Multi-time scale load power prediction method and system

Abstract

The invention relates to a multi-time-scale load power prediction method and system, and the method comprises the following steps: obtaining the historical data of a load, and dividing the historical data into load data of a plurality of time scales; according to the load data of different time scales, a corresponding prediction algorithm is adopted, and a load prediction model of the corresponding time scale is constructed and trained; and loading the real-time data of the load into the load prediction model of the corresponding time scale to obtain a prediction result of the corresponding time scale. Compared with the prior art, the method has the advantages of giving consideration to prediction precision and prediction efficiency and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of energy systems, and in particular to a multi-time-scale load power forecasting method and system. Background Art

[0002] Under the general background of advocating green, environmental protection and low carbon, the proportion of new energy in the energy system is getting higher and higher. However, there are many challenges in the load of submerged arc furnace in the energy system. The power fluctuation of submerged arc furnace load is large, which leads to the instability and unreliability of the energy system. CN108647812A discloses a short-term power load forecasting method based on principal component adaptive BP neural network. The model is constructed by adaptive BP neural network method. The historical data of the power plant is used as input to establish a short-term forecasting method of the power plant load by combining the principal component with the adaptive BP neural network algorithm. This method reduces the amount of calculation, but because it ignores more features in the original data, the prediction accuracy is affected. CN102705303A discloses a method for fault location of hydraulic servo system based on residual and two-stage neural network. The initial data is directly used as input parameters, the amount of calculation is large, and the prediction efficiency is low. In view of the deficiencies in the aforementioned prior art, it is of great practical significance to accurately and efficiently predict the power of submerged arc furnace load. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-time scale load power forecasting method and system in order to overcome the defects of the above-mentioned prior art.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] A multi-time scale load power forecasting method comprises the following steps:

[0006] Obtain historical load data, and divide the historical data into load data of multiple time scales;

[0007] According to the load data of different time scales, the corresponding prediction algorithm is adopted to construct and train the load prediction model of the corresponding time scale;

[0008] The real-time load data is loaded into the load forecasting model of the corresponding time scale to obtain the forecast results of the corresponding time scale.

[0009] The load history data include the historical true value data of the submerged arc furnace, the amount of smelting material added, the quality of manganese ore entering the furnace, the temperature of the heating device and the upper limit of the submerged arc furnace load.

[0010] The historical load data is divided into load data of multiple time scales after data preprocessing, wherein the preprocessing includes data anomaly processing, data normalization and data resampling.

[0011] The data anomaly processing is specifically as follows: using the statistical method 3σ principle to detect outliers on the input data, and filling the detected abnormal data in one of four ways: direct deletion, historical simultaneous true value filling, previous and next term filling, and linear interpolation filling.

[0012] The filling method for abnormal data is determined by the filling effect backtesting verification method. The specific steps are as follows:

[0013] 1) Four methods are used to fill data respectively, 4 types of filling data are obtained, and candidate filling curves are obtained by fitting;

[0014] 2) obtaining a preset number of candidate load curves ranked from high to low in similarity to the historical ore-fired furnaces;

[0015] 3) Taking the average of the candidate load curves, we can obtain a real electric load curve of an electric arc furnace;

[0016] 4) Calculate the similarity between the candidate filling curve and the real electric load curve of the quasi-submersible arc furnace respectively, and select the filling method adopted by the candidate filling curve with the highest similarity as the final exception handling method.

[0017] The data resampling is configured through granularity according to the requirements of the time scale, and automatically performs upsampling or downsampling.

[0018] The load data of different time scales are used to train the load forecasting model after feature engineering and feature selection, wherein the feature engineering includes statistical features, statistical features at the same moment in a cycle, and domain knowledge features, and the feature selection adopts one of filtering, wrapping or embedding methods.

[0019] The statistical features include the maximum value, minimum value, slope, kurtosis, skewness, cumulative area, and peak-to-peak value features extracted based on the time window, wherein the slope is used to reflect the climbing rate and descending rate of the electric load of the blast furnace; the statistical features at the same moment in the cycle are the statistical features of the same moment points extracted within the same period, which are used to analyze the discrete degree and trend of load changes at specific time points; the domain knowledge features include production scheduling and maintenance planning. The production scheduling is divided into three types: reduced production, normal, and mass production. The maintenance plan is divided into two types: maintenance and no maintenance. The production scheduling and maintenance plan are encoded according to the type, and relevant domain features are extracted according to different types. Among them, the domain knowledge features of the production scheduling plan include load peak, load duration, load rising slope, load falling slope, area under the load curve, and load change smoothness.

[0020] The load forecasting model has two built-in algorithms, XGBoost and LightGBM. When production needs to meet real-time requirements, the forecasting model is switched to LightGBM for real-time load forecasting through parameter configuration. When production needs to meet model prediction accuracy requirements, the forecasting model is switched to XGBoost for real-time load forecasting through parameter configuration.

[0021] A multi-time scale load power forecasting system, comprising:

[0022] A data acquisition module is used to acquire historical load data and divide the historical data into load data of multiple time scales;

[0023] The prediction model construction and training module is used to adopt the corresponding prediction algorithm according to the load data of different time scales, and to construct and train the load prediction model of the corresponding time scale;

[0024] The prediction module is used to load the real-time load data into the load prediction model of the corresponding time scale to obtain the prediction result of the corresponding time scale.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) The present invention performs feature engineering and feature selection on the historical data of the electric arc furnace load to obtain effective load features, thereby training a load prediction model to improve the prediction accuracy of the model; at the same time, the load prediction model has two built-in algorithms, which can be selected according to the needs of actual applications, thereby taking into account both prediction accuracy and prediction efficiency.

[0027] (2) The present invention adopts four abnormal data filling methods when preprocessing load data, and determines the adopted method through the filling effect backtesting verification method, thereby reducing the error caused by improper abnormal processing and improving the model prediction accuracy.

[0028] (3) The statistical features extracted by the present invention in feature engineering can reflect the trend of electricity consumption changes in the electric arc furnace. By learning the past change characteristics of the sequence and discovering the operation rules of the sequence, the trend of the future sequence can be predicted. The statistical features extracted at the same time points in the same period are helpful for analyzing the discrete degree and trend of load changes at specific time points. In addition, the field characteristics of the electric arc furnace are also added, so that the characteristics of the electric arc furnace in this application scenario can be analyzed more accurately and specifically. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the method of the present invention;

[0030] Figure 2A flowchart of data preprocessing in one embodiment;

[0031] Figure 3 is a diagram of a resampling process in one embodiment;

[0032] Figure 4 It is a schematic diagram of load curves for different production scheduling plans;

[0033] Figure 5 A schematic diagram of a feature bucketing method of a histogram algorithm in an embodiment;

[0034] Figure 6 A schematic diagram of a Leaf-wise tree growth strategy in an embodiment;

[0035] Figure 7 It is a schematic diagram of a multi-time scale blast furnace load prediction process in an embodiment. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0037] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0038] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a kind of", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. The terms "first", "second", "third" and the like involved in the present application are merely used to distinguish similar objects and do not represent a specific ordering of the objects.

[0039] This embodiment provides a multi-time scale load power forecasting method, such as Figure 1 As shown, the following steps are included:

[0040] S1, obtaining historical load data, and dividing the historical data into load data of multiple time scales.

[0041] S11, obtain historical data of load

[0042] The input data for the prediction of the load of the submerged arc furnace is mainly data related to the submerged arc furnace. Ironmaking in the submerged arc furnace is a complex process, and there are many factors that affect power consumption. In order to reduce the complexity of the model as much as possible, this embodiment selects several data with large influencing factors as model input, mainly including the historical true value data of the submerged arc furnace, the amount of smelting material added, the quality of manganese ore entering the furnace, the temperature of the heating device, and the upper limit of the load of the submerged arc furnace.

[0043] Historical truth data: The model can learn the changing patterns of electricity consumption during ironmaking in the submerged arc furnace from the historical truth of the submerged arc furnace.

[0044] Smelting material charge amount: the actual mass of ore added to each submerged arc furnace, in kg or tons;

[0045] Quality of manganese ore added to the furnace: The quality of the ore added to the submerged arc furnace is expressed by the percentage of iron in the ore, which is generally 36%-40%. The quality of the ore will also affect the power consumption of the submerged arc furnace;

[0046] Heating device temperature: In the iron-making process, molten iron is usually added directly to the electric furnace for iron-making, which can greatly achieve energy-saving effects. The temperature of the molten iron added to the electric furnace is the heating device temperature;

[0047] Submerged arc furnace load upper limit: The rated power of the submerged arc furnace is used as the load prediction upper limit to constrain the prediction result from exceeding the upper limit.

[0048] S12, data preprocessing

[0049] The historical load data is divided into load data of multiple time scales after data preprocessing. In this embodiment, Figure 2 As shown, the preprocessing includes data resampling, data outlier processing and data normalization in sequence.

[0050] S121, data resampling

[0051] The implementation of data resampling function mainly depends on the specific business needs. By configuring the granularity in the initialization parameters, the algorithm automatically performs resampling. The sampling process is as follows: Figure 3 shown.

[0052] S122, Exception handling

[0053] Since the electric arc furnace scenario has high requirements for the real-time performance of the algorithm, this embodiment uses the statistical method 3σ principle to detect outliers on the input data. The 3σ principle refers to three times the standard deviation, that is, starting from the mean (μ) of the normal distribution, extending to both sides of the range of 3 standard deviations (σ), that is, the range between μ-3σ and μ+3σ, as the normal data range, and data outside this range is regarded as abnormal data.

[0054] The detected abnormal data is filled in by one of four methods: direct deletion, historical simultaneous true value filling, previous and next item filling, and linear interpolation filling.

[0055] This embodiment determines the filling method for filling abnormal data through the filling effect backtesting verification method, and the specific steps are as follows:

[0056] 1) Four methods are used to fill data respectively, 4 types of filling data are obtained, and candidate filling curves are obtained by fitting;

[0057] 2) Obtain the TopK candidate load curves ranked from high to low in similarity to the historical ore-fired furnaces, where K is a preset number;

[0058] 3) Taking the average of K candidate load curves, we can obtain a real electric load curve of an electric arc furnace;

[0059] 4) Calculate the similarity between the candidate filling curve and the real electric load curve of the submerged arc furnace (MAPE is selected as the similarity measurement indicator in this embodiment), and select the filling method used by the candidate filling curve with the highest similarity as the final exception handling method. The exception handling method proposed in the present invention can reduce the error caused by improper exception handling.

[0060] S123, data normalization

[0061] In this embodiment, Z-Score standardization is used to perform data normalization processing. In other embodiments, other normalization methods may also be used, which are not limited in this embodiment.

[0062] S13, Feature Engineering

[0063] In order to meet the real-time requirements of ferroalloy load prediction on site, a machine learning algorithm is used, and the machine learning algorithm requires manual feature extraction. Therefore, this embodiment extracts the maximum, minimum, slope, kurtosis, skewness, cumulative area, peak-to-peak value and other statistical features based on the time window, where the slope can reflect the climbing rate and downhill rate of the electric load of the submerged arc furnace, thereby capturing the trend of the change in electricity consumption of the submerged arc furnace. By learning the past change characteristics of the sequence and discovering the operation law of the sequence, the trend of the future sequence can be predicted; the electricity consumption trend of the submerged arc furnace changes periodically, generally 6 hours for one furnace, and the periodic characteristics are of great significance for modeling. Extracting the statistical features of the same time points in the same period helps to analyze the discrete degree and trend of the load change at a specific time point. The characteristics of the submerged arc furnace field mainly refer to other information such as production scheduling and maintenance planning. The production scheduling plan has three types: production reduction, normal, and mass production, and the maintenance plan has two types: maintenance and no maintenance. In order to enable the model to learn the domain characteristics of the ore-fired furnace, the production schedule, maintenance plan and other information are encoded. For example, the production reduction, normal production and mass production in the production schedule can be encoded as [0, 1, 2] respectively, and the maintenance and no maintenance in the maintenance plan are encoded as [0, 1]. The category characteristics of the domain are uniformly encoded and input into the model for learning. Among them, there are inconsistencies in the load change rules of each production type in the production schedule, but the area under the load curve remains consistent, such as Figure 4 As shown. When obtaining the production schedule for the next N days, extract relevant field features according to the production schedule type, automatically obtain historical load data under the corresponding production schedule type, and perform feature extraction based on the historical load data to extract load peak, load duration, load rising slope, load falling slope, area under the load curve, load change smoothness, etc. These features are crucial for load prediction of different production schedule types.

[0064] S14, feature selection

[0065] This embodiment embeds three feature selection methods: filtering, wrapping, and embedding. The significance of feature selection is to avoid the dimensionality disaster and eliminate irrelevant or redundant features, thereby reducing the number of features, improving model accuracy, and reducing the running time. It reduces the difficulty of learning tasks, selects truly relevant features to simplify the model, and assists in understanding the process of data generation. The filtering method evaluates the importance of features by the statistical characteristics of the features themselves, without considering the subsequent machine learning algorithm; the advantages of the filtering method are simple and efficient calculations, and can quickly eliminate obviously irrelevant features; the wrapping method is a feature selection method that combines the feature selection process with the machine learning algorithm, and evaluates the effectiveness of different feature subsets by training and testing a specific machine learning algorithm. The computational cost is high and affects the model efficiency; the embedding is a method that simultaneously performs feature selection during the training process of the machine learning algorithm, embeds the feature selection process into the model construction process, and uses some characteristics of the model itself for automatic feature selection, which requires specific machine learning algorithm support, and the selection results often depend on the specific model. In the scenario of an electric arc furnace, many parameters are input into the electric arc furnace. Feature engineering will be performed based on the original data, which will generate a large number of features. Considering the timeliness and algorithm modeling flexibility, this embodiment selects filtering as the feature screening method. By calculating the variance of the sample data, the features with low variance are removed, and the effective features are screened out using statistical information. At the same time, the features are also reduced in dimension.

[0066] S2, using the corresponding prediction algorithm according to the load data of different time scales, construct and train the load prediction model of the corresponding time scale.

[0067] In this embodiment, the load forecasting model has two built-in algorithms, XGBoost and LightGBM. XGBoost has higher prediction accuracy and LightGBM has faster speed. When production needs to meet real-time requirements, the prediction model is switched to LightGBM for real-time load forecasting through parameter configuration. When production needs to meet model prediction accuracy requirements, the prediction model is switched to XGBoost for real-time load forecasting through parameter configuration, which can meet the load forecasting requirements of different scenarios.

[0068] ①XGBoost, the full name of which is Extreme Gradient Boosting, is a gradient boosting model. The principle of the model is:

[0069] (1) Assuming that K trees are trained, the (final) prediction value for the i-th sample for:

[0070]

[0071] Among them, f k (xi ) is the model prediction result of the kth tree for sample i.

[0072] (2) The objective function is (loss function + model complexity (penalty))

[0073]

[0074] Among them, the first term is the loss function, and the second term is the regularization term, which represents the complexity of all K trees; is the predicted value, y i is the true value, n is the number of samples, and Ω is the complexity.

[0075] ②LightGBM is an efficient, fast, distributed gradient boosting (GB) framework based on the gradient boosting decision tree (GBDT). The main idea is to use weak classifiers (decision trees) for iterative training to obtain the optimal model. The model has the advantages of good training effect and low overfitting. The principle of LGBM algorithm:

[0076] (1) Based on histogram algorithm

[0077] The basic idea of ​​the histogram is to discretize the continuous feature data into K integers, construct a histogram with a width of K, and use the discretized values ​​as indexes to calculate statistics in the histogram when traversing the data. According to the discrete values ​​of the histogram, traverse to find the optimal segmentation point, such as Figure 5 shown.

[0078] (2) Leaf-wise algorithm

[0079] Each time, from all the leaf nodes of the current decision tree, find a leaf node with the largest split gain, and then split it, and repeat this cycle. Figure 6 shown.

[0080] (3) Single-side gradient sampling algorithm

[0081] The idea of ​​the unilateral gradient sampling algorithm is to sort the training set data according to the gradient, preset a ratio, and retain the data samples with gradients higher than this ratio among all samples; data samples with gradients lower than this ratio will not be discarded directly, but a sampling ratio will be set to proportionally extract samples from samples with smaller gradients. In order to compensate for the impact on the sample distribution, the unilateral gradient sampling algorithm will multiply the data set with smaller gradients by a coefficient when calculating the information gain to amplify it. In this way, when calculating the information gain, the algorithm can pay more attention to the sample data that is "not fully trained".

[0082] GBDT (Gradient Boosting Decision Tree), also known as Gradient Boosting Decision Tree, is an iterative decision tree algorithm, also known as MART (Multiple Additive Regression Tree). It constructs a set of weak learners (trees) and accumulates the results of multiple decision trees as the final prediction output. This algorithm effectively combines decision trees with ensemble ideas.

[0083] In order to better measure and select a load forecasting model suitable for the electric arc furnace scenario, this embodiment uses three measurement indicators: Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) to test the effect of the model.

[0084] MAPE is the average percentage difference between the predicted value and the actual observed value, which can be used to measure the relative size of the prediction error. The calculation formula of MAPE is:

[0085]

[0086] Among them, y i is the actual true value, is the predicted value, and n is the number of samples.

[0087] RMSE measures the standard deviation of the difference between the predicted value and the true value. The calculation formula of RMSE is:

[0088]

[0089] MAE is the mean absolute error between the predicted value and the true value, which has good interpretability. Its calculation formula is:

[0090]

[0091] S3, loading the real-time load data into the load forecasting model of the corresponding time scale to obtain the forecasting result of the corresponding time scale.

[0092] This embodiment combines the AI ​​algorithm in the field of artificial intelligence with the alchemy industry, and uses the algorithm to predict the power load of the submerged arc furnace, providing an important reference for production planning and scheduling. Figure 7As shown, to meet the multi-scale and multi-granularity prediction requirements of the load, it is only necessary to define the prediction length and granularity before the algorithm starts data processing, resample the data during data preprocessing to obtain the granularity data required for the task, and input the data into the model for feature extraction, calculate all historical data features, use the full amount or a certain length of features for training, and intercept the features of the latest specified prediction length for prediction.

[0093] The above is an introduction to the method embodiment. The following is a further explanation of the solution of the present invention through a system embodiment.

[0094] This embodiment provides a multi-time scale load power forecasting system, including:

[0095] A data acquisition module is used to acquire historical load data and divide the historical data into load data of multiple time scales;

[0096] The prediction model construction and training module is used to adopt the corresponding prediction algorithm according to the load data of different time scales, and to construct and train the load prediction model of the corresponding time scale;

[0097] The prediction module is used to load the real-time load data into the load prediction model of the corresponding time scale to obtain the prediction result of the corresponding time scale.

[0098] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0099] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A multi-time scale load power forecasting method, characterized in that: The following steps are involved: Obtain historical load data, and divide the historical data into load data of multiple time scales; According to the load data of different time scales, the corresponding prediction algorithm is adopted to construct and train the load prediction model of the corresponding time scale; The real-time load data is loaded into the load forecasting model of the corresponding time scale to obtain the forecast results of the corresponding time scale.

2. A multi-time scale load power forecasting method according to claim 1, characterized in that: The load history data include the historical true value data of the submerged arc furnace, the amount of smelting material added, the quality of manganese ore entering the furnace, the temperature of the heating device and the upper limit of the submerged arc furnace load.

3. The multi-time scale load power forecasting method according to claim 1, characterized in that: The historical load data is divided into load data of multiple time scales after data preprocessing, wherein the preprocessing includes data anomaly processing, data normalization and data resampling.

4. A multi-time scale load power forecasting method according to claim 3, characterized in that: The data anomaly processing is specifically as follows: using the statistical method 3σ principle to detect outliers on the input data, and filling the detected abnormal data in one of four ways: direct deletion, historical simultaneous true value filling, previous and next term filling, and linear interpolation filling.

5. A multi-time scale load power forecasting method according to claim 4, characterized in that: The filling method for abnormal data is determined by the filling effect backtesting verification method. The specific steps are as follows: 1) Four methods are used to fill data respectively, 4 types of filling data are obtained, and candidate filling curves are obtained by fitting; 2) obtaining a preset number of candidate load curves ranked from high to low in similarity to the historical ore-fired furnaces; 3) Taking the average of the candidate load curves, we can obtain a real electric load curve of an electric arc furnace; 4) Calculate the similarity between the candidate filling curve and the real electric load curve of the quasi-submersible arc furnace respectively, and select the filling method adopted by the candidate filling curve with the highest similarity as the final exception handling method.

6. A multi-time scale load power forecasting method according to claim 3, characterized in that: The data resampling is configured through granularity according to the requirements of the time scale, and automatically performs upsampling or downsampling.

7. The multi-time scale load power forecasting method according to claim 1, characterized in that: The load data of different time scales are used to train the load forecasting model after feature engineering and feature selection, wherein the feature engineering includes statistical features, statistical features at the same moment in a cycle, and domain knowledge features, and the feature selection adopts one of filtering, wrapping or embedding methods.

8. A multi-time scale load power forecasting method according to claim 7, characterized in that: The statistical features include the maximum value, minimum value, slope, kurtosis, skewness, cumulative area, and peak-to-peak value features extracted based on the time window, wherein the slope is used to reflect the climbing rate and descending rate of the electric load of the blast furnace; the statistical features at the same moment in the cycle are the statistical features of the same moment points extracted within the same period, which are used to analyze the discrete degree and trend of load changes at specific time points; the domain knowledge features include production scheduling and maintenance planning. The production scheduling is divided into three types: reduced production, normal, and mass production. The maintenance plan is divided into two types: maintenance and no maintenance. The production scheduling and maintenance plan are encoded according to the type, and relevant domain features are extracted according to different types. Among them, the domain knowledge features of the production scheduling plan include load peak, load duration, load rising slope, load falling slope, area under the load curve, and load change smoothness.

9. The multi-time scale load power forecasting method according to claim 1, characterized in that: The load forecasting model has two built-in algorithms, XGBoost and LightGBM. When production needs to meet real-time requirements, the forecasting model is switched to LightGBM for real-time load forecasting through parameter configuration. When production needs to meet model prediction accuracy requirements, the forecasting model is switched to XGBoost for real-time load forecasting through parameter configuration.

10. A multi-time scale load power forecasting system, characterized in that: include: A data acquisition module is used to acquire historical load data and divide the historical data into load data of multiple time scales; The prediction model construction and training module is used to adopt the corresponding prediction algorithm according to the load data of different time scales, and to construct and train the load prediction model of the corresponding time scale; The prediction module is used to load the real-time load data into the load prediction model of the corresponding time scale to obtain the prediction result of the corresponding time scale.

Images