize the advantages of the algorithms to achieve better prediction effect.

[0049] A power transaction market-oriented price prediction system performs prediction result evaluation on the day-ahead clearing price prediction result, so that the data is convenient for observation and auxiliary transaction decision. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flow chart of the power transaction market-oriented price prediction method of the present application;

[0051] Figure 2 A MAPE evaluation result graph of the three day-ahead clearing price prediction methods of the present application;

[0052] Figure 3 An ACC evaluation result graph of the three day-ahead clearing price prediction methods of the present application. DETAILED DESCRIPTION

[0053] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0054] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to the process, method, product, or device.

[0055] A power price prediction method for a power transaction market, as shown in Figure 1 specifically comprising the following steps:

[0056] Step S1, collecting a data set related to the day-ahead clearing price prediction;

[0057] The data set includes day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, unified calling load, day-ahead must-on and must-off unit, weather type description, wind speed, legal holiday / weekend or not, and power coal price.

[0058] The internal data includes day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, unified calling load, and day-ahead must-on and must-off unit.

[0059] The external data includes weather type description, wind speed, legal holiday / weekend or not, and power coal price.

[0060] The data type case: the day-ahead must-on and must-off unit, weather type description, and legal holiday / weekend or not are text data, and the rest of the data set is numerical type.

[0061] The data granularity case: the day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, and unified calling load are minute level, i.e., 1 piece / 15 minutes; the day-ahead must-on and must-off unit, legal holiday / weekend or not, and power coal price are day level, i.e., 1 piece / day; the weather type description and wind speed are hour level, i.e., 1 piece / 3 hours.

[0062] The data range case: the data in the last year is collected; in particular, the weather type description and wind speed data in the external data are collected from the meteorological data of all counties or districts in the region covered by the power transaction market.

[0063] The granularity processing of the day-ahead must-on and must-off unit, legal holiday / weekend or not, and power coal price adopts the mean filling method and is consistent with the granularity of the day-ahead clearing price; the granularity processing of the weather type description and wind speed adopts the downward filling method and is consistent with the granularity of the day-ahead clearing price; the granularity of the new energy output prediction, day-ahead tie-line plan, and unified calling load is consistent with the granularity of the day-ahead clearing price.

[0064] Step S21, summing the capacities of the day-ahead must-on and must-off units to obtain the total capacity of the must-on unit and the total capacity of the must-off unit, respectively.

[0065] The capacity value corresponding to the must-on and must-off unit is queried according to the online public information, and for the unit for which the corresponding capacity information cannot be found, the default value is set, and the default value is 200 MW.

[0066] Step S22, converting the weather type description data of the text type into the numerical type by using a self-defined conversion table, so as to obtain a weather index value W; wherein the self-defined conversion table adopts a corresponding relationship mapping method, and the self-defined conversion table is specifically shown in Table 1:

[0067] Table 1

[0068] Serial number Weather type description Conversion value 1 Sunny 1 2 Cloudy 2 3 Overcast 3 4 Shower / thunder shower 4 5 Light rain / snow 5 6 Moderate rain / snow 6 7 Heavy rain / snow 7 8 Heavy rain / very heavy rain / snowstorm 8

[0069] For the weather description condition similar to the compound type of sunny to cloudy and cloudy to overcast appearing in the weather forecast, the mean value method is adopted to process it. First, the weather types contained in the above table are extracted by using the text matching method. Second, the extracted weather types are converted into corresponding values by referring to the above table. Finally, the average value of the corresponding values of each type is taken as the final conversion value of the weather description, that is, the weather index value W.

[0070] Step S23, processing the legal holiday / weekend data; for the legal holiday / weekend, the value is 1, and for the non-legal holiday / weekend, the value is 0. The processed legal holiday / weekend data is represented by H.

[0071] Step S24, aggregating the weather type description and wind speed data; since the weather type description and wind speed data obtained are data of multiple places, if these meteorological data are all input into the model as features, it will cause the feature dimension to be too high, which will affect the model training result. Therefore, it is necessary to aggregate the data of multiple places into one feature data, so as to realize the purpose of dimension reduction. The aggregation process of the weather type description and wind speed data is as follows:

[0072] Step S241, obtaining the weather description data and wind speed data of each county or district in the recent 30 days, and processing the weather description data according to the method of step S22 to obtain the corresponding weather index value W; at the same time, obtaining the day-ahead clearing price P of the corresponding time node, so as to constitute a data set X, which includes the weather index value W, the wind speed S and the day-ahead clearing price P; the specific expression is as follows:

[0073]

[0074] wherein a represents a certain county, and a = 1,...,k, t represents a certain time, t = 1,...,n, w a,t represents the weather index value of the a county at the t time, s a,t represents the wind speed value of the a county at the t time; p t represents the day-ahead clearing price at the t time;

[0075] Step S242, respectively solving the Pearson correlation coefficients between the weather index values of each district and the day-ahead clearing prices and the Pearson correlation coefficient between the wind speed value and the day-ahead clearing price The specific calculation formula is as follows:

[0076]

[0077] wherein, represents the average value of the weather index value at each time point in a county; represents the average value of the day-ahead clearing price at each time point; represents the average value of the wind speed value at each time point in a county, p t represents the day-ahead clearing price at the tth moment;

[0078] Step S243, normalizing the weather weight value and the wind speed weight value in step S242 to obtain the weather weight value and the wind speed weight value The specific calculation formula is as follows:

[0079]

[0080] Step S244, merging the weather index value and the wind speed data of multiple counties at the tth moment into a comprehensive weather index w t and a comprehensive wind speed value s t at the tth moment by using the weighted summation method, so as to obtain the comprehensive weather index W' and the comprehensive wind speed value S'; the specific calculation method is as follows:

[0081]

[0082] Step S245, data granularity unification processing; since the granularity of the day-ahead clearing price data collection is 1 strip / 15 minutes, it is necessary to align the granularity of the day-ahead must-start and must-stop unit, whether it is a statutory holiday / weekend, the power coal price, the weather type description, the wind speed data with the granularity of the day-ahead clearing price, that is, to process them into 1 strip / 15 minutes, wherein the new energy output prediction, the day-ahead tie line plan, and the unified power load are all 1 strip / 15 minutes consistent with the day-ahead clearing price; among them, for the day-ahead must-start and must-stop unit, whether it is a statutory holiday / weekend, the power coal price, the mean filling method is adopted, that is, the value at any time within d days is the value collected in d days; for the granularity processing of the weather type description and the wind speed, the downward filling method is adopted, that is, from the tth moment to the t+1th moment, the value at all time points is the value corresponding to the tth moment;

[0083] Step S3, a day-ahead clearing price prediction model is constructed and the day-ahead clearing price prediction model is trained; the day-ahead clearing price prediction model based on the fusion of the random forest regression algorithm and the polynomial regression algorithm comprises three units, which are respectively: a power price prediction algorithm unit based on the random forest algorithm, a power price prediction algorithm unit based on the polynomial regression, and a data fusion algorithm unit based on the linear regression; wherein the input values of the power price prediction algorithm unit based on the random forest algorithm and the power price prediction algorithm unit based on the polynomial regression are new energy output prediction E, day-ahead tie-line plan T, total load L, total capacity of day-ahead compulsory units C on , total capacity of day-ahead compulsory units C off , comprehensive weather index W, comprehensive wind speed S, whether it is a statutory holiday / weekend H, power coal price T, and the output value is day-ahead clearing price prediction value LP; the input value of the data fusion algorithm unit based on the linear regression is the day-ahead clearing price prediction value LP, and the output value is the day-ahead clearing price prediction value P'; the prediction results of the power price prediction algorithm unit based on the random forest algorithm and the power price prediction algorithm unit based on the polynomial regression are input into the data fusion algorithm unit based on the linear regression for training; wherein the maximum number of iterations of the power price prediction algorithm based on the random forest algorithm is set to 100;

[0084] Step S31, the processed data is divided into a training set and a validation set in a ratio of 7:3; the processed training set and validation set are input into the power price prediction algorithm unit based on the polynomial regression to determine the parameter order; wherein, for the determination of the key parameter order in the polynomial regression algorithm unit, a dynamic parameter selection method is used to determine, specifically, the order range k is set, wherein k e {2, 3, 4}, and the value of the order k is traversed in turn to train the model, and the trained model is verified on the validation set, and at the same time, the evaluation index MAE is used to evaluate the verification result, and then the order corresponding to the minimum MAE value is selected as the final order, wherein the MAE calculation formula is as follows:

[0085]

[0086] Wherein, p t represents the day-ahead clearing price at the t-th moment, and p t ' represents the day-ahead clearing price prediction value at the t-th moment.

[0087] Step S32, the processed training set is trained in the power price prediction algorithm unit based on the random forest algorithm to obtain the prediction result of the training set;

[0088] Step S4, collect the data required on the operation day, and process the data required on the operation day according to the method in step S2, so as to input the processed data into the trained model for day-ahead clearing price prediction, and then obtain the day-ahead clearing price prediction result, and evaluate the day-ahead clearing price prediction result;

[0089] The broken line chart is used to visualize the time sequence, and in order to quantitatively describe the advantages and disadvantages of the prediction result, two evaluation indexes are proposed, the result evaluation indexes are the mean absolute percentage error MAPE and the accuracy ACC, and the calculation formulas are as follows:

[0090]

[0091]

[0092] Among them, The average value of the day-ahead clearing price at n time points is represented by p t The day-ahead clearing price prediction value at the t time point is represented by acc, and the process variable is represented by acc.

[0093] A power transaction market-oriented price prediction system, comprising:

[0094] A data collection module for collecting a data set related to day-ahead clearing price prediction, the data set including day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, unified power load, day-ahead must-start and must-stop unit, weather type description, wind speed, legal holiday / weekend, and power coal price;

[0095] A data processing module for processing the data collected by the data collection module, the granularity of the day-ahead must-start and must-stop unit, legal holiday / weekend, and power coal price is processed by the mean filling method to be consistent with the granularity of the day-ahead clearing price, the granularity of the weather type description and wind speed is processed by the downward filling method to be consistent with the granularity of the day-ahead clearing price, and the granularity of the new energy output prediction, day-ahead tie-line plan, and unified power load is consistent with the granularity of the day-ahead clearing price;

[0096] A model construction module for constructing a day-ahead clearing price prediction model and training the day-ahead clearing price prediction model;

[0097] A day-ahead clearing price prediction module for collecting data required on the operation day, processing the data required on the operation day, and inputting the processed data into the trained model for day-ahead clearing price prediction.

[0098] The power price prediction method for the electricity trading market has been applied to the day-ahead clearing price prediction of a certain electricity trading market. Meanwhile, in order to verify the advantages and disadvantages of the power price prediction method for the electricity trading market proposed by the present application, the present application selects two kinds of day-ahead clearing price prediction models currently widely used for comparison. Specifically, method 1 uses a random forest algorithm to realize day-ahead clearing price prediction, and method 2 uses a neural network algorithm to realize day-ahead clearing price prediction. Among them, the data set is selected from the data published by a certain electricity trading market, the weather data is from the public data of the meteorological bureau, the holiday data is from the data published by the state, and the power coal price is from the network; the comparison process adopts a sliding window mode, that is, the algorithm is executed at 21 o'clock every night to predict the day-ahead price at 96 time points (every 15 minutes a point) in the next day; the cycle length is 30 consecutive days; the evaluation index selects the mean absolute percentage error MAPE and the accuracy ACC proposed by the present application, and the final result is as shown in Table 2. Figure 2 、 Figure 3 as shown in Table 2.

[0099] Table 2

[0100]

[0101]

[0102] From the above verification and evaluation results, it can be seen that the method proposed by the present application is superior to method 1 and method 2 in terms of monthly average ACC and monthly average MAPE. At the same time, during the verification period, the minimum ACC of the method of the present application is greater than that of method 1 and method 2, and the maximum MAEP is less than that of method 1 and method 2. Based on the above data, it can be seen that the method proposed by the present application effectively improves the accuracy of day-ahead clearing price prediction.

[0103] In order to improve the accuracy of day-ahead clearing price prediction in the electricity trading market, the present application provides a day-ahead clearing price prediction method for the electricity trading market. The method first collects internal and external data sets related to day-ahead clearing price prediction. On this basis, the matching method, the relationship mapping method and the weighted average method are used to process the collected day-ahead must start and must stop units, legal holidays / weekends and weather type description data sets. Secondly, a day-ahead clearing price prediction model is constructed, which combines two different algorithms to further utilize the advantages of each algorithm to achieve better prediction results. At the same time, the processed data set is used for model training and day-ahead clearing price prediction. Finally, the day-ahead clearing price prediction result is visualized and the historical prediction result is evaluated, thereby providing data support for power traders of new energy power generation enterprises.

[0104] The above merely illustrates the technical idea of the present application, and cannot limit the protection scope of the present application. Any modification made according to the technical idea of the present application on the basis of the technical scheme falls within the protection scope of the present application.

Claims

1. A method for electricity price forecasting for electricity trading market, characterized by, Specifically comprising the following steps: Step S1, collect the data set related to the day-ahead clearing price prediction, the data set including day-ahead clearing price, new energy output prediction, day-ahead tie-line plan, unified calling load, day-ahead must-run and must-stop unit, weather type description, wind speed, legal holiday / weekend or not, and dynamic coal price; Step S2, the granularity processing of the day-ahead must-run and must-stop unit, legal holiday / weekend or not, and dynamic coal price adopts the mean filling method and is consistent with the granularity of the day-ahead clearing price, the granularity processing of the weather type description and wind speed adopts the downward filling method and is consistent with the granularity of the day-ahead clearing price, and the granularity of the new energy output prediction, day-ahead tie-line plan, and unified calling load is consistent with the granularity of the day-ahead clearing price; Step S3, constructing a day-ahead clearing price prediction model and training the day-ahead clearing price prediction model; The day-ahead clearing price prediction model comprises three units, namely, a price prediction algorithm unit based on a random forest algorithm, a price prediction algorithm unit based on a polynomial regression, and a data fusion algorithm unit based on a linear regression; The input values of the electricity price prediction algorithm unit based on the random forest algorithm and the electricity price prediction algorithm unit based on the polynomial regression are new energy output prediction , day-ahead tie-line plan , total load of general power consumption , total capacity of day-ahead compulsory on-line units , total capacity of day-ahead compulsory off-line units , comprehensive weather index , comprehensive wind speed , whether it is a statutory holiday / weekend , power coal price , and the output value is a day-ahead clearing price prediction value ; The data fusion algorithm unit based on linear regression takes the day-ahead clearing price prediction value as the input value and outputs the day-ahead clearing price prediction value . ​ Step S4, collecting the data required on the operation day, processing the data required on the operation day according to the method in step S2, inputting the processed data into the trained model to predict the day-ahead clearing price, and evaluating the day-ahead clearing price prediction result.

2. The method of claim 1, wherein the power transaction market-oriented price forecasting method is characterized by, The step S2 specifically comprises the following steps: Step S21, summing the capacities of the day-ahead must-run and must-stop units to obtain the total capacity of the must-run unit and the total capacity of the must-stop unit respectively; Step S22, converting the weather type description data of the text type into the numerical type, thereby obtaining the weather index value ; Step S23, processing the statutory holiday / weekend data; for the statutory holiday / weekend, assign the value 1, otherwise assign the value 0, the processed statutory holiday / weekend data is represented as Holiday Weekend Data Step S24, performing aggregation processing on the weather type description and wind speed data.

3. The method of claim 2, wherein the power transaction market-oriented price forecasting method is characterized by, The step S24 specifically comprises the following steps: Step S241, acquire weather description data and wind speed data of each county or district in the recent 30 days, and process the weather description data according to the method of step S22 to obtain the corresponding weather index value W ; at the same time, acquire the day-ahead clearing price of the corresponding time node P , thereby forming a data set X , the data set X including weather index W , wind speed S and day-ahead clearing price P ; the specific expression is as follows: in, a Indicates a certain county or district, and , t Indicates a certain moment, , express a County / District t Weather index value at any given time. express a County / District t Wind speed at any given time; Indicates the first t The day-end clearing price at the specified time; Step S242, respectively solve the Pearson correlation coefficient between each county weather index value and the day-ahead clearing price and the Pearson correlation coefficient between the wind speed value and the day-ahead clearing price ; the specific calculation formula is as follows: wherein, represents a the average of the weather index values at each time point in the county area; represents the average of the day-ahead clearing prices at each time point; represents a the average of the wind speed values at each time point in the county area, represents the day-ahead clearing price at the t moment. Step S243, normalizing the weather weight value in step S242 and to obtain the weather weight value of each county and the wind speed weight value of each county; the specific calculation formula is as follows: Step S244, the weather index value and the wind speed data of the time are combined into one t comprehensive weather index value of the time t and the comprehensive wind speed value , so as to obtain the comprehensive weather index value and the comprehensive wind speed value ; the specific calculation manner is as follows:​ Step S245, data granularity unification processing; for the processing of the day-ahead must-run and must-stop unit, legal holiday / weekend or not, and dynamic coal price, the mean filling method is adopted, that is, the value at any time within d days is the value collected on d days; for the granularity processing of the weather type description and wind speed, the downward filling method is adopted, that is, within the range from t to t+1, the value at all time points is the value corresponding to t.

4. The method of claim 1, wherein the power transaction market-oriented price forecasting method is characterized by, The step S3 is specifically performed according to the following steps: Step S31, dividing the processed data into a training set and a validation set according to a ratio of 7:3; inputting the processed training set and validation set into the price prediction algorithm unit based on the polynomial regression to determine the parameter order; Step S32, training the processed training set in the price prediction algorithm unit based on the random forest algorithm to obtain the prediction result of the training set.

5. The method of claim 4, wherein the power transaction market-oriented price forecasting method is characterized by, The parameter order is determined in a dynamic parameter selection manner, and an order range is set k , wherein , and the order values are sequentially traversed k The trained model is verified on the verification set, and the evaluation index MAE is used to evaluate the verification result, and then the order corresponding to the minimum MAE value is selected as the final order, wherein the MAE calculation formula is as follows: wherein, represents the day-ahead clearing price at the t time point, represents the day-ahead clearing price at the t time point.

6. The method of claim 4, wherein the power transaction market-oriented price forecasting method is characterized by, The step S3 inputs the prediction results of the price prediction algorithm unit based on the random forest algorithm and the price prediction algorithm unit based on the polynomial regression into the data fusion algorithm unit based on the linear regression for training; wherein the maximum iteration number of the price prediction algorithm unit based on the random forest algorithm is set to 100.

7. The method of claim 1, wherein the method is a power transaction market-oriented electricity price forecasting method. The step S4 selects the mean absolute percentage error MAPE and the accuracy ACC as the evaluation indexes of the day-ahead clearing price prediction result, and the calculation formulas are as follows: in, express n The average clearing electricity price of the previous time period; Indicates the first t The day-ahead clearing price forecast at time acc is a process variable.

8. A power price forecasting system for a power trading market, characterized by, It comprises: The data collection module is configured to collect a data set related to the day-ahead clearing price prediction, the data set including the day-ahead clearing price, the new energy output prediction, the day-ahead tie-line plan, the utility power load, the day-ahead compulsory on-off unit, the weather type description, the wind speed, the statutory holiday / weekend, and the dynamic coal price. The data processing module is configured to process the data collected by the data collection module, and to process the granularity of the day-ahead compulsory on-off unit, the statutory holiday / weekend, and the dynamic coal price by using the mean filling method to be consistent with the granularity of the day-ahead clearing price, to process the granularity of the weather type description and the wind speed by using the downward filling method to be consistent with the granularity of the day-ahead clearing price, and to process the granularity of the new energy output prediction, the day-ahead tie-line plan, and the utility power load to be consistent with the granularity of the day-ahead clearing price. The model construction module is configured to construct the day-ahead clearing price prediction model and to train the day-ahead clearing price prediction model. The day-ahead clearing price prediction model includes three units, namely, a price prediction algorithm unit based on a random forest algorithm, a price prediction algorithm unit based on a polynomial regression, and a data fusion algorithm unit based on a linear regression. The input values ​​for the electricity price prediction algorithm units based on random forest and multinomial regression are the power output prediction values ​​of new energy sources. The planned connection line , unified dispatch of electrical load Total capacity of generating units that must be started up by the next few days Total capacity of units that must be shut down in the coming days Comprehensive Weather Index Overall wind speed Statutory holidays / weekends or not thermal coal prices The output value is the day-ahead clearing price forecast. ; The data fusion algorithm unit based on linear regression takes the day-ahead clearing price prediction value as the input value and outputs the day-ahead clearing price prediction value . ​ The day-ahead clearing price prediction module is configured to collect data required on a running day and to process the data required on the running day, and to input the processed data into the trained model to perform the day-ahead clearing price prediction.

Images