A power system load prediction method, system, device and medium based on meteorological big data

Abstract

This invention discloses a method, system, equipment, and medium for power system load forecasting based on meteorological big data. The method includes: acquiring historical load data and meteorological data; constructing meteorological enhancement features based on the meteorological data; generating load classification features based on the response relationship between the historical load data and the meteorological enhancement features; inputting the historical load data, the meteorological enhancement features, and the load classification features into a backbone forecasting network to extract time-series forecast features and obtain backbone forecast features; fusing the backbone forecast features and the meteorological enhancement features to generate initial short-term power load forecast values; and constraining the initial short-term power load forecast values ​​to obtain the final short-term power load forecast result, thereby improving the engineering usability of the forecast results.

Description

Technical Field

[0001] This invention relates to the field of power system load forecasting technology, and in particular to a method, system, equipment and medium for power system load forecasting based on meteorological big data. Background Technology

[0002] As the power system's electricity consumption structure and load characteristics continue to change, short-term power load forecasting plays an increasingly important role in grid dispatching, operation control, electricity planning, and load management. Electricity load is not only influenced by historical electricity consumption patterns but also by meteorological factors such as temperature, humidity, wind speed, rainfall, solar radiation intensity, irradiance, air pressure, and weather type. Particularly under meteorological conditions such as high temperatures, low temperatures, continuous rain, and persistent high humidity, air conditioning, heating, cooling, lighting, and some production and residential loads will change significantly, making short-term load curves more volatile and uncertain. Therefore, in power system load forecasting, how to fully utilize meteorological big data and accurately characterize the impact of meteorological factors on load changes is a crucial issue for improving the accuracy and stability of short-term load forecasting.

[0003] Current short-term power load forecasting technologies still have shortcomings. Some methods rely heavily on historical load series, failing to adequately consider meteorological factors. Even when meteorological variables such as temperature and humidity are introduced, they are often treated as ordinary input features, making it difficult to characterize temperature-humidity coupling, high and low temperature threshold responses, the cumulative effects of continuous meteorological conditions, and meteorological lag effects. Furthermore, different load types exhibit varying sensitivities to meteorological factors, and single time-series models struggle to simultaneously account for local fluctuations, long-term dependencies, and the identification of key time steps, resulting in insufficient forecast accuracy and stability under complex meteorological conditions.

[0004] To address the aforementioned issues, this invention proposes a power system load forecasting model based on meteorological big data. This model uses historical load data and meteorological data as primary inputs. First, it constructs meteorological coupling features, meteorological cumulative features, and meteorological lag features to characterize the nonlinear, persistent, and delayed effects of meteorological factors on short-term load. Then, it generates load meteorological sensitivity classification features based on the response relationship between historical load and meteorological enhancement features. Further, it constructs a hybrid backbone forecasting network (TCN-LSTM-Attention) combining a temporal convolutional network, a long short-term memory network, and an attention mechanism. The temporal convolutional network (TCN) extracts local fluctuations and multi-scale temporal features of short-term load, the long short-term memory network (LSTM) extracts long-term dependencies in the load sequence, and the attention mechanism identifies key historical time steps that significantly contribute to the current forecast result. Finally, it generates the final short-term power load forecast result through multi-source dynamic fusion and output constraint mechanisms. Therefore, this invention can more fully integrate the impact of meteorological big data on load changes while preserving the temporal patterns of historical loads, improving the accuracy and engineering applicability of short-term power load forecasting under complex meteorological conditions. Summary of the Invention

[0005] In view of the above-mentioned existing problems, the present invention provides a method, system, equipment and medium for power system load forecasting based on meteorological big data, which solves the problems that existing load models are difficult to characterize complex meteorological influences, adapt to different loads and have poor forecast accuracy and stability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a power system load forecasting method based on meteorological big data, comprising: Acquire historical load data and meteorological data, and construct meteorological enhancement features based on the meteorological data; Based on the response relationship between the historical load data and the meteorological enhancement features, load classification features are generated; The historical load data, the meteorological enhancement features, and the load classification features are input into the backbone prediction network to extract time-series prediction features and obtain the backbone prediction features. The main forecast features and the meteorological enhancement features are fused to generate an initial forecast value for short-term power load; The initial short-term power load forecast is constrained to obtain the final short-term power load forecast result.

[0007] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, the construction of meteorological enhancement features includes: Meteorological coupling features are constructed based on multiple meteorological variables in meteorological data; Based on the meteorological coupling characteristics within continuous historical periods, meteorological cumulative characteristics are constructed using a nonlinear cumulative method. Based on meteorological coupling characteristics under multiple preset lag times, meteorological lag characteristics are constructed using a multi-scale lag weighting method. Meteorological enhancement features are obtained by combining meteorological coupling features, meteorological accumulation features, and meteorological lag features.

[0008] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, the generated load classification features include: Based on historical load changes and meteorological enhancement characteristics in historical load data, calculate the response intensity of the target power load to meteorological factors; Based on the response intensity, the target power load objects are divided into different meteorological sensitivity categories; The meteorological sensitivity categories are used to generate load classification features through category embedding mapping.

[0009] The beneficial effect of this preferred technical solution is that by distinguishing the sensitivity of load to meteorological conditions, it enhances feature adaptation and improves the prediction generalization performance.

[0010] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, the extraction of time-series forecasting features to obtain the main forecasting features includes: Historical load data, meteorological enhancement features, and load classification features are combined as input features for the backbone prediction network; Input the features from the backbone prediction network into a temporal convolutional network to extract multi-scale local temporal features; Multi-scale local temporal features are input into a long short-term memory network to extract temporal memory features; The temporal memory features are input into the attention mechanism for weighted processing to generate the backbone prediction features.

[0011] The beneficial effect of this preferred technical solution is that by extracting time series information at multiple scales, it can focus on key time periods and enrich the predictive representation.

[0012] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, generating initial short-term power load forecasts includes: The main forecast features and meteorological enhancement features are combined into a fusion input feature; The dynamic fusion weights of each feature channel in the fused input features are calculated using a gated weight network. The input features are weighted and fused based on dynamic fusion weights, and the initial forecast value of short-term power load is generated through the prediction output layer.

[0013] The beneficial effect of this preferred technical solution is that it improves prediction accuracy by adaptively allocating feature weights and fully integrating multi-dimensional information.

[0014] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, the constraint on the initial forecast value of the short-term power load includes: Determine the upper and lower load limits corresponding to the current forecast sample, as well as the maximum load change threshold corresponding to the current forecast sample; The initial short-term power load forecast, the upper and lower load limits corresponding to the current forecast sample, and the maximum load change threshold corresponding to the current forecast sample are input into the output constraint function to obtain the final short-term power load forecast result.

[0015] As a preferred embodiment of the power system load forecasting method based on meteorological big data described in this invention, the final short-term power load forecasting result includes: single-step forecasting result or multi-step forecasting result. In response to performing a single-step forecast, it outputs the load forecast value for the next forecast period. In response to multi-step forecasting, it outputs a load forecast sequence for multiple consecutive forecast periods in the future.

[0016] Secondly, the present invention provides a system for a power system load forecasting method based on meteorological big data, comprising: The data acquisition and feature construction module is used to acquire historical load data and meteorological data, and construct meteorological enhanced features based on the meteorological data. The load classification feature generation module generates load classification features based on the response relationship between the historical load data and the meteorological enhancement features. The feature extraction module is used to input the historical load data, the meteorological enhancement features, and the load classification features into the backbone prediction network, extract the time-series prediction features, and obtain the backbone prediction features. The feature fusion module is used to fuse the main forecast features and the meteorological enhancement features to generate an initial forecast value for short-term power load. The prediction result generation module is used to constrain the initial predicted value of the short-term power load and obtain the final short-term power load prediction result.

[0017] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the power system load forecasting method based on meteorological big data.

[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the power system load forecasting method based on meteorological big data.

[0019] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention not only uses meteorological data as a common input variable, but also further constructs meteorological coupling features, meteorological cumulative features, and meteorological lag features, which can more fully characterize the nonlinear, continuous, and delayed effects of meteorological factors on power load. This invention uses temperature and humidity cross-terms, high-temperature threshold response terms, and low-temperature threshold response terms to characterize the complex coupling effects between temperature, humidity, and other meteorological factors, helping to improve the prediction accuracy of the model under typical weather conditions such as high temperature, high humidity, and low temperature. This invention constructs meteorological cumulative features through a nonlinear cumulative function with attenuation weights, enabling the model to express the cumulative impact of continuous high temperature, continuous low temperature, continuous rainfall, or continuous radiation changes on the load. This invention constructs meteorological lag features through a multi-scale lag weighting mechanism, enabling the model to adaptively identify the intensity of meteorological influences at different lag times, improving the model's ability to express delayed load responses. This invention uses load meteorological sensitivity classification features, enabling the model to distinguish the differences in response to meteorological factors by different load objects or different load states, improving the model's adaptability to different prediction scenarios. This invention employs a TCN-LSTM-Attention backbone prediction network, capable of simultaneously capturing local fluctuations, multi-scale trends, long-term dependencies, and key historical time step information in load sequences. Through a multi-source dynamic fusion mechanism, it adaptively determines the contribution weights of backbone prediction features, meteorological coupling features, meteorological cumulative features, meteorological lag features, and load classification features, improving the model's generalization ability under complex meteorological conditions. Furthermore, this invention uses an output constraint mechanism to reasonably constrain the initial prediction results, reducing the risk of load exceeding limits, abnormal changes, or deviations from historical statistical patterns, thus improving the engineering usability of the prediction results. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A schematic diagram of the overall process logic of a power system load forecasting method based on meteorological big data, provided as an embodiment of the present invention; Figure 2This is a schematic diagram illustrating the generation of load meteorological sensitivity classification features in a power system load forecasting method based on meteorological big data, provided as an embodiment of the present invention. Figure 3 This is a schematic diagram of the backbone prediction network feature extraction process for a power system load prediction method based on meteorological big data, provided as an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figures 1-3 As an embodiment of the present invention, a power system load forecasting method based on meteorological big data is provided, comprising: S100: Acquire historical load data and meteorological data, and construct meteorological enhancement features based on the meteorological data; In this embodiment of the invention, acquiring historical load data and meteorological data includes acquiring historical load data of the target power load object within a continuous historical period according to a preset sampling interval. Historical load data includes one or more of the following: active power sequence, electricity consumption sequence, maximum load, minimum load, or a historical load curve formed from the above data.

[0024] Acquire meteorological data within the time and spatial range corresponding to historical load data. Meteorological data includes multiple parameters such as temperature, humidity, wind speed, rainfall, solar radiation intensity, irradiance, air pressure, and weather type.

[0025] Historical load data and meteorological data were timestamped according to a unified time scale. Missing data were repaired using linear interpolation, nearest-neighbor imputation, or statistical imputation. Outlier data were processed using threshold judgment, sliding window smoothing, or statistical rule removal. After data cleaning, the historical load data and meteorological data were normalized to obtain a unified time series sample.

[0026] For any sample to be predicted, the historical load sequence is denoted as: The meteorological characteristic sequence is denoted as: in, For the historical load sequence within the input window, The input window contains a sequence of meteorological features. Enter the window length.

[0027] S200: Generates load classification features based on the response relationship between historical load data and meteorological enhancement features; S300: Input historical load data, meteorological enhancement features, and load classification features into the backbone prediction network, extract time-series prediction features, and obtain backbone prediction features; In one optional embodiment, the backbone prediction network can be a Transformer-Long Short-Term Memory hybrid backbone network (Transformer-LSTM). The Transformer encoder performs multi-head self-attention on the load and meteorological sequences to capture global long-distance temporal correlations; the output is fed into the LSTM to extract local temporal fluctuation features; and the two types of features are fused to obtain the backbone prediction features. In another optional embodiment, the backbone prediction network can also be a convolutional bidirectional gated recurrent (CNN-BiGRU) attention network, a one-dimensional convolutional neural network (CNN) to extract multi-scale short-term load fluctuation features; a bidirectional gated recurrent unit (BiGRU) to traverse the sequence bidirectionally and capture temporal dependencies; and a single-layer attention layer to filter key temporal nodes and output backbone features. In this embodiment of the invention, the backbone prediction network includes a TCN-LSTM-Attention backbone prediction network.

[0028] The TCN-LSTM-Attention backbone prediction network consists of a TCN temporal convolutional network, an LSTM long short-term memory network, and an attention mechanism. It is used to extract multi-scale local temporal features, long-term dependency features, and key time step features required for short-term load prediction.

[0029] S400: It integrates the main forecast features and meteorological enhancement features to generate the initial forecast value of short-term power load; S500: Constrains the initial short-term power load forecast to obtain the final short-term power load forecast result.

[0030] It should be noted that this invention utilizes meteorological enhancement features to fully exploit the coupling of temperature and humidity, the accumulation of extreme weather, and the nonlinear effects of meteorological lag, generating load classification features to distinguish the differences in meteorological sensitivity among industrial, commercial, and residential loads. A TCN-LSTM-Attention network synchronously captures local load fluctuations and long-term temporal patterns, adaptively weighting and fusing various features, and correcting anomaly predictions through boundary constraints. Compared to traditional models, prediction errors are significantly reduced under complex weather conditions, adaptable to multiple types of power supply areas, avoids prediction jumps and out-of-bounds, and improves the reliability and engineering practicality of power grid dispatch data.

[0031] In this embodiment of the invention, step S100 includes the following sub-steps A1-A4; In A1: Meteorological coupling features are constructed based on multiple meteorological variables in the meteorological data; In A2: Based on the meteorological coupling characteristics within continuous historical time periods, meteorological cumulative characteristics are constructed using a nonlinear cumulative method; In A3: Meteorological lag features are constructed based on meteorological coupling characteristics under multiple preset lag times using a multi-scale lag weighting method; In A4: Meteorological coupling features, meteorological cumulative features, and meteorological lag features are combined to obtain meteorological enhancement features.

[0032] In this embodiment of the invention, a meteorological coupling feature reflecting the comprehensive influence of meteorological factors is constructed based on one or more meteorological variables among temperature, humidity, wind speed, rainfall, irradiance, and air pressure.

[0033] The meteorological coupling characteristics are represented as follows: in, Indicates meteorological coupling characteristics, Indicates temperature. Indicates humidity. Indicates the temperature and humidity cross term. Indicates rainfall, solar radiation intensity, or irradiance. Indicates wind speed. Indicates air pressure. Indicates the cooling-sensitive temperature threshold. Indicates the heating-sensitive temperature threshold. to For model parameters, Let f(x) represent a positive function that satisfies the following conditions: , Used to characterize the nonlinear enhancement effect on the load after the temperature exceeds the cooling sensitivity threshold. It is used to characterize the nonlinear enhancement effect on the load when the temperature is below the heating sensitivity threshold.

[0034] In one optional embodiment, the nonlinear accumulation method can be a piecewise threshold power accumulation method, which divides the meteorological intensity into three threshold intervals: high, medium, and low; different intervals are assigned differentiated power nonlinear transformations; and the accumulated features are obtained by weighted power operations on the meteorological coupling features within the window and then summed. In another optional embodiment, the nonlinear accumulation method can also be the time-series quantile nonlinear accumulation method, which extracts the 5%, 50%, and 95% quantiles of meteorological coupling features within the accumulation window; the quantiles are respectively subjected to sigmoid nonlinear mapping; and the meteorological accumulation features are generated by weighted fusion according to extreme weather weights. In this embodiment of the invention, the nonlinear accumulation method includes a nonlinear accumulation method with attenuation weights; Specifically, based on the meteorological coupling characteristics over a continuous historical period, a nonlinear cumulative method with attenuation weights is used to construct meteorological cumulative characteristics to reflect the cumulative impact of continuous high temperature, continuous low temperature, continuous high humidity, continuous rainfall, or continuous radiation changes on short-term loads.

[0035] The cumulative meteorological characteristics are expressed as follows: in, Indicates the cumulative characteristics of meteorological events. Indicates the cumulative weather window length. Indicates the number of cumulative windows. Meteorological coupling characteristics at a historical moment, Indicates the first The decay weight corresponding to each historical moment This represents a nonlinear mapping function.

[0036] Decay weight satisfy: Nonlinear mapping function satisfy: in, Indicates the attenuation coefficient. and These are the model parameters.

[0037] In one optional embodiment, the multi-scale lag weighting method can be a hierarchical lag convolution weighting method, which sets multiple groups of fixed lag step size groups and divides the lag scale into short, medium and long three layers; each layer uses independent one-dimensional convolution to extract local patterns of the corresponding lag meteorological coupling features; after the convolution output is activated by nonlinear transformation, the meteorological lag features are obtained by hierarchical weighting and summation. In another alternative embodiment, the multi-scale lag weighting method can also be a dynamic lag clustering weighting method, which adaptively selects the effective lag duration based on the correlation between historical load and meteorological conditions; clusters similar lag steps into multi-scale lag clusters; assigns weights according to the correlation coefficient of each cluster, and generates lag features through weighted fusion. In this embodiment of the invention, the multi-scale lag weighting method includes the load-association adaptive multi-scale lag weighting method; Specifically, based on the meteorological coupling characteristics under multiple preset lag times, a multi-scale lag weighting method is used to construct meteorological lag characteristics to reflect the delayed response of the load to meteorological changes.

[0038] The weather lag characteristic is expressed as follows: in, Indicates the lag characteristic of meteorology. Indicates the preset lag step size. Indicates the first Meteorological coupling characteristics corresponding to each lag time Indicates the first The weight of the lag effect corresponding to each lag time.

[0039] Lagging effect weight satisfy: in, Indicates the first Historical load characteristics corresponding to each lag time , , and These are the model parameters.

[0040] The meteorological enhancement characteristics are represented as follows: in, It is a meteorological coupling characteristic. This is a meteorological accumulation characteristic. This is a characteristic of weather lag.

[0041] It should be noted that by constructing three types of meteorological characteristics, the coupling of temperature and humidity, continuous meteorological accumulation, and meteorological lag effects are accurately characterized, and the nonlinear influence of meteorology is fully explored, thus significantly improving the accuracy of load forecasting.

[0042] In this embodiment of the invention, step S200 includes the following sub-steps B1-B3; In B1: Based on the historical load changes and meteorological enhancement characteristics in the historical load data, calculate the response intensity of the target power load to meteorological factors; In B2: Based on the response intensity, the target power load objects are divided into different meteorological sensitivity categories; In B3: Meteorological sensitivity categories are generated as load classification features through category embedding mapping.

[0043] In embodiments of the present invention, such as Figure 2 As shown, based on historical load changes, meteorological coupling characteristics, meteorological cumulative characteristics, and meteorological lag characteristics, the response intensity of the target power load to meteorological factors is calculated and expressed as follows: in, Indicates the first Response scores corresponding to meteorological sensitivity This indicates the historical load variation. , , These represent the changes in meteorological coupling characteristics, meteorological cumulative characteristics, and meteorological lag characteristics, respectively. This represents the activation function. and These are model parameters; Based on the response intensity, the target power load objects are divided into different meteorological sensitivity categories, which can be divided into high meteorological sensitivity loads, medium meteorological sensitivity loads and low meteorological sensitivity loads; When the correlation between historical load changes and meteorological enhancement characteristic changes is high, the corresponding load status can be classified as highly meteorologically sensitive; when there is a certain relationship between load changes and meteorological enhancement characteristic changes but it is not significant, it can be classified as moderately meteorologically sensitive; when load changes are mainly affected by historical electricity consumption patterns and have a weak relationship with meteorological changes, it can be classified as low meteorologically sensitive.

[0044] Meteorological sensitivity categories are converted into load classification features that can be input into the forecasting network for subsequent backbone forecasting and multi-source dynamic fusion.

[0045] The generated classification weights are represented as follows: The characteristics of deep integrated load classification are represented as follows: in, Indicates the first Classification weights for meteorological sensitivity categories, Indicates the number of meteorological sensitivity categories. Indicates the first Embedding vectors corresponding to meteorological sensitivity categories, This indicates the characteristics of the comprehensive load classification.

[0046] It should be noted that by differentiating the load's sensitivity to weather conditions, and adapting to different electricity consumption scenarios, the forecast generalization capability is improved.

[0047] In this embodiment of the invention, step S300 includes the following sub-steps C1-C4; In C1: Historical load data, meteorological enhancement features, and load classification features are combined as input features for the backbone prediction network; In C2: The input features of the backbone prediction network are input into the temporal convolutional network to extract multi-scale local temporal features; In C3: Multi-scale local temporal features are input into the long short-term memory network to extract temporal memory features; In C4: Temporal memory features are input into the attention mechanism for weighted processing to generate backbone prediction features.

[0048] Specifically, such as Figure 3 As shown, historical load sequences, meteorological coupling characteristics, meteorological cumulative characteristics, meteorological lag characteristics, and load classification characteristics are combined to form the input characteristics of the backbone forecasting network, which are expressed as follows: in, This represents the input features of the backbone prediction network. Represents the historical load sequence. Indicates meteorological coupling characteristics, Indicates the cumulative characteristics of meteorological events. Indicates the lag characteristic of meteorology. This indicates the characteristics of load classification.

[0049] We utilize causal convolution and dilated convolution structures to perform temporal convolution processing on the input features of the backbone prediction network in order to capture short-term load fluctuations, local spikes, and multi-scale change trends.

[0050] Temporal convolutional networks (TCNs) are represented by the following causal dilated convolution form: in, Indicates the first Layer TCN output features, This represents the input features of the previous layer. Indicates the kernel length. Indicates the first Layer expansion coefficient, Represents the convolution kernel parameters. Indicates the bias term. This represents a non-linear activation function.

[0051] In one optional embodiment, the attention mechanism can be a gated self-attention mechanism, which generates a gated weight matrix on the temporal features output by the LSTM; filters irrelevant temporal nodes through gating, retaining key meteorological periods; and weights the temporal features with the gated weights to output the backbone features. In another optional embodiment, the attention mechanism can also be a multi-head sparse attention mechanism, which divides the temporal correlation weights of multi-head parallel computation, retains only highly correlated lagging nodes; multi-head features are spliced ​​and fused to compress redundant temporal information; and the final backbone features are obtained by weighting after normalization. In this embodiment of the invention, the attention mechanism includes the Attention mechanism; Specifically, the multi-scale local temporal features output by TCN are input into a Long Short-Term Memory (LSTM) network to obtain temporal memory features. Then, the Attention mechanism is used to weight different historical time steps to generate backbone prediction features.

[0052] The temporal memory features extracted by the LSTM (Long Short-Term Memory) network are represented as follows: in, Indicates the output characteristics of TCN. This indicates the hidden state in the previous moment. Represents the Gate of Oblivion Indicates the input gate. Indicates candidate memory states, Indicates the current memory state. Indicates the previous memory state. Indicates the output gate. Indicates the output features of LSTM. This indicates element-wise multiplication.

[0053] The attention mechanism generates the following backbone prediction features: in, Indicates the first Attention weights corresponding to each historical time step Indicates the first The output features of the LSTM at each historical time step Indicates weather enhancement characteristics, Indicates the main prediction features, , , and These are the model parameters.

[0054] It should be noted that by capturing multi-scale load fluctuations through TCN, retaining long-term temporal memory through LSTM, and combining attention with meteorological characteristics to weight key periods, multi-dimensional features are comprehensively extracted to improve the predictive representation capability.

[0055] In this embodiment of the invention, step S400 includes the following sub-steps D1-D3; In D1: The backbone prediction features and meteorological enhancement features are combined into fused input features; In D2: The dynamic fusion weights of each feature channel in the fused input features are calculated through a gated weight network; In D3: The fusion input features are weighted and fused based on dynamic fusion weights, and the initial forecast value of short-term power load is generated through the prediction output layer.

[0056] In this embodiment of the invention, the main forecast feature, meteorological coupling feature, meteorological cumulative feature, meteorological lag feature, and load classification feature are combined into a fusion input feature, represented as follows: By using a gated weighted network to calculate the contribution weights of various feature channels to the current short-term load forecasting task, the feature contributions under different meteorological conditions and different load sensitivity conditions can be adaptively adjusted.

[0057] The dynamic fusion weights are represented as follows: in, Indicates the weight of the main prediction features. Represents the weights of meteorological coupling features. Indicates the weight of cumulative meteorological characteristics. Indicates the weight of meteorological lag characteristics. This represents the weight of the load classification features.

[0058] Based on dynamic fusion weights, various features are weighted and fused, and the initial forecast value of short-term power load is generated through the prediction output layer.

[0059] The weighted fusion feature is represented as: in, Indicates the weighted fusion characteristics, , , and This represents the projection parameters used to unify the feature dimensions.

[0060] The initial forecast value for short-term power load is expressed as follows: in, This represents the initial forecast value of short-term electricity load. , , and These are the model parameters.

[0061] It should be noted that by using a gating network to adaptively allocate the weights of various features, adapting to different weather and load scenarios, and then unifying the dimensions and fusing them together, the accuracy of initial load forecasting can be improved by making full use of multi-dimensional information.

[0062] In this embodiment of the invention, step S500 includes the following sub-steps E1-E2. In E1: Determine the upper and lower load limits corresponding to the current forecast sample, as well as the maximum load change threshold corresponding to the current forecast sample; In E2: The initial short-term power load forecast, the upper and lower load limits corresponding to the current forecast sample, and the maximum load change threshold corresponding to the current forecast sample are input into the output constraint function to obtain the final short-term power load forecast result.

[0063] In this embodiment of the invention, the upper load limit, lower load limit, and maximum load change threshold are defined as follows: in, This represents the scenario conditions comprised of the forecast period, date type, season type, and weather conditions. This indicates the 95th percentile. This indicates the conditional 5th percentile. Indicates the conditional standard deviation. This represents the historical load variation. This represents the change in meteorological coupling characteristics. , and To adjust the parameters.

[0064] In the training process of the power system load forecasting model, a composite loss function is used to optimize the model parameters. The composite loss function includes a mean squared error term, a mean absolute error term, a relative error term, and a classification weight regularization term.

[0065] The composite loss function is expressed as: in, This represents the model training loss function. This represents the actual load value. This indicates the final short-term electricity load forecast result. Indicates the number of training samples. Indicates the first Classification weights for meteorological sensitivity categories, Indicates the first Historical average weight of meteorological sensitivity categories This represents a constant used to prevent the denominator from being zero. , , and This represents the loss weight.

[0066] The initial short-term power load forecast, upper load limit, lower load limit, and maximum load change threshold are input into the output constraint function to obtain the final short-term power load forecast result.

[0067] The final short-term power load forecast results are expressed as follows: in, This indicates the final short-term electricity load forecast result. This represents the initial forecast value of short-term electricity load. Indicates the lower limit of load. Indicates the maximum load. This represents the reference load value for the previous forecast period or an adjacent historical period. This indicates the threshold for the maximum load change. , and This represents the constraint penalty coefficient.

[0068] In this embodiment of the invention, after completing steps E1-E2, step S500 also includes steps E3-E4. In E3: In response to performing a single-step forecast, the load forecast value for the next forecast period is output; In E4: In response to multi-step forecasting, it outputs a load forecast sequence for multiple consecutive forecast periods in the future.

[0069] In this embodiment of the invention, after the model training is completed, the model can output single-step short-term load prediction results or multi-step short-term load prediction results according to the actual prediction task.

[0070] When performing a single-step forecast, the model outputs a load forecast value for the next forecast period based on historical load data and meteorological data within the current input window.

[0071] When performing multi-step forecasting, the model outputs a load forecast sequence for multiple consecutive forecast periods in the future, represented as: in, This represents a multi-step short-term load forecast sequence. This indicates the number of prediction steps.

[0072] It should be noted that by constraining the predicted value through scenario statistical thresholds and using a composite loss optimization model, single or multi-step prediction can be supported, avoiding out-of-bounds jumps in results and meeting the needs of power grid engineering.

[0073] Example 2: The above is an illustrative scheme of a power system load forecasting method based on meteorological big data according to this embodiment. It should be noted that the technical solution of this power system load forecasting method based on meteorological big data belongs to the same concept as the technical solution of the power system load forecasting method based on meteorological big data described above. Details not described in detail in the technical solution of the power system load forecasting method based on meteorological big data in this embodiment can be found in the description of the technical solution of the power system load forecasting method based on meteorological big data described above.

[0074] This embodiment describes a power system load forecasting method based on meteorological big data, comprising: The data acquisition and feature construction module is used to acquire historical load data and meteorological data, and construct meteorological enhanced features based on the meteorological data. The load classification feature generation module generates load classification features based on the response relationship between the historical load data and the meteorological enhancement features. The feature extraction module is used to input the historical load data, the meteorological enhancement features, and the load classification features into the backbone prediction network, extract the time-series prediction features, and obtain the backbone prediction features. The feature fusion module is used to fuse the main forecast features and the meteorological enhancement features to generate an initial forecast value for short-term power load. The prediction result generation module is used to constrain the initial predicted value of the short-term power load and obtain the final short-term power load prediction result.

[0075] This embodiment also provides a computer device applicable to a power system load forecasting method based on meteorological big data, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a power system load forecasting method based on meteorological big data, as proposed in the above embodiments.

[0076] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power system load forecasting method based on meteorological big data as proposed in the above embodiment.

[0077] The storage medium proposed in this embodiment and the method for power system load forecasting based on meteorological big data proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0078] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computing device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A power system load forecasting method based on meteorological big data, characterized in that, include: Acquire historical load data and meteorological data, and construct meteorological enhancement features based on the meteorological data; Based on the response relationship between the historical load data and the meteorological enhancement features, load classification features are generated; The historical load data, the meteorological enhancement features, and the load classification features are input into the backbone prediction network to extract time-series prediction features and obtain the backbone prediction features. The main forecast features and the meteorological enhancement features are fused to generate an initial forecast value for short-term power load; The initial short-term power load forecast is constrained to obtain the final short-term power load forecast result.

2. The power system load forecasting method based on meteorological big data as described in claim 1, characterized in that, The construction of meteorological enhancement features includes: Meteorological coupling features are constructed based on multiple meteorological variables in meteorological data; Based on the meteorological coupling characteristics within continuous historical periods, meteorological cumulative characteristics are constructed using a nonlinear cumulative method. Based on meteorological coupling characteristics under multiple preset lag times, meteorological lag characteristics are constructed using a multi-scale lag weighting method. Meteorological enhancement features are obtained by combining meteorological coupling features, meteorological accumulation features, and meteorological lag features.

3. The power system load forecasting method based on meteorological big data as described in claim 2, characterized in that, The generated load classification features include: Based on historical load changes and meteorological enhancement characteristics in historical load data, calculate the response intensity of the target power load to meteorological factors; Based on the response intensity, the target power load objects are divided into different meteorological sensitivity categories; The meteorological sensitivity categories are used to generate load classification features through category embedding mapping.

4. The power system load forecasting method based on meteorological big data as described in claim 3, characterized in that, Extracting time-series prediction features yields backbone prediction features including: Historical load data, meteorological enhancement features, and load classification features are combined as input features for the backbone prediction network; Input the features from the backbone prediction network into a temporal convolutional network to extract multi-scale local temporal features; Multi-scale local temporal features are input into a long short-term memory network to extract temporal memory features; The temporal memory features are input into the attention mechanism for weighted processing to generate the backbone prediction features.

5. The power system load forecasting method based on meteorological big data as described in claim 4, characterized in that, The generation of initial short-term power load forecasts includes: The main forecast features and meteorological enhancement features are combined into a fusion input feature; The dynamic fusion weights of each feature channel in the fused input features are calculated using a gated weight network; The input features are weighted and fused based on dynamic fusion weights, and the initial forecast value of short-term power load is generated through the prediction output layer.

6. The power system load forecasting method based on meteorological big data as described in claim 5, characterized in that, Constraining the initial forecast of the short-term power load includes: Determine the upper and lower load limits corresponding to the current forecast sample, as well as the maximum load change threshold corresponding to the current forecast sample; The initial short-term power load forecast, the upper and lower load limits corresponding to the current forecast sample, and the maximum load change threshold corresponding to the current forecast sample are input into the output constraint function to obtain the final short-term power load forecast result.

7. A power system load forecasting method based on meteorological big data as described in claim 1 or 6, characterized in that, The final short-term power load forecast results can be obtained as either single-step or multi-step forecast results. In response to performing a single-step forecast, it outputs the load forecast value for the next forecast period. In response to multi-step forecasting, it outputs a load forecast sequence for multiple consecutive forecast periods in the future.

8. A system for power system load forecasting based on meteorological big data, comprising applying the power system load forecasting method based on meteorological big data as described in any one of claims 1-7, characterized in that, include: The data acquisition and feature construction module is used to acquire historical load data and meteorological data, and construct meteorological enhanced features based on the meteorological data. load A classification feature generation module is used to generate load classification features based on the response relationship between the historical load data and the meteorological enhancement features. The feature extraction module is used to input the historical load data, the meteorological enhancement features, and the load classification features into the backbone prediction network, extract the time-series prediction features, and obtain the backbone prediction features. The feature fusion module is used to fuse the main forecast features and the meteorological enhancement features to generate an initial forecast value for short-term power load. The prediction result generation module is used to constrain the initial predicted value of the short-term power load and obtain the final short-term power load prediction result.

9. A computer device, characterized in that, include: A memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the power system load forecasting method based on meteorological big data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of a power system load forecasting method based on meteorological big data as described in any one of claims 1 to 7.

Images