Load rate trend prediction method and system based on power grid operation data

Abstract

The invention discloses a load rate trend prediction method and system based on power grid operation data, and relates to the technical field of power load prediction, and the method comprises the steps: creating a data set, and dividing the data set into a training set and a verification set; a Prophet model and a GBDT model are constructed, and a prediction result is obtained; and sorting prediction results into a new data set, training a Meta model to obtain a final fusion prediction result, and substituting the final fusion prediction result into a load rate formula to calculate a predicted load rate. According to the method, two different types of models, namely Prophet and GBDT, are combined, so that model ensemble is realized, and the prediction accuracy can be improved. Prophet is suitable for capturing a trend and a periodic mode of a time sequence, and GBDT is suitable for fitting a complex nonlinear relationship. The combination of the two enables the model to be more powerful and universal. By constructing a new feature data set and training a Meta model to integrate Prophet and GBDT prediction results, complementary information between the two models can be learned, and better prediction is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric load prediction, and in particular to a method and system for predicting a load rate trend based on power grid operation data. Background Art

[0002] With the development of power systems, the importance of power load forecasting has become increasingly prominent. Accurate power load forecasting is of great significance to the economic operation and safe dispatch of power grids. Traditional power load forecasting mainly relies on manual experience, and the accuracy of the forecast results is difficult to guarantee. In order to improve the accuracy of power load forecasting, many scholars have explored various new forecasting methods.

[0003] At present, the more common methods for power load forecasting include regression analysis, time series method, neural network method, etc. The regression analysis method establishes a statistical model based on various factors that affect the load and makes predictions through the quantitative relationship between variables. However, this method has a poor fitting effect on the nonlinear relationship of the complex. Time series methods such as ARMA and ARIMA use the laws of the time series itself for prediction, which can better reflect the periodic change law of the load, but the impact of random factors is not considered enough. Neural network methods such as BP neural network have self-learning and adaptive capabilities and can fit complex nonlinear relationships, but their "black box" operation mechanism makes the results difficult to explain. Summary of the invention

[0004] In view of the problems existing in the existing power load forecasting, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the traditional power load prediction mainly relies on manual experience for prediction, the accuracy of the prediction result is difficult to guarantee, and the nonlinear fitting in the prior art is poor.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting a load rate trend based on power grid operation data, which comprises:

[0008] Create a data set and divide it into a training set and a validation set;

[0009] Construct Prophet and GBDT models to obtain prediction results;

[0010] The prediction results are organized into a new data set, the Meta model is trained, the final fusion prediction results are obtained, and they are substituted into the load rate formula to calculate the predicted load rate.

[0011] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, the Prophet model is shown as follows:

[0012] y(t)=g(t)+s(t)+h(t)

[0013] Among them, g(t) represents the trend term, which indicates the trend of the time series in the non-periodic aspect; s(t) represents the periodic term, or seasonal term, which is measured in weeks or years; h(t) represents the holiday term, which indicates the impact of potential holidays with non-fixed periods in the time series on the predicted value;

[0014] The trend term is based on a piecewise linear function as shown below:

[0015] g(t)=(k+a(t) T δ)·t+(m+a(t) T γ)

[0016] a(t)=(a 1 (t),...,a s (t)) T ,δ=(δ 1 ,...,δ S ) T ,γ=(γ 1 ,...,γ S ) T

[0017] Among them, k represents the growth rate, δ represents the change in the growth rate, m represents the offset, and γ is the offset corresponding to each changepoint.

[0018] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, the Fourier series form of the periodic term is as follows:

[0019]

[0020] Where N represents the number of such periods that you want to use in the model, and P represents the period of the time series;

[0021] When P=7,N=10:

[0022]

[0023] When P=7, N=3:

[0024]

[0025] The periodic term is shown as follows:

[0026] s(t)=X(t)β

[0027] Among them, the initialization of β is β~Normal(0,σ 2 ), X(t) is the periodic function of the time series, and β represents the seasonal effect.

[0028] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, the holiday item is as follows:

[0029]

[0030] in, and κ=(κ 1 ,...,κ L ) T , D i represents the period before and after the holiday, κ i It represents the impact range of holidays, L is the number of holidays, Z(t) is the time representation of holidays, and t is the number of holidays.

[0031] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, wherein: the GDBT model includes initializing the learner, calculating the residual, calculating the best fit value and updating the strong learner;

[0032] The initialization learner is shown as follows:

[0033]

[0034] Among them, F 0 (x) is the model of the 0th tree, Y i is the true value, and c is the residual.

[0035] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, the calculation residual is shown as follows:

[0036]

[0037] Among them, M is the data training set, and the maximum number of iterations is m;

[0038] The calculation of the best fit value is shown as follows:

[0039]

[0040] Among them, y i is the true value, f m-1 (x i )+γ is the residual;

[0041] The updated strong learner is shown as follows:

[0042]

[0043] The final GDBT model is obtained by arranging the above, as shown in the following formula:

[0044]

[0045] As a preferred solution of the load rate trend prediction method based on power grid operation data of the present invention, obtaining the final fusion prediction result includes the following steps:

[0046] The prediction results of the GDBT model and the Prophet model are used as new features to construct a data set;

[0047] Train the Meta model using the newly constructed dataset;

[0048] Verify the Meta model and adjust hyperparameters;

[0049] Use Meta models to make predictions on unknown data.

[0050] In a second aspect, an embodiment of the present invention provides a load rate trend prediction system based on power grid operation data, which includes:

[0051] The data preparation module is used to build a training data set. By accessing multi-source heterogeneous data, a data set containing historical load, weather, and holiday characteristics is generated.

[0052] The base model training module is used to train the two basic prediction models of Prophet and GBDT;

[0053] The model fusion module is used to take the prediction results of the base model as new features, train the Meta regression model to fuse the results, and obtain the final prediction;

[0054] The load rate calculation module is used to calculate the predicted load rate according to the current value and the line parameters.

[0055] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, it implements any step of the above-mentioned load rate trend prediction method based on power grid operation data.

[0056] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned method for predicting a load rate trend based on power grid operation data is implemented.

[0057] The beneficial effect of the present invention is that it combines two different types of models, Prophet and GBDT, to achieve a model ensemble, which can improve the accuracy of prediction. Prophet is suitable for capturing trends and periodic patterns in time series, and GBDT is suitable for fitting complex nonlinear relationships. The combination of the two can make the model more powerful and versatile. By constructing a new feature data set and training the Meta model to integrate the prediction results of Prophet and GBDT, the complementary information between the two models can be learned to achieve better predictions. In addition, the Prophet model adds holiday terms, which can better consider the impact of holidays on load and improve the accuracy of predictions. The Prophet model adds holiday terms, which can better consider the impact of holidays on load and improve the accuracy of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0059] Figure 1 The figure is a flow chart of a method for predicting load rate trend based on power grid operation data.

[0060] Figure 2 This is a comparison chart between the predicted value and the true value in the GBDT algorithm regression task of the load rate trend prediction method based on power grid operation data.

[0061] Figure 3 The input and output diagram of the Prophet algorithm for load rate trend prediction based on power grid operation data. DETAILED DESCRIPTION

[0062] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0063] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0064] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0065] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0066] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0067] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0068] Example 1

[0069] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a load rate trend prediction method based on power grid operation data, comprising:

[0070] S1: Create a data set and divide it into a training set and a validation set;

[0071] By accessing the massive amount of power grid operation data in the past five years, the power grid operation data is matched with the power grid model data. In the model data, the line data and the main transformer data are extracted. The line ID in the operation data is matched with the line ID, and the main transformer ID is matched with the main transformer ID in the operation data to extract the current data of all lines and main transformers. Through the historical date data, the holiday data, weather data, and major festival / activity data in the past five years are obtained, and the data sets of occurrence time, current, equipment, date type, start time, end time, and weather data are generated in chronological order.

[0072] S2: Build Prophet and GBDT models to get prediction results;

[0073] The Prophet algorithm uses a decomposable time series model, which has three main model components: growth (growth trend), seasonality (seasonal trend), and holidays (the impact of holidays on the predicted value). Therefore, the focus is on how to calculate these three parts. The calculation formula is as follows:

[0074] y(t)=g(t)+s(t)+h(t)

[0075] Among them, g(t) represents the trend term, which represents the non-periodic trend of the time series; s(t) represents the periodic term, or seasonal term, which is generally in weeks or years; h(t) represents the holiday term, which represents the impact of potential non-fixed-periodic holidays in the time series on the predicted value.

[0076] The g(t) trend term is based on a piecewise linear function (linear growth), and its formula is as follows:

[0077] g(t)=(k+a(t) T δ)·t+(m+a(t) T γ)

[0078] a(t)=(a 1 (t),...,a s (t)) T ,δ=(δ 1 ,...,δ S ) T ,γ=(γ 1 ,...,γ S ) T

[0079] Among them, k represents the growth rate, δ represents the change in the growth rate, and m represents the offset.

[0080] It is assumed that S change points have been placed and the position of the change point is at timestamp S j,1≤j≤S, then at these timestamps, we need to give the change in growth rate, that is, at timestamp S j We can assume that there is such a vector: δ∈R S , where δ i Indicates that at timestamp S j If the initial growth rate is replaced by k, then the growth rate at timestamp t is Through an indicator function a(t)∈{0,1} S ,that is The growth rate at timestamp t is k+a(t) T δ.

[0081] In the Prophet algorithm, there are three important indicators:

[0082] changepoint_range: the location of the change point

[0083] n_changepoint: the number of change points

[0084] changepoint_prior_scale: rate of change of growth

[0085] Among them, changepoint_range refers to the percentage. It is necessary to set the change point in the time series as long as the previous changepoint_range. In the default function, changepoint_range = 0.8. n_changepoint represents the number of change points. In the default function, n_changepoint = 25. changepoint_prior_scale represents the distribution of the change point growth rate, δ j ~Laplace(0,τ), where τ is change_point_scale.

[0086] In this patent, the selection of change points is based on the historical data of the first 80% of the time series, and then 25 change points are found by equal division, and the growth rate of the change points satisfies the Laplace distribution δ j ~Laplace(0,0.05). Therefore, when τ approaches zero, δ j It also tends to zero, and the growth function at this time will become a linear function of the entire segment.

[0087] s(t) is the periodic term. Since the time series may contain seasonal trends of various periodic types such as days, weeks, months, and years, the Fourier series can be used to approximate this periodic property.

[0088] Use Fourier series to simulate the periodicity of time series: Assume that P represents the period of the time series, P = 365.25 represents a yearly period, and P = 7 represents a weekly period.

[0089] Its Fourier series is in the form of:

[0090]

[0091] Among them, N represents the number of such cycles that we hope to use in the model. A larger N value can fit a more complex seasonal function, but it will also bring more overfitting problems.

[0092] According to empirical values, for a sequence with a yearly period (P = 365.25), N = 10; for a sequence with a weekly period (P = 7), N = 3

[0093] When N=10,

[0094]

[0095] When N=3,

[0096]

[0097] Therefore, the seasonal term of the time series is:

[0098] s(t)=X(t)β

[0099] Among them, the initialization of β is β~Normal(0,σ 2 ), where σ is controlled by seasonality_prior_scale, that is, σ = seasonality_prior_scale. The larger the value, the more obvious the seasonal effect; the smaller the value, the less obvious the seasonal effect. In the code, seasonality_mode also corresponds to two modes, namely addition and multiplication, and the default is addition.

[0100] h(t) Holiday effect. In actual power grid operation, holidays or some major events will have a great impact on the load rate of the power grid, and these time points often do not have periodicity. Analysis of these points is extremely necessary. This patent collects special holidays. In addition to holidays, it also includes major events in previous years, special power consumption periods, and power supply guarantee periods, such as: Double Eleven, College Entrance Examination, etc. Because each holiday or a known major event has a different degree of influence on the time series, such as the Spring Festival and National Day are seven-day holidays, while Labor Day and other holidays are shorter. Therefore, the holiday model regards the impact of different holidays at different time points as independent models, and different before and after window values ​​can be set for different holidays, indicating that the holiday will affect the time series of a period of time before and after.

[0101] For the i-th holiday, D i In order to express the holiday effect, a corresponding indicator function is needed, and a parameter κ is also needed. i To indicate the impact of holidays.

[0102] Assuming there are L holidays, the holiday effect model is:

[0103]

[0104] in, and κ=(κ 1 ,...,κ L ) T .

[0105] Among them, κ~Normal(0,v 2 ) and the normal distribution is affected by the indicator v = holidays_prior_scale. The default value is 10. When the value is larger, it means that holidays have a greater impact on the model; when the value is smaller, it means that holidays have a smaller effect on the model. This parameter can be adjusted by yourself.

[0106] Furthermore, the data obtained in step 1 is imported into the GBDT model to obtain a predictive GBDT model.

[0107] GDBT is Boosted trees, which is an additive model composed of k base models. Assuming that the tree model to be trained in the tth iteration is ft(xi), then:

[0108]

[0109] First initialize the learner:

[0110]

[0111] For m=1,2,…,M, for each sample i=1,2,…,Ni i=1,2,…,Ni=1,2,…,N, calculate the negative gradient, that is, the residual:

[0112]

[0113] The residual obtained in the previous step is used as the new true value of the sample, and the data (xi, rim), i = 1, 2, ..., N is used as the training data of the next tree, and a new regression tree fm (x) is obtained, and its corresponding leaf node area is Rjm, j = 1, 2, ..., J. Where J is the number of leaf nodes of the regression tree t.

[0114] Calculate the best fit value for the leaf region j=1,2,..,J:

[0115]

[0116] Update the strong learner:

[0117]

[0118] The final result after learning is as follows:

[0119]

[0120] Bring in a large amount of historical data and improve the accuracy of the algorithm after model training

[0121] S3: Organize the prediction results into a new data set, train the Meta model, obtain the final fusion prediction results, and substitute them into the load rate formula to calculate the predicted load rate.

[0122] The prediction results of Prophet and GBDT are used as new features and merged into a new dataset. The new dataset is used to train a Meta model (secondary model), whose input is the prediction results of Prophet and GBDT.

[0123] Prepare the dataset and create a new dataset that contains the prediction results of the base models (Prophet and GBDT) as features, as well as the actual target values ​​(ground truth).

[0124] Divide the dataset: Divide the dataset into a training set and a validation set. The training set is used to train the base model, and the validation set is used to train the Meta model and adjust hyperparameters.

[0125] Training base model: Use the training set to train the Prophet and GBDT models to obtain their prediction results.

[0126] Construct Meta features: Use the prediction results of the base model as new features to construct a new dataset. The feature of each sample is the prediction result of the base model, and the target value is the actual target value.

[0127] Training Meta model: Use the newly constructed data set (including the base model prediction results and target values) to train the Meta model. Meta model This patent uses a linear regression model.

[0128] Validation and Adjustment: Use the validation set to validate the Meta model and adjust the hyperparameters of the Meta model based on performance metrics (mean square error, accuracy). This can be done through techniques such as cross-validation.

[0129] Prediction: Use the Meta model to predict unknown data. In the prediction stage, first use Prophet and GBDT to obtain the prediction results of the base model, and then use these results as the input of the Meta model to obtain the final fusion prediction results.

[0130] Through the load rate calculation formula: line load rate = current value / current limit*100%, main transformer load rate = current value / main transformer rated current, substitute the predicted value into the calculation to predict the load rate.

[0131] Furthermore, this embodiment also provides a load rate trend prediction system based on power grid operation data, including:

[0132] The data preparation module is used to build a training data set. By accessing multi-source heterogeneous data, a data set containing historical load, weather, and holiday characteristics is generated.

[0133] The base model training module is used to train the two basic prediction models of Prophet and GBDT;

[0134] The model fusion module is used to take the prediction results of the base model as new features, train the Meta regression model to fuse the results, and obtain the final prediction;

[0135] The load rate calculation module is used to calculate the predicted load rate according to the current value and the line parameters.

[0136] This embodiment also provides a computer device, which is suitable for the case of a load rate trend prediction method based on power grid operation data, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the load rate trend prediction method based on power grid operation data as proposed in the above embodiment.

[0137] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0138] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting a load rate trend based on power grid operation data as proposed in the above embodiment is implemented.

[0139] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0140] Example 2

[0141] Reference Figure 1 to Figure 3 , which is the second embodiment of the present invention, and this embodiment provides a load rate trend prediction method based on power grid operation data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0142] In order to verify the effectiveness of the proposed load rate prediction method based on Prophet and GBDT models, we collected the power grid operation data of a certain region in the past five years as an experimental data set. The data set contains 300 samples, each of which includes features such as "date", "load current", "line number", "main transformer number", "weather", and "holiday". In order to construct the holiday effect term h(t) in the Prophet model, we collected the statutory holiday dates in the past five years. We used the data of the past four years as the training set and the data of the most recent year as the test set.

[0143] During the model training process, we set the relevant hyperparameters of the Prophet model as: changepoint_prior_scale=0.05, seasonality_prior_scale=10, holidays_prior_scale=0.1. This can enhance the Prophet model's ability to fit the seasonality and holiday effects of the data. We set the GBDT model to 50 decision trees, with a maximum depth of 5 and a learning rate of 0.1. This can prevent overfitting. We chose linear regression for the Meta model.

[0144] The specific experimental steps are as follows:

[0145] The collected power grid operation dataset is split into training set and test set in a ratio of 7:3.

[0146] The Prophet model and the GBDT model are trained using the training set respectively, and the load current prediction results of the two models for the test set samples are obtained.

[0147] The predicted currents of the two models are used as new features and combined with the actual currents of the test set samples to form a new data set.

[0148] Train a linear regression Meta model using the new dataset.

[0149] The Meta model is used to predict the test set to obtain the final predicted current.

[0150] Substitute the predicted current into the formula to calculate the predicted line load factor and main transformer load factor.

[0151] The prediction errors of different models were calculated, and the prediction effects of this method were compared with those of a single Prophet or GBDT model.

[0152] The experimental results are shown in the following table:

[0153] Table 1 Experimental results

[0154] Sample No. Actual current / A Prophet predicted current / A GBDT predicted current / A Meta model predicted current / A 1 150 155 152 149 2 160 158 163 161 ... ... ... ... ... 298 176 173 178 175 299 172 169 175 171 300 168 166 170 167

[0155] It can be seen from the data in the table that the predicted values ​​of the current by the Prophet model, GBDT model and Meta fusion model are relatively close to the actual values, which verifies that the designed method can better predict the load current.

[0156] Among them, the Prophet model takes into account the impact of periodicity and holidays, and has a better prediction effect for some special time points. However, when the load pattern has complex changes, the prediction error of the Prophet model is large.

[0157] The GBDT model can be continuously optimized through iterative training, and has a stronger ability to fit complex changes in loads. However, the algorithm itself has a certain variance error.

[0158] The Meta fusion model combines the advantages of the two models and models both the overall trend and local details of load changes. The experimental results show that the Meta model has the lowest overall prediction current error, which verifies the conclusion that model fusion can improve prediction accuracy.

[0159] After calculating the load factor, it can be found that the load factor prediction error of the Meta model is more than 10% lower than that of the single model. This further proves that this method can significantly improve the accuracy of load prediction, provide support for safe and economical dispatch of power grids, and has great practical value.

[0160] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting load rate trend based on power grid operation data, characterized in that: include, Create a data set and divide it into a training set and a validation set; Construct Prophet and GBDT models to obtain prediction results; The prediction results are organized into a new data set, the Meta model is trained, the final fusion prediction results are obtained, and they are substituted into the load rate formula to calculate the predicted load rate.

2. The method for predicting load rate trend based on power grid operation data according to claim 1, characterized in that: The Prophet model is shown as follows: y(t)=g(t)+s(t)+h(t) Among them, g(t) represents the trend term, which indicates the changing trend of the time series on the non-periodic side, s(t) represents the periodic term, and h(t) represents the holiday term. The trend term is based on a piecewise linear function as shown below: g(t)=(k+a(t) T δ)·t+(m+a(t) T γ) a(t)=(a1(t),...,a s (t)) T ,δ=(δ1,...,δ S ) T ,γ=(γ1,...,γ S ) T Among them, k represents the growth rate, δ represents the change in the growth rate, m represents the offset, and γ is the offset corresponding to each changepoint.

3. The method for predicting load rate trend based on power grid operation data according to claim 2, characterized in that: The Fourier series form of the periodic term is shown below: Where N represents the number of such periods that you want to use in the model, and P represents the period of the time series; When P=7,N=10: When P=7, N=3: The periodic term is shown as follows: s(t)=X(t)β Among them, the initialization of β is β~Normal(0,σ 2 ), X(t) is the periodic function of the time series, and β represents the seasonal effect.

4. The method for predicting load rate trend based on power grid operation data according to claim 3, characterized in that: The holiday item is shown as follows: in, and κ=(κ1,...,κ L ) T , D i represents the time T before and after the holiday, κ i It represents the impact range of holidays, L is the number of holidays, Z(t) is the time representation of holidays, and t is the number of holidays.

5. The method for predicting load rate trend based on power grid operation data according to claim 4, characterized in that: The GDBT model includes initializing the learner, calculating the residual, calculating the best fit value, and updating the strong learner; The initialization learner is shown as follows: Among them, F0(x) is the model of the 0th tree, Y i is the true value, and c is the residual.

6. The method for predicting load rate trend based on power grid operation data according to claim 5, characterized in that: The calculation residual is shown as follows: Among them, M is the data training set, and the maximum number of iterations is m; The calculation of the best fit value is shown as follows: Among them, y i is the true value, f m-1 (x i )+γ is the residual; The updated strong learner is shown as follows: The final GDBT model is obtained by arranging the above, as shown in the following formula:

7. The method for predicting load rate trend based on power grid operation data according to claim 6, characterized in that: Obtaining the final fusion prediction result includes the following steps: The prediction results of the GDBT model and the Prophet model are used as new features to construct a data set; Train the Meta model using the newly constructed dataset; Verify the Meta model and adjust hyperparameters; Use Meta models to make predictions on unknown data.

8. A load rate trend prediction system based on power grid operation data, based on the load rate trend prediction method based on power grid operation data according to any one of claims 1 to 7, characterized in that: include, The data preparation module is used to build a training data set. By accessing multi-source heterogeneous data, a data set containing historical load, weather, and holiday characteristics is generated. The base model training module is used to train the two basic prediction models of Prophet and GBDT; The model fusion module is used to take the prediction results of the base model as new features, train the Meta regression model to fuse the results, and obtain the final prediction; The load rate calculation module is used to calculate the predicted load rate according to the current value and the line parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the load rate trend prediction method based on power grid operation data according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the load rate trend prediction method based on power grid operation data according to any one of claims 1 to 7 are implemented.

Images