CatBoost-based Prediction Method, Device and Storage Medium for Payload Spare Capacity Requirement

Abstract

The present invention discloses a method, device and storage medium for predicting the net load reserve capacity demand based on CatBoost. The method includes: obtaining the prediction feature information of each prediction point within the period to be predicted; inputting the prediction feature information into a load / new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error and new energy prediction error corresponding to each prediction point; subtracting the load prediction error and new energy prediction error of each prediction point to obtain the prediction result of the net load reserve capacity demand of each prediction point; and obtaining the upper and lower limits of the prediction interval of the net load reserve capacity demand of each prediction point according to the cumulative distribution function of the net load reserve capacity demand prediction error at a certain confidence level. The present invention can be used for evaluating the net load reserve capacity demand for the day-ahead and intra-day, with high prediction stability, accuracy, operation speed and operation efficiency, and strong universality.

Description

Technical Field

[0001] The present invention relates to the technical field of net load reserve capacity demand assessment, and particularly to a method, device and storage medium for predicting net load reserve capacity demand based on CatBoost. Background Art

[0002] Reserve demand stems from the uncertainties in power grid operation. On the one hand, it is unpredictable power grid accidents (contingency reserve). On the other hand, it is the inaccuracy of supply and demand forecasting (load reserve), which mainly comes from load forecasting errors in traditional power grids. In China, a scientific reserve demand assessment method has not been formed in the power grid, and the demand for grid upward and downward regulation resources brought by the uncertainty of new energy power generation cannot be comprehensively considered. The power grid reserve problem involves operation problems at different levels such as frequency control, security control and economic dispatching. However, the current reserve management mode determined according to a certain proportion of the load and the maximum unit capacity is difficult to meet the power grid security and economic operation requirements under the new situation. Therefore, it is necessary to carry out systematic and market-oriented research and exploration on the reserve management of the power grid.

[0003] With the increase in the penetration rate of new energy, the power grid needs to more accurately grasp the prediction error characteristics of the net load curve of the power grid and establish a net load reserve demand assessment technology adapted to the development of new energy.

[0004] In recent years, many scholars have studied the optimal allocation of reserve capacity under the integration of new energy to obtain an effective reserve plan that is both scientific and economical. Currently, the existing research is mainly divided into two types:

[0005] One is the stochastic programming method. Monte Carlo simulation is used to generate operation scenarios composed of various uncertain factors such as wind power and load, and intelligent algorithms (such as particle swarm optimization algorithm) are used to solve the reserve capacity configuration.

[0006] The other is the convolution method. Various uncertain factors such as wind power / solar power, load, and unit outage are modeled using probability distribution functions. Through the convolution principle, the ideal supply margin function of the system is obtained, and the reserve requirement is obtained by combining the probability of loss of load. The deterministic unit commitment model is solved to obtain the reserve capacity configuration.

[0007] In the stochastic programming method, the calculation of Monte Carlo simulation scenario generation is large, which is not suitable for problems with more types of uncertain factors. The convolution method has a small calculation amount, but the uncertain factors that can be considered are limited.

[0008] The net load reserve capacity is divided into upward reserve capacity and downward reserve capacity. In a traditional (non-new energy) power system, the load forecasting error is relatively small, and the power sources are basically adjustable conventional units such as thermal power and hydropower units, and the downward regulation capacity of the units is relatively abundant. Therefore, only the basic upward reserve capacity required due to equipment failures and load forecasting errors needs to be set. In view of the uncertainty of large-scale new energy, this method adds additional upward and downward reserve requirements due to the grid connection of wind power / solar power on the original reserve basis.

[0009] In summary, there are currently many methods for predicting the output of wind power / solar power. Limited by multiple uncertain factors such as wind / solar forecasting and load forecasting, there is currently no general method to accurately evaluate the net load reserve capacity demand in a certain area. Summary of the Invention

[0010] The present invention provides a method, device and storage medium for predicting the net load reserve capacity demand based on CatBoost to solve the problem that the existing methods for predicting the load reserve capacity demand are difficult to make accurate predictions under the action of multiple uncertain factors such as wind / solar forecasting and load forecasting.

[0011] A method for predicting the net load reserve capacity demand based on CatBoost includes:

[0012] S1: Obtain the prediction feature information of each prediction point within the time period to be predicted;

[0013] S2: Input the prediction feature information of each prediction point into the load / new energy prediction error prediction model constructed based on the CatBoost integrated learning method to obtain the load prediction error and new energy prediction error corresponding to each prediction point;

[0014] S3: Subtract the load prediction error and new energy prediction error of each prediction point to obtain the prediction result of the net load reserve capacity demand of each prediction point;

[0015] S4: Based on the prediction results of the net load reserve capacity demand of each prediction point, at a certain confidence level, according to the cumulative distribution function of the net load reserve capacity demand prediction error, obtain the upper and lower limits of the net load reserve capacity demand prediction interval of each prediction point;

[0016] Wherein, the cumulative distribution function of the net load reserve capacity demand prediction error is obtained by statistically analyzing the errors of the net load reserve capacity demand prediction results within a preset time period before the time period to be predicted.

[0017] Further, the prediction feature information includes the time, month, solar term, holiday, load prediction value and new energy prediction value corresponding to each prediction point.

[0018] Further, the step S2 specifically includes:

[0019] Divide the prediction feature information of each prediction point into load error prediction feature information and new energy error prediction feature information; the load error prediction feature information includes time, month, solar term, holiday, and load prediction value, and the new energy error prediction feature information includes time, month, solar term, holiday, and new energy prediction value;

[0020] Input the load error prediction feature information into the load prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error corresponding to each prediction point;

[0021] Input the new energy error prediction feature information into the new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the new energy prediction error corresponding to each prediction point.

[0022] Further, the load prediction error prediction model and the new energy prediction error prediction model are obtained through the following method:

[0023] Obtain historical multivariate time series samples containing load error prediction feature information and load prediction error, and obtain historical multivariate time series samples containing new energy error prediction feature information and new energy prediction error. Normalize the samples respectively to obtain the load prediction error training set and the new energy prediction error training set;

[0024] With the load error prediction feature information as the input and the load prediction error as the output, construct a load prediction error prediction model based on the CatBoost ensemble learning method, and use the load prediction error training set for training to obtain the final load prediction error prediction model;

[0025] With the new energy error prediction feature information as the input and the new energy prediction error as the output, construct a new energy prediction error prediction model based on the CatBoost ensemble learning method, and use the new energy prediction error training set for training to obtain the final new energy prediction error prediction model.

[0026] Further, the cumulative distribution function of the net load reserve capacity demand prediction error is obtained through the following method:

[0027] Use the constructed load / new energy prediction error prediction model to predict the load prediction error and new energy prediction error at each time point within the preset duration before the time period to be predicted;

[0028] Subtract the load prediction error and new energy prediction error at each time point to obtain the net load reserve capacity demand prediction result at each time point;

[0029] Obtain the true values of the net load reserve capacity requirements at each time point and compare them with the predicted results of the net load reserve capacity requirements at each time point to obtain the prediction errors of the net load reserve capacity requirements at each time point, and construct a prediction error database for the net load reserve capacity requirements;

[0030] Based on the prediction error database of the net load reserve capacity requirements, use the non-parametric kernel density estimation method to obtain the probability density function of the prediction errors of the net load reserve capacity requirements and the cumulative distribution function of the prediction errors of the net load reserve capacity requirements.

[0031] Furthermore, use the Gaussian kernel function to perform kernel density estimation on the prediction errors of the net load reserve capacity requirements. The probability density function of the prediction errors of the net load reserve capacity requirements constructed by the non-parametric kernel density estimation method is:

[0032]

[0033]

[0034] In the formula, n is the number of samples in the prediction error database of the net load reserve capacity requirements; ε i is the prediction error of the net load reserve capacity requirements of the i-th sample; h is the bandwidth; K(δ) is the Gaussian kernel function, and δ is the independent variable of the kernel function;

[0035] According to the probability density function f(ε) of the prediction errors of the net load reserve capacity requirements, integrate it to obtain the cumulative distribution function F(ε) of the prediction errors of the net load reserve capacity requirements.

[0036] Furthermore, the step S4 includes: arbitrarily given α, and 0 < α < 1. Under the confidence level of 1 - α, according to the cumulative distribution function of the prediction errors of the net load reserve capacity requirements, and combined with the predicted results of the net load reserve capacity requirements at each prediction point, obtain the prediction interval [L i , U i of the net load reserve capacity requirements that satisfies the confidence level of 1 - α at each prediction point, and is expressed as follows:

[0037]

[0038] In the formula, L i and U i are the upper and lower limits of the prediction interval of the net load reserve capacity requirements respectively; α1 = α / 2; α2 = 1 - α / 2; is the inverse function of the cumulative distribution function of the prediction errors of the net load reserve capacity requirements, and E pre is the predicted result of the net load reserve capacity requirements.

[0039] On the second aspect, a device for predicting the net load reserve capacity requirements based on CatBoost is provided, including:

[0040] A feature acquisition module, configured to acquire the prediction feature information of each prediction point within the to-be-predicted period;

[0041] A prediction error acquisition module, configured to input the prediction feature information of each prediction point into a load / new energy prediction error prediction model constructed based on the CatBoost ensemble learning method, and obtain the load prediction error and new energy prediction error corresponding to each prediction point;

[0042] A spare demand point prediction result acquisition module, configured to subtract the load prediction error and new energy prediction error of each prediction point to obtain the prediction result of the net load spare capacity demand of each prediction point;

[0043] A spare demand interval acquisition module, configured to obtain the upper and lower limits of the net load spare capacity demand prediction interval of each prediction point based on the prediction result of the net load spare capacity demand of each prediction point, at a certain confidence level, according to the cumulative distribution function of the net load spare capacity demand prediction error;

[0044] Wherein, the cumulative distribution function of the net load spare capacity demand prediction error is obtained by statistically analyzing the errors of the net load spare capacity demand prediction results within a preset time period before the to-be-predicted period.

[0045] In a third aspect, an electronic device is provided, including:

[0046] A memory, which stores a computer program;

[0047] A processor, configured to execute the computer program to implement the above-mentioned method for predicting the net load spare capacity demand based on CatBoost.

[0048] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program, and is characterized in that when the computer program is executed by a processor, the above-mentioned method for predicting the net load spare capacity demand based on CatBoost is implemented.

[0049] Beneficial effects

[0050] The present invention provides a method, device and storage medium for predicting the net load spare capacity demand based on CatBoost, having the following advantages:

[0051] (1) The research on wind power / solar power prediction and load prediction is relatively mature nowadays, but the prediction errors still cannot be overcome. With the large-scale grid connection of new energy, there is an urgent need to effectively evaluate the net load reserve demand. The present invention simultaneously considers the prediction errors of wind power / solar power and load, overcomes the influence of multiple prediction errors, and directly evaluates the net load reserve capacity demand. For a high-proportion new energy power system, the present invention fully considers the strong uncertainty of new energy, constructs a CatBoost model, analyzes the relationship between the net load reserve capacity demand and variables such as the predicted values of wind power / solar power and load, and improves the prediction accuracy of the net load reserve capacity demand.

[0052] (2) The main pain points solved by the present invention are to efficiently and reasonably process categorical features, gradient deviation, and prediction deviation problems, thereby reducing the occurrence of overfitting, and further improving the accuracy and generalization ability of the algorithm. Selecting variables that can comprehensively reflect the net load reserve capacity demand to construct the input feature set can improve the stability, accuracy, operation speed, and operation efficiency of the model.

[0053] (3) Due to the complex non-linear time-varying characteristics, large fluctuation characteristics, and strong uncertainty of the net load reserve capacity demand in a high-proportion new energy power system, the present invention performs probability prediction based on point prediction, and further provides the upward reserve and downward reserve at the moment to be predicted, so as to provide more complete and effective information for the power grid planning and dispatching departments and ensure the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 is the overall flowchart of the method for predicting the net load reserve capacity demand based on CatBoost in the embodiment of the present invention;

[0056] Figure 2 is the schematic diagram of the CatBoost algorithm provided by the embodiment of the present invention;

[0057] Figure 3 is the interval prediction result diagram of the method for predicting the daily net load reserve capacity demand based on CatBoost provided by Embodiment 1 of the present invention;

[0058] Figure 4 is the interval prediction result diagram of the method for predicting the daily net load reserve capacity demand based on CatBoost provided by Embodiment 2 of the present invention;

[0059] Figure 5 It is the interval prediction result graph of the intra-day net load reserve capacity demand prediction method based on CatBoost provided in Embodiment 3 of the present invention;

[0060] Figure 6 It is the interval prediction result graph of the intra-day net load reserve capacity demand prediction method based on CatBoost provided in Embodiment 4 of the present invention. Detailed implementation manners

[0061] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts fall within the scope protected by the present invention.

[0062] Aiming at the problem that the existing load reserve capacity demand prediction methods are difficult to make accurate predictions under the action of multiple uncertain factors such as wind / solar prediction and load prediction, the present invention provides a net load reserve capacity demand prediction method, device and storage medium based on CatBoost, which can achieve a relatively accurate assessment of the net load reserve capacity demand at two time scales of day-ahead (such as one day in advance) and intra-day (such as 4 hours in advance). The point prediction model of the load / new energy prediction error is constructed by using the CatBoost ensemble learning algorithm to obtain the point prediction result of the net load reserve capacity. Further, the probability density function and cumulative distribution function of the net load reserve capacity prediction error in each period are established by using non-parametric kernel density estimation, and the interval prediction result of the net load reserve capacity demand is obtained. The present invention can obtain the probability density function of the net load reserve capacity demand at the moment to be predicted and the prediction interval at any confidence level, with high prediction stability, accuracy, operation speed and operation efficiency, and can achieve dynamic estimation, which helps the grid operation dispatching personnel to formulate the starting mode in advance, identify when the system may be short of reserve, make emergency plans in advance, or take measures to improve the system stable operation ability.

[0063] The technical solutions of the present invention will be further described below in conjunction with several embodiments.

[0064] Embodiment 1

[0065] As Figure 1 shown, this embodiment provides a net load reserve capacity demand prediction method based on CatBoost, and the specific steps are as follows:

[0066] S1: Dataset construction and division. Construct a day-ahead net load reserve capacity demand prediction dataset for a certain province in the northwest from 2018 to 2021, including input feature xi and the corresponding output feature y i ; the input feature x i = [x i1 , x i2 , x i3 , x i4 , x i5 T is the historical moment x, month x, solar term x, whether it is a holiday x, and load / new energy prediction value x of a certain province on the predicted day obtained from the power grid control center i1 , month x i2 , solar term x i3 , whether it is a holiday x i4 and load / new energy prediction value x i5 ; the output feature y i is the load / new energy prediction error on the predicted day. It should be noted that when x represents the load prediction value, x i5 = [x i , x i1 , x i2 , x i3 , x i4 , x i5 T constitutes the load error prediction feature information, and the corresponding output feature y i represents the load prediction error; when x i5 represents the new energy prediction value, x i = [x i1 , x i2 , x i3 , x i4 , x i5 T constitutes the new energy error prediction feature information, and the corresponding output feature y i represents the new energy prediction error; both the load prediction value and the new energy prediction value are predicted based on existing technologies, and the present invention does not involve improvements to the technologies for predicting the load prediction value and the new energy prediction value herein. Taking 15 minutes as the time interval, a multivariate time series sample including information such as moment, month, solar term, holiday, and load / new energy prediction value and load / new energy prediction error is established, and the sample is normalized; in this embodiment, in order to verify the effects of the technical solutions of the present invention at the same time, the time series sample before February 18, 2021 is used as the training set, and this training set includes a load prediction error training set and a new energy prediction error training set; the time series sample from February 18 to 28, 2021 is used as the test set, and this test set includes a load prediction error test set and a new energy prediction error test set; rolling prediction is performed with a step size of 24h (96 moments).

[0067] ​​​S2: Based on the CatBoost ensemble learning method, train using the load prediction error training set and the new energy prediction error training set respectively to construct a load prediction error prediction model and a new energy prediction error prediction model; then use the load prediction error prediction model and the new energy prediction error prediction model to perform point predictions on the load prediction error and new energy prediction error day by day from February 18th to 28th, 2021. CatBoost is a GBDT framework with fewer parameters, supporting categorical variables and high accuracy, implemented based on symmetric decision trees as the base learners. The main pain point it addresses is the efficient and reasonable handling of categorical features. In addition, CatBoost also solves the problems of gradient bias and prediction drift, thereby reducing the occurrence of overfitting, and further improving the accuracy and generalization ability of the algorithm. Figure 2 This is the schematic diagram of the CatBoost ensemble learning algorithm and will not be elaborated here.

[0068] S3: Subtract the prediction results of the load prediction error and new energy prediction error obtained in S2 to obtain the prediction results of the net load reserve capacity demand at each moment day by day from February 18th to 28th, 2021. Define the load / new energy prediction error as the deviation between the true value and the predicted value of the load / new energy at a certain moment; define the net load reserve capacity demand R net as the load prediction error ε load and the new energy prediction error ε ne difference, that is: R net = ε load - ε ne .

[0069] S4: Before performing point predictions on the load prediction errors and new energy prediction errors day by day from February 18th to 28th, 2021, first use the load prediction error prediction model and the new energy prediction error prediction model to perform point predictions on the load prediction errors and new energy prediction errors at each moment within a preset time period (such as one year) before the prediction date, and then obtain the prediction results of the net load reserve capacity requirements at each moment; obtain the true values of the net load reserve capacity requirements at each moment (obtained by subtracting the new energy prediction error from the load prediction error in step S1), and compare them with the prediction results of the net load reserve capacity requirements at each moment to obtain the prediction errors of the net load reserve capacity requirements at each moment, thereby constructing a database of prediction errors of the net load reserve capacity requirements. Using the non-parametric kernel density estimation method, obtain the probability density function and cumulative distribution function of the prediction errors of the net load reserve capacity requirements. There are various structures for the kernel function, mainly divided into non-smooth kernels and smooth kernels. The kernel density estimation under the non-smooth kernel function cannot reflect the differences between adjacent data. In order to obtain a relatively smooth model, in this embodiment, the Gaussian kernel function is selected to perform kernel density estimation on the prediction errors of the net load reserve capacity requirements. The probability density function of the prediction errors of the net load reserve capacity requirements constructed by the kernel density estimation method is:

[0070]

[0071]

[0072] In the formula: n is the number of samples in the database of prediction errors of the net load reserve capacity requirements; ε i is the prediction error of the i-th sample; h is the bandwidth; K(δ) is the Gaussian kernel function, and δ is the independent variable of the kernel function.

[0073] According to the probability density function f(ε) of the prediction errors of the net load reserve capacity requirements, integrate it to obtain the cumulative distribution function F(ε) of the prediction errors of the net load reserve capacity requirements.

[0074] S5: Given α = 0.1, at a confidence level of 90%, according to the cumulative distribution function of the prediction errors of the net load reserve capacity requirements and combined with the prediction results of the net load reserve capacity requirements at each moment day by day from February 18th to 28th, 2021 obtained in step S3, the confidence intervals of the net load reserve capacity requirements at each moment day by day from February 18th to 28th, 2021 that meet the confidence level of 90% can be obtained L i and U i are the upper and lower limits of the prediction interval of the net load reserve capacity requirements respectively; α1 = 0.05; α2 = 0.95; is the inverse function of F(ε); E pre is the prediction result of the net load reserve capacity requirements at each moment day by day from February 18th to 28th, 2021.

[0075] To measure the reserve adequacy of the present invention during the entire scheduling period, an upward reserve coverage rate index is used to quantify the upward reserve safety margin. The calculation method of the upward reserve coverage rate index is as follows:

[0076]

[0077] In the formula, is used to represent the upward reserve coverage rate; N t is the number of scheduling periods existing during the scheduling period; c i is a 0-1 variable. If the upward reserve evaluation result is greater than the actual reserve demand of the power grid, then c i = 1; otherwise, c i = 0.

[0078] The Pearson Correlation Coefficient (PCC) is used as an index for the synchronous change of the reserve evaluation result and the actual reserve demand, and it is an important index to measure the coordination of two sets of data (evaluation result and actual demand). Its calculation method is as follows:

[0079]

[0080] In the formula, is the upward reserve synchronization coefficient during the scheduling period; and are respectively the actual reserve demand quantity in the i-th scheduling period, the average value and the standard deviation of the actual reserve demand during the entire scheduling period; and are respectively the evaluation result of the upward reserve capacity in the i-th period, the average value and the standard deviation of the evaluation results of the upward reserve capacity during the entire scheduling period.

[0081] The expression of the invalid reserve IR is:

[0082] The method of the present invention is compared with the traditional method, and the comparison results of the evaluation result indexes are shown in Table 1:

[0083] Table 1 Comparison of evaluation result indexes

[0084]

[0085] The interval prediction results of the day-ahead net load reserve capacity demand based on CatBoost from February 18th to 28th, 2021 are as Figure 3As shown, it can be seen that the traditional deterministic reserve criterion is too rough and cannot meet the upward reserve demand in some periods. Since the actual reserve capacity exceeds 1000 MW in many cases, to ensure safety, CatBoost evaluation can effectively reduce the upward reserve redundancy and has synchronous followability. While reducing ineffective reserves, the coverage rate and synchronous followability are better. As can be seen from Table 1, in Example 1 of the present invention, the upward reserve coverage rate is as high as 79.50%, the ineffective reserve is 530.77 MW, and the synchronous coefficient is 0.54.

[0086] Example 2

[0087] This embodiment provides a method for predicting the net load reserve capacity demand based on CatBoost. The difference from Example 1 is that when constructing the training set and the test set, the time series samples before April 12, 2021 are used as the training set, and the period from April 12 to 30, 2021 is used as the test set. The other principle processes are the same as those in Example 1 and will not be elaborated here.

[0088] The method of the present invention and the traditional method are compared, and the comparison results of the evaluation result indicators are shown in Table 2:

[0089] Table 2 Comparison of evaluation result indicators

[0090]

[0091] The prediction results of the daily net load reserve capacity demand interval based on CatBoost from April 12 to 30, 2021 are as Figure 4 shown. Combining with Table 2, in terms of comprehensive indicators, the CatBoost ensemble learning method of the present invention has better performance than the traditional method. In this scenario, the situation where the actual reserve capacity exceeds 1000 MW is relatively less than that in Example 1. Due to giving priority to safety, the reserve redundancy increases, but it has more excellent dynamic followability and upward reserve coverage rate, and can better avoid the implementation of orderly power consumption due to insufficient upward reserve in the power grid.

[0092] In summary, the method for evaluating the net load reserve capacity demand based on CatBoost of the present invention has good prediction accuracy, high coverage rate and synchronization at the daily time scale, and can meet the application scenarios.

[0093] Example 3

[0094] This embodiment provides a method for predicting the net load reserve capacity demand based on CatBoost, which is different from Embodiment 1 in that: since the power consumption load of the provincial power grid soared on August 2, 2021, the time series samples before August 1, 2021 are used as the training set, and the peak load period from August 1 to 7, 2021 is used as the test set, and rolling prediction is carried out with a step size of 4h (16 moments). The other principle processes are the same as those in Embodiment 1 and will not be elaborated here.

[0095] The method of the present invention is compared with the traditional method, and the comparison results of the evaluation result indicators are shown in Table 3:

[0096] Table 3 Comparison of evaluation result indicators

[0097]

[0098] The interval prediction results of the day-ahead net load reserve capacity demand based on CatBoost from August 1 to 7, 2021 are as Figure 5 shown. It can be seen from Figure 5 that the upward reserve and downward reserve coverage of the present invention are good and have followability; combined with the indicators in Table 3, while ensuring an upward reserve coverage rate of 96.58%, the present invention greatly reduces the ineffective reserve and has a high synchronization. It has excellent performance in terms of good followability, high accuracy and upward ineffective reserve, which is of great help for intraday scheduling in the peak load scenario in summer.

[0099] Embodiment 4

[0100] This embodiment provides a method for predicting the net load reserve capacity demand based on CatBoost, which is different from Embodiment 1 in that, since there have been many cases where traditional reserves could not meet the actual reserve demand in 2021, the time series samples before September 21, 2021 are used as the training set, and the period from September 21 to 27, 2021 when traditional reserves are insufficient is used as the test set, and rolling prediction is carried out with a step size of 4h (16 moments). The other principle processes are the same as those in Embodiment 1 and will not be elaborated here.

[0101] The method of the present invention is compared with the traditional method, and the comparison results of the evaluation result indicators are shown in Table 4:

[0102] Table 4 Comparison of evaluation result indicators

[0103]

[0104] The interval prediction results of the day-ahead net load reserve capacity demand based on CatBoost from September 21 to 27, 2021 are as Figure 6As shown in combination with Table 4, the synchronization coefficient of the present invention is 0.36, and the coverage rate is as high as 95.98%. When the traditional reserve is insufficient, it can follow the coverage well; when the traditional reserve is excessive, it can follow effectively without waste. The ineffective reserve of the present invention is greatly reduced compared with the traditional method, which can provide more effective information for the intraday scheduling of the power grid.

[0105] In summary, the method for evaluating the net load reserve capacity demand based on CatBoost of the present invention also has good prediction accuracy on the intraday time scale, with a relatively high coverage rate and synchronization, greatly reducing the ineffective reserve, and can meet the application scenarios.

[0106] Embodiment 5

[0107] This embodiment provides a device for predicting the net load reserve capacity demand based on CatBoost, including:

[0108] A feature acquisition module, configured to acquire the prediction feature information of each prediction point within the period to be predicted;

[0109] A prediction error acquisition module, configured to input the prediction feature information of each prediction point into a load / new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error and new energy prediction error corresponding to each prediction point;

[0110] A reserve demand point prediction result acquisition module, configured to subtract the load prediction error and new energy prediction error of each prediction point to obtain the prediction result of the net load reserve capacity demand of each prediction point;

[0111] A reserve demand interval acquisition module, configured to obtain the upper and lower limits of the net load reserve capacity demand prediction interval of each prediction point based on the prediction result of the net load reserve capacity demand of each prediction point and according to the cumulative distribution function of the net load reserve capacity demand prediction error at a certain confidence level;

[0112] Wherein, the cumulative distribution function of the net load reserve capacity demand prediction error is obtained by statistically analyzing the errors of the net load reserve capacity demand prediction results within a preset time period before the period to be predicted.

[0113] It should be understood that the functional unit modules in this embodiment can be concentrated in one processing unit, or each unit module can exist physically alone, or two or more unit modules can be integrated into one unit module, and can be implemented in the form of hardware or software.

[0114] Embodiment 6

[0115] This embodiment provides an electronic device, including:

[0116] A memory, which stores a computer program;

[0117] A processor for executing the computer program to implement the CatBoost-based net load reserve capacity demand prediction method as described above.

[0118] Embodiment 7

[0119] This embodiment provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the CatBoost-based net load reserve capacity demand prediction method as described above.

[0120] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0121] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0122] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0123] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are performed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1Steps of functions specified in one or more boxes.

[0124] It can be understood that the same or similar parts in the above embodiments can be referred to each other, and for the content not detailed in some embodiments, reference can be made to the same or similar content in other embodiments.

[0125] Any process or method description shown in the flowchart or described in other ways herein can be understood as representing a module, segment, or part of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present invention includes additional implementations, where the functions can be executed in a substantially simultaneous manner or in the reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the technical field to which the embodiments of the present invention belong.

[0126] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the net load reserve capacity requirement based on CatBoost, characterized in that, Including: S1: Obtain the prediction feature information of each prediction point during the period to be predicted; S2: Input the prediction feature information of each prediction point into the load / new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error and new energy prediction error corresponding to each prediction point; S3: Subtract the load prediction error and new energy prediction error of each prediction point to obtain the prediction result of the net load reserve capacity demand of each prediction point; S4: Based on the prediction result of the net load reserve capacity demand of each prediction point, at a certain confidence level, according to the cumulative distribution function of the net load reserve capacity demand prediction error, obtain the upper and lower limits of the net load reserve capacity demand prediction interval of each prediction point; Wherein, the cumulative distribution function of the net load reserve capacity demand prediction error is obtained by statistically analyzing the errors of the net load reserve capacity demand prediction results during a preset time period before the period to be predicted; The specific steps of S2 include: Divide the prediction feature information of each prediction point into load error prediction feature information and new energy error prediction feature information; the load error prediction feature information includes time, month, solar term, holiday, load prediction value, and the new energy error prediction feature information includes time, month, solar term, holiday, new energy prediction value; Input the load error prediction feature information into the load prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error corresponding to each prediction point; Input the new energy error prediction feature information into the new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the new energy prediction error corresponding to each prediction point; The cumulative distribution function of the net load reserve capacity demand prediction error is obtained by the following method: Use the constructed load / new energy prediction error prediction model to predict the load prediction error and new energy prediction error at each time point during a preset time period before the period to be predicted; Subtract the load prediction error and new energy prediction error at each time point to obtain the prediction result of the net load reserve capacity demand at each time point; Obtain the true value of the net load reserve capacity demand at each time point and compare it with the prediction result of the net load reserve capacity demand at each time point to obtain the prediction error of the net load reserve capacity demand at each time point, and construct a net load reserve capacity demand prediction error database; Based on the net load reserve capacity demand prediction error database, use the non-parametric kernel density estimation method to obtain the probability density function of the net load reserve capacity demand prediction error and the cumulative distribution function of the net load reserve capacity demand prediction error.

2. The method for predicting the net load reserve capacity requirement based on CatBoost according to claim 1, wherein, The prediction feature information includes the time, month, solar term, holiday, load prediction value and new energy prediction value corresponding to each prediction point.

3. The method for predicting the net load reserve capacity demand based on CatBoost according to claim 1, characterized in that, The load prediction error prediction model and the new energy prediction error prediction model are obtained by the following method: Obtain historical multivariate time series samples containing load error prediction feature information and load prediction errors, and obtain historical multivariate time series samples containing new energy error prediction feature information and new energy prediction errors, and normalize the samples respectively to obtain a load prediction error training set and a new energy prediction error training set; Taking the load error prediction feature information as the input and the load prediction error as the output, a load prediction error prediction model is constructed based on the CatBoost ensemble learning method and trained using the load prediction error training set to obtain the final load prediction error prediction model; Taking the new energy error prediction feature information as the input and the new energy prediction error as the output, a new energy prediction error prediction model is constructed based on the CatBoost ensemble learning method and trained using the new energy prediction error training set to obtain the final new energy prediction error prediction model.

4. The method for predicting the net load reserve capacity requirement based on CatBoost according to claim 1, wherein The Gaussian kernel function is used to perform kernel density estimation on the net load reserve capacity demand prediction error, and the probability density function of the net load reserve capacity demand prediction error constructed by the non-parametric kernel density estimation method is: where n is the number of samples in the net load reserve capacity demand prediction error database; ε i is the net load reserve capacity demand prediction error of the i-th sample; h is the bandwidth; K(δ) is the Gaussian kernel function, and δ is the independent variable of the kernel function; According to the probability density function f(ε) of the net load reserve capacity demand prediction error, its integral is taken to obtain the cumulative distribution function F(ε) of the net load reserve capacity demand prediction error.

5. The method for predicting the net load reserve capacity requirement based on CatBoost according to claim 4, wherein The step S4 includes: arbitrarily giving α, where 0 < α < 1, and under the confidence level of 1 - α, according to the cumulative distribution function of the net load reserve capacity demand prediction error, and combining the net load reserve capacity demand prediction results of each prediction point, obtaining the net load reserve capacity demand prediction interval [L i , U i that satisfies the confidence level of 1 - α for each prediction point, and is expressed as follows: where L i and U i are the upper and lower limits of the net load reserve capacity demand forecast interval respectively; α1 = α / 2; α2 = 1 - α / 2; is the inverse function of the cumulative distribution function of the net load reserve capacity demand forecast error, and E pre is the net load reserve capacity demand forecast result.

6. A net load reserve capacity demand prediction device based on CatBoost, characterized in that, Used to implement the net load reserve capacity demand prediction method based on CatBoost as described in claim 1, including: A feature acquisition module for acquiring the prediction feature information of each prediction point within the to-be-predicted time period; A prediction error acquisition module for inputting the prediction feature information of each prediction point into the load / new energy prediction error prediction model constructed based on the CatBoost ensemble learning method to obtain the load prediction error and new energy prediction error corresponding to each prediction point; A spare demand point prediction result acquisition module for subtracting the load prediction error and new energy prediction error of each prediction point to obtain the net load reserve capacity demand prediction result of each prediction point; A spare demand interval acquisition module for, based on the net load reserve capacity demand prediction results of each prediction point, obtaining the upper and lower limits of the net load reserve capacity demand prediction interval of each prediction point according to the cumulative distribution function of the net load reserve capacity demand prediction error at a certain confidence level; Wherein, the cumulative distribution function of the net load reserve capacity demand prediction error is obtained by statistically analyzing the prediction result errors of the net load reserve capacity demand within a preset time period before the to-be-predicted time period.

7. An electronic device, characterized in that, Including: A memory that stores a computer program; A processor for executing the computer program to implement the net load reserve capacity demand prediction method based on CatBoost as described in any one of claims 1 to 5.

8. A computer-readable storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the net load reserve capacity demand prediction method based on CatBoost as described in any one of claims 1 to 5.

Images