Power demand and weather forecast error collaborative prediction method and device and related equipment

Abstract

The invention discloses a power demand and weather forecast error collaborative prediction method and device and related equipment. The method comprises the following steps: constructing a multi-source collaborative data set containing a weather forecast error propagation sequence; a multi-task learning model is constructed and trained, the model comprises a shared feature coding layer, a load prediction main task branch and a meteorological error prediction auxiliary task branch, and the auxiliary task branch is used for predicting a meteorological prediction error propagation sequence; and outputting a prediction result of the power load and corresponding uncertainty quantitative information according to real-time data by using the trained multi-task learning model. According to the invention, the accuracy of collaborative prediction of the electric power demand and the weather forecast error is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power engineering, and in particular to a power demand and meteorological forecast error collaborative prediction method and device and related equipment. BACKGROUND

[0002] Under the background of new power system construction and energy transformation, the coupling relationship between power load and meteorological conditions is increasingly close and dynamic. Meteorological elements (such as temperature, humidity, wind speed, and illumination) not only directly affect the temperature control load of residents and businesses, but also indirectly affect the supply and demand balance of the power grid by affecting the output characteristics of renewable energy (such as wind power and photovoltaic power). Therefore, numerical weather prediction has become a key input for modern power systems to conduct short-term and ultra-short-term load prediction. Accurate prediction of the influence of meteorological condition changes on power demand is of great significance to ensuring power grid safety, optimizing dispatching decisions, and improving economic operation level.

[0003] However, existing power load prediction methods still have significant limitations in deep integration and utilization of meteorological information. First, the mainstream methods usually treat meteorological factors such as temperature as static or independent parameters in the feature engineering stage, failing to fully model the dynamic nonlinear relationship between them and load demand that changes with season, time period, and region. This simplification leads to a decline in prediction performance during key meteorological transitions such as sudden temperature changes, cold waves, and heat waves. Second, the more fundamental defect is that existing prediction models are generally built on a single task learning framework, i.e., only the final load prediction value is used as the optimization target. This framework completely ignores the inherent uncertainty and error of meteorological forecasts. In actual applications, numerical weather prediction inevitably has deviations, and existing models lack explicit modeling capabilities for error propagation mechanisms. When forecast errors occur, they are directly transmitted and even amplified in the prediction model as incorrect inputs, severely weakening the reliability and robustness of load prediction results.

[0004] In summary, existing technologies lack effective modeling and utilization mechanisms for meteorological forecast errors, making it difficult to guarantee the accuracy and reliability of power load prediction under the realistic condition of meteorological forecast uncertainty. How to collaboratively model power demand and meteorological forecast errors to achieve robust improvement in prediction performance has become a technical problem that needs to be solved in this field. SUMMARY

[0005] Embodiments of the present application provide a power demand and meteorological forecast error collaborative prediction method, device and related equipment to improve the accuracy of power demand and meteorological forecast error collaborative prediction.

[0006] To solve the above technical problems, the present application provides a power demand and meteorological forecast error collaborative prediction method, comprising: Construct a multi-source collaborative dataset containing the propagation sequence of weather forecast errors; A multi-task learning model is constructed and trained. The model includes a shared feature encoding layer, a main task branch for load prediction, and an auxiliary task branch for meteorological error prediction. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors. Using the trained multi-task learning model, the predicted power load and corresponding uncertainty quantification information are output based on real-time data.

[0007] Optionally, the construction of a multi-source collaborative dataset containing weather forecast error propagation sequences includes: Acquire historical power load data, numerical weather forecast data, and meteorological observation data for the corresponding time periods; Based on the meteorological forecast data and the actual meteorological observation data, a dynamic meteorological forecast error propagation sequence is constructed.

[0008] Optionally, the step of constructing a dynamic weather forecast error propagation sequence based on the weather forecast data and the actual weather observation data includes: The difference between the forecast and actual values ​​of each meteorological station at different time steps is calculated to form the original error sequence; A graph structure representing the spatial relationships between meteorological stations is constructed, and a spatial propagation operator of error is learned from the original error sequence using a graph convolutional network; Based on the physical mechanism equation of error propagation, the spatial propagation operator is used to discretize the physical mechanism equation to generate the dynamic weather forecast error propagation sequence.

[0009] Optionally, the method further includes: Based on the dynamic weather forecast error propagation sequence, a weather-load demand coupling feature is constructed; The meteorological-load demand coupling features are input into the multi-task learning model.

[0010] Optionally, the construction and training of the multi-task learning model includes training the model using a phased strategy; In the first stage, the shared feature encoding layer and the main load prediction task branch are trained using the load prediction task; In the second stage, the main task branch for load forecasting and the auxiliary task branch for meteorological error forecasting are trained collaboratively.

[0011] Optionally, during the second phase of training, the weights of the load forecasting main task branch and the meteorological error forecasting auxiliary task branch in the joint loss function are dynamically adjusted based on the meteorological uncertainty index assessed in the historical training cycle.

[0012] Optionally, the step of outputting the predicted power load and the corresponding uncertainty quantification information based on real-time data includes: Based on the error results predicted by the auxiliary task branch, calculate the meteorological uncertainty index; Based on the meteorological uncertainty index and the preset mapping relationship, the disturbance magnitude of the load forecast is determined; Based on the load forecast value output by the main load forecasting task branch and the disturbance amplitude, a forecast confidence interval is generated.

[0013] To address the aforementioned technical problems, this application also provides a device for collaborative prediction of electricity demand and weather forecast errors, comprising: The data construction module is used to build a multi-source collaborative dataset containing the propagation sequence of weather forecast errors; The model processing module is used to build and train a multi-task learning model. The model includes a shared feature encoding layer, a main task branch for load forecasting, and an auxiliary task branch for meteorological error forecasting. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors. The prediction output module is used to output the prediction results of power load and the corresponding uncertainty quantification information based on real-time data using the trained model.

[0014] To address the aforementioned technical problems, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned method for co-predicting electricity demand and weather forecast errors.

[0015] To address the aforementioned technical problems, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned method for co-predicting electricity demand and weather forecast errors.

[0016] The method, apparatus, computer equipment, and storage medium for collaborative prediction of power demand and weather forecast errors provided in this invention construct and utilize a dynamic weather forecast error propagation sequence to enable a task learning model to collaboratively model the evolution of load demand and forecast errors. This significantly improves the accuracy and robustness of load forecasting when there are deviations in the input weather forecast, and outputs prediction results with quantified uncertainty, providing a more reliable decision-making basis for power dispatch. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. 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.

[0018] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the method for co-predicting electricity demand and weather forecast errors according to this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the power demand and weather forecast error co-prediction device according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 ,like Figure 1As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0023] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0024] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0025] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0026] It should be noted that the power demand and weather forecast error co-prediction method provided in this application embodiment is executed by the server, and correspondingly, the power demand and weather forecast error co-prediction device is set in the server.

[0027] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included. The terminal devices 101, 102, and 103 in this embodiment can specifically correspond to application systems in actual production.

[0028] Please see Figure 2 , Figure 2 This invention illustrates a method for collaborative prediction of electricity demand and weather forecast errors, provided by an embodiment of the present invention, which is applied to... Figure 1 Taking the server-side as an example, the details are as follows: S201: Construct a multi-source collaborative dataset containing the propagation sequence of weather forecast errors.

[0029] Specifically, historical power load data is first collected from the power grid dispatching system, with a sampling frequency of no less than 15 minutes. At the same time, numerical weather forecast data for the same period and real-time observation data from corresponding meteorological stations are obtained from the meteorological department. The meteorological elements involved include at least temperature, humidity, wind speed, air pressure, and solar radiation.

[0030] To ensure data quality, a systematic quality control process was implemented on the raw data: the spatiotemporal KNN method was used to impute missing values, outliers were detected and corrected based on the mean and standard deviation within a sliding window, and all variables were standardized using Z-scores to eliminate the influence of dimensions. Building upon data alignment and cleaning, the core task was to construct a dynamic weather forecast error propagation sequence. This process first calculated the difference between the forecast and actual observation values ​​for each meteorological station at each historical time step, forming a cross-site raw error vector sequence.

[0031] Subsequently, the meteorological station network is abstracted into a graph structure, where nodes represent stations and edge weights integrate information such as geographical distance between stations, regional prevailing wind direction, and correlation coefficients with historical errors. A graph convolutional network (GCN) is then trained on this graph using historical error sequences, thereby learning the propagation pattern of errors in the spatial neighborhood through data-driven learning, which is equivalent to obtaining a discretized spatial propagation operator.

[0032] Furthermore, the operator is embedded into the partial differential equations describing the physical mechanism of diffusion or advection processes, and the explicit Euler method is used for time discretization to derive the recursive relationship of the error from the current moment to the next moment. Iterating through this recursive relationship generates a complete dynamic sequence that simultaneously encodes the spatiotemporal evolution of the error. In addition, this embodiment constructs a series of meteorological-load coupling features based on this dynamic sequence. For example, it combines temperature observations with their corresponding error propagation components to generate an "effective temperature," and then calculates the "propagation-enhanced degree-day coefficient"; and it combines the sensitivity of local load to temperature with the error sequence to construct "disturbance-load coupling features," etc. Simultaneously, multi-period sine-cosine encoding is performed on the time information, and holiday and workday type identifiers are introduced. Finally, all processed features are organized chronologically and divided into training, validation, and test sets.

[0033] This embodiment constructs a dynamic weather forecast error propagation sequence, enabling the prediction model to anticipate the spatiotemporal evolution trend of errors in advance, thereby significantly improving the foresight and adaptability of load forecasting under the influence of moving weather systems. Based on the coupling characteristics derived from this sequence, the dynamic correlation between weather disturbances and electricity demand can be more precisely characterized, effectively improving the model's accuracy in capturing key inflection points of load changes. Simultaneously, multi-source data alignment and quality control throughout the entire process effectively enhance the stability and reliability of the entire process from data input to model output.

[0034] In one specific optional implementation, step S201, namely, constructing a multi-source collaborative dataset containing weather forecast error propagation sequences, includes: Acquire historical power load data, numerical weather forecast data, and meteorological observation data for the corresponding time periods; Based on meteorological forecast data and actual meteorological observation data, a dynamic meteorological forecast error propagation sequence is constructed.

[0035] Step S201 aims to construct a collaborative dataset that deeply integrates multi-source information, which serves as the basis for subsequent model training and prediction.

[0036] First, three types of historical data are collected and integrated: power load data, meteorological observation data, and numerical weather prediction data. Power load data is typically obtained from the power grid dispatch system, such as load records of a regional power grid over the past year, sampled at a frequency of 15 minutes. Numerical weather prediction data comes from meteorological departments and covers forecast values ​​for elements such as temperature, humidity, wind speed, air pressure, and solar radiation for specific future periods. Correspondingly, meteorological observation data requires obtaining actual observation values ​​within the same spatiotemporal range from meteorological observation stations to ensure strict temporal and spatial alignment between forecasts and actual conditions. To ensure data usability, rigorous quality control is necessary. Specifically, this includes imputing missing values ​​using the spatiotemporal KNN method, correcting outliers exceeding a preset threshold (e.g., ±3 standard deviations) using neighborhood averaging, and standardizing all variables using Z-scores to eliminate the influence of dimensions, providing a stable and reliable data foundation for subsequent modeling.

[0037] Secondly, based on the aligned meteorological forecast and actual observation data, a dynamic sequence reflecting the spatiotemporal evolution of errors is constructed. Specifically, for each meteorological station at each time step, the difference between the forecast and actual values ​​is calculated, forming the original station error time series. To characterize the spatial correlation and propagation of errors, meteorological stations are abstracted as nodes in a graph structure. Based on factors such as geographical distance between stations, prevailing wind direction in the region, and correlation coefficients of historical error sequences, the weights of edges between nodes are comprehensively calculated to construct a weighted adjacency matrix. A graph convolutional network is used to learn this graph structure and its attached historical error data, adaptively extracting the propagation pattern of errors in the spatial neighborhood. This is equivalent to learning a data-driven spatial propagation operator. Furthermore, by combining the mechanistic equations describing physical processes such as matter diffusion (whose spatial differential terms are discretized and approximated by the learned propagation operator) and using explicit Euler method for time discretization, the recursive relationship of errors from the current time to the next time step can be derived. Through iterative calculations based on this recursive relationship, a complete dynamic meteorological forecast error propagation sequence that simultaneously contains information on spatial diffusion and temporal evolution is ultimately generated.

[0038] This embodiment introduces a dynamic error propagation sequence, enabling the model to obtain forward-looking information about how forecast biases evolve over time and space, rather than just knowing static instantaneous errors. This significantly enhances the adaptability and early warning capability of the prediction model when facing mobile weather systems such as cold waves and typhoons. At the same time, by collaboratively constructing and aligning load, weather forecast, actual conditions, and derived error sequences, it provides unified features for subsequent multi-task learning models, allowing the model to simultaneously learn the patterns of load changes and error evolution. This supports the collaborative prediction of both from the data source, effectively reducing the risk of model training instability caused by data noise, missing data, or differences in magnitude, and is conducive to improving the robustness and reliability of predictions.

[0039] In one specific optional implementation, a dynamic weather forecast error propagation sequence is constructed based on weather forecast data and actual weather observation data, including: The difference between the forecast and actual values ​​of each meteorological station at different time steps is calculated to form the original error sequence; A graph structure representing the spatial relationships between meteorological stations is constructed, and a spatial propagation operator of error is learned from the original error sequence using a graph convolutional network; Based on the physical mechanism equations of error propagation, the spatial propagation operator is used to discretize the physical mechanism equations and generate dynamic weather forecast error propagation sequences.

[0040] First, for each meteorological station, such as station A located in the eastern part of the city, the difference between the predicted and actual temperature values ​​at a specific time each day (e.g., 14:00) is calculated. If the predicted temperature for a certain day is 25°C and the actual temperature is 26.5°C, then the error at that moment is -1.5°C. This calculation is extended to all stations and continuous time steps, thus forming a raw error sequence that varies over time and covers the entire region. This step transforms the abstract uncertainty of weather forecasts into a concrete and quantifiable data basis.

[0041] Secondly, to characterize the spatial correlation and propagation of errors, a spatial relationship graph with meteorological stations as nodes needs to be constructed. For example, station A and station B are 50 kilometers apart, the prevailing wind direction in the region is easterly, and historical data shows that their temperature errors are often positively correlated. Based on this information, the weights of the edges connecting the two stations can be calculated by integrating factors such as geographical distance, wind direction angle, and historical error correlation coefficients, thereby constructing a weighted graph adjacency matrix. Subsequently, a graph convolutional network is used to learn this graph structure and the original error sequences attached to it. Specifically, through multi-layer graph convolution operations, the network can adaptively extract the patterns of how errors influence each other between adjacent stations from the data, and its output can be equivalent to a data-driven spatial propagation operator. This operator quantifies the intensity and manner in which errors spread from one station to its surrounding neighborhood.

[0042] Finally, the spatial propagation operator learned above is combined with the mechanistic equations describing the physical process of error diffusion. By using the discretized propagation operator learned from a graph convolutional network and discretizing the time derivative using the explicit Euler method, a recursive formula that can be used for numerical computation is obtained. For example, this formula can be used to deduce the error estimate vectors for each station at the next time step, based on the error vectors of each station at the current time step. By iterating in this way, a complete dynamic weather forecast error propagation sequence that simultaneously reflects temporal evolution and spatial diffusion characteristics can be generated.

[0043] This implementation integrates discrete, isolated site forecast errors into a unified field with spatiotemporal continuity and physical consistency. This allows subsequent forecasting models not only to recognize current forecast biases at a specific location but also to proactively anticipate how these biases will affect surrounding areas over time and along spatial paths (such as downwind). For example, when a cold wave front forecast is inaccurate, this sequence can plot the propagation direction and intensity changes of the error several hours in advance, providing crucial, forward-looking information on the evolution of uncertainties for downstream power load forecasting models. This significantly enhances the model's robustness and accuracy during weather system transitions.

[0044] S202: Construct and train a multi-task learning model. The model includes a shared feature encoding layer, a main task branch for load forecasting, and an auxiliary task branch for meteorological error prediction. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors.

[0045] The core of this embodiment lies in constructing a multi-task learning model architecture capable of collaboratively learning the patterns of electricity demand and the evolution of meteorological errors. First, the multi-source collaborative features (including historical load features, meteorological-error coupling features, and time-cycle features) constructed in step S201 are organized into a unified feature vector at each time step, and a continuous historical window of length L is extracted to form the model input sequence. This input sequence is fed into a shared feature encoding layer, which is typically composed of a Long Short-Term Memory network or a Transformer encoder, and integrates a spatiotemporal attention mechanism.

[0046] Specifically, the temporal attention mechanism can adaptively weight the importance of different moments in the historical window; for example, the model may assign higher weights to meteorological disturbance features from hours before a cold wave arrives. Spatial attention, on the other hand, can highlight the impact of key meteorological elements in the feature dimension. The encoding layer outputs a context vector that integrates the deep coupling relationships of multi-source information, serving as a shared input for the two task branches.

[0047] The main load forecasting branch takes the aforementioned context vector as input and performs nonlinear mapping through several fully connected layers to directly output the predicted power load for a specific future time period (e.g., the next 24 hours). The auxiliary meteorological error forecasting branch takes the same context vector as input, but its output target is the predicted value of the dynamic meteorological forecast error propagation sequence constructed in S201 for a future time period. The two branches are trained collaboratively using a joint loss function, which is a weighted sum of the mean squared error loss of the main task and the mean squared error loss of the auxiliary task.

[0048] In this embodiment, multi-model task training employs a progressive strategy: First, the shared coding layer and the main task branch are pre-trained using only the load forecasting task, allowing them to initially grasp basic temporal patterns. Subsequently, auxiliary task branches are introduced for joint training, and the weights of the two task losses in the total loss are dynamically adjusted based on the meteorological uncertainty index evaluated in the previous training cycle. For example, when the model predicts that future error propagation will intensify (high uncertainty), the weight of the auxiliary task loss will be appropriately increased during training, driving the model to pay more attention to learning error evolution patterns.

[0049] This embodiment achieves collaborative modeling of meteorological information and its uncertainties through a shared coding layer and a dual-branch structure, and provides a built-in calibration mechanism by leveraging explicit error prediction of auxiliary tasks, thereby improving the overall accuracy and robustness of load forecasting. At the same time, the progressive and dynamic weight adjustment strategy effectively coordinates the multi-task learning process, improving training efficiency and model convergence stability.

[0050] In one specific optional implementation, the method further includes: Based on the dynamic weather forecast error propagation sequence, the weather-load demand coupling characteristics are constructed; The meteorological-load demand coupling characteristics are input into the multi-task learning model.

[0051] In one specific alternative implementation, a multi-task learning model is constructed and trained, including training the model using a phased strategy; In the first stage, the shared feature coding layer and the main load prediction task branch are trained using the load prediction task. In the second phase, the main task branch of joint load forecasting and the auxiliary task branch of meteorological error forecasting will undergo collaborative training.

[0052] In one specific optional implementation, during the second phase of training, the weights of the main load forecasting task branch and the auxiliary meteorological error forecasting task branch in the joint loss function are dynamically adjusted based on the meteorological uncertainty index assessed in the historical training cycle.

[0053] In this specific optional embodiment, further refinement of the method involves three key aspects: feature engineering, model training strategy, and optimization mechanism, which will be described in detail below.

[0054] First, at the feature construction level, this embodiment systematically constructs a series of meteorological-load demand coupling features based on dynamic weather forecast error propagation sequences. Specifically, it combines the original meteorological observations with the corresponding error propagation sequence components to generate "effective meteorological variables." Taking temperature as an example, the effective temperature... T eff ( t (Based on actual temperature) T obs ( t ) and temperature error propagation sequence components ΔT err ( t According to the formula T eff ( t ) =T obs ( t ) + γ ΔT err ( t ) was calculated to obtain, where γ This is the modulation coefficient, used to adjust the intensity of the forward-looking error effect. Based on this effective temperature, a "propagation-enhanced degree-day coefficient" can be further constructed. For example, when the effective temperature... T eff ( t (Higher than the refrigeration reference temperature) T cool At that time, the number of days of enhanced cooling was transmitted. PCDD ( t ) = max ( 0,T eff ( t ) − Tcool ) 。 This indicator, based on traditional degree-days, incorporates future temperature change trends predicted by error propagation, thus providing a more forward-looking representation of additional cooling demand caused by impending temperature rises. Furthermore, it constructs a load sensitivity coupling characteristic, obtaining sensitivity by calculating the rate of change of load with temperature within a local time window. S ( t This is coupled with the error propagation intensity to form a characteristic. F sens (t)= S (t)·|| ΔT err ( tThis feature amplifies the potential impact of meteorological errors on load forecasting during highly sensitive periods. All constructed coupled features are organized together with the original load history and time-coded features to form the input vector of the multi-task learning model.

[0055] Secondly, regarding the model training strategy, this embodiment adopts a phased and progressive strategy. The first phase is the pre-training phase, in which only the load forecasting main task branch is enabled. The aforementioned constructed feature vectors are input into the model, but only the mean squared error loss between the load forecast value and the actual value is calculated. This loss is then used to optimize the parameters of the shared feature encoding layer and the load forecasting main task branch through the backpropagation algorithm. The purpose of this phase is to enable the shared encoding layer to learn the basic temporal evolution law and core influencing factors of power load first, and establish a stable primary feature representation. The second phase is the collaborative training phase. After the shared encoding layer parameters have initially converged, the load forecasting main task branch and the meteorological error prediction auxiliary task branch are enabled simultaneously. The two branches share the context features output by the same encoding layer, but calculate the load forecasting loss and the error sequence prediction loss separately.

[0056] Finally, at the optimization mechanism level, this embodiment introduces a dynamic weight adjustment mechanism based on hysteresis uncertainty assessment during the second-stage collaborative training process. This mechanism aims to address the circular dependency or instability issues that may arise from the direct coupling between uncertainty quantification and weight adjustment during training. Specifically, at the end of each training cycle (e.g., a complete data traversal cycle), a comprehensive meteorological uncertainty index is calculated using the current model's prediction results of the meteorological error propagation sequence on the validation set.

[0057] This embodiment encodes the dynamic error propagation trend as input features and combines phased collaborative training with adaptive weight adjustment. This method effectively improves the model's prediction accuracy for load changes during meteorological transition periods, the stability of the training process, and its generalization ability in the face of different forecast uncertainties.

[0058] S203: Utilize the trained multi-task learning model to output the predicted power load and corresponding uncertainty quantification information based on real-time data.

[0059] In one specific optional implementation, the output of the predicted power load and the corresponding uncertainty quantification information based on real-time data includes: Calculate the meteorological uncertainty index based on the error results of the auxiliary task branch prediction; Based on the mapping relationship between meteorological uncertainty indicators and preset parameters, the disturbance magnitude of load forecast is determined; Based on the load forecast values ​​and disturbance magnitudes output by the main load forecasting task branch, a forecast confidence interval is generated.

[0060] Specifically, upon receiving the real-time collected power load sequence, the latest numerical weather forecast, and corresponding meteorological data, the system first generates the input feature vector for the current moment online according to the feature construction process described in S201, and feeds it into the pre-trained multi-task learning model. The model's forward propagation process executes two tasks in parallel: the main load forecasting branch outputs the predicted power load point sequence for future periods (e.g., the next 24 hours); the meteorological error prediction auxiliary branch outputs the predicted values ​​of the dynamic meteorological forecast error propagation sequence for the corresponding period.

[0061] Building upon this foundation, this embodiment further implements an uncertainty quantification process. First, based on the future error prediction sequence output by the auxiliary task branch, a comprehensive meteorological uncertainty index is calculated. Specifically, for example, the average L2 norm or variance of the predicted values ​​for key future periods (e.g., the next 6-12 hours) can be calculated as a quantification index; a higher index value indicates greater uncertainty in the weather forecast over future periods. Subsequently, through a pre-defined or online-learned mapping function, the uncertainty index is converted into an estimate of the disturbance magnitude of the load forecast values. Finally, by combining the disturbance magnitude calculated from the point prediction values ​​output by the main task branch, confidence intervals for the load forecast at each future time point are constructed. The system's final output includes the point prediction values, confidence intervals, and the corresponding uncertainty index levels.

[0062] This embodiment transforms the meteorological error cognition obtained through internal model collaborative learning into externally interpretable and actionable decision support information. By outputting quantitative prediction confidence intervals instead of single, definitive values, it significantly improves the information content and practicality of the prediction results.

[0063] In this embodiment, a multi-source collaborative dataset containing meteorological forecast error propagation sequences is constructed. A multi-task learning model is built and trained, comprising a shared feature encoding layer, a main task branch for load forecasting, and an auxiliary task branch for meteorological error prediction. The auxiliary task branch is used to predict the meteorological forecast error propagation sequences. Using the trained multi-task learning model, the predicted power load and corresponding uncertainty quantification information are output based on real-time data. By constructing and utilizing dynamic meteorological forecast error propagation sequences, the task learning model collaboratively models the evolution of load demand and forecast errors. This significantly improves the accuracy and robustness of load forecasting when there are deviations in the input meteorological forecast, and outputs prediction results with uncertainty quantification, providing a more reliable decision-making basis for power dispatch.

[0064] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0065] Figure 3A schematic diagram of a device for co-predicting power demand and weather forecast errors, corresponding one-to-one with the power demand and weather forecast error co-prediction method described in the above embodiments, is shown. Figure 3 As shown, the functional modules of this power demand and weather forecast error co-prediction device are described in detail below: Data construction module 301 is used to construct a multi-source collaborative dataset containing the propagation sequence of weather forecast errors; The model processing module 302 is used to build and train a multi-task learning model. The model includes a shared feature encoding layer, a main task branch for load prediction, and an auxiliary task branch for meteorological error prediction. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors. The prediction output module 303 is used to output the prediction results of power load and the corresponding uncertainty quantification information based on real-time data using the trained model.

[0066] Specific limitations regarding the power demand and weather forecast error co-prediction device can be found in the limitations of the power demand and weather forecast error co-prediction method described above, and will not be repeated here. Each module in the aforementioned power demand and weather forecast error co-prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0067] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0068] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components connected to the memory 41, processor 42, and network interface 43 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0069] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0070] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or D-interface display memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and the external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as the program code of the method for co-predicting electricity demand and weather forecast errors. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0071] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run program code stored in the memory 41 or process data, for example, to run program code for a method for co-predicting electricity demand and weather forecast errors.

[0072] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0073] This application also provides another embodiment, namely, a computer-readable storage medium storing an interface display program that can be executed by at least one processor to cause the at least one processor to perform the steps of the above-described method for co-predicting electricity demand and weather forecast errors.

[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0075] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for collaborative prediction of electricity demand and weather forecast errors, characterized in that, include: Construct a multi-source collaborative dataset containing the propagation sequence of weather forecast errors; A multi-task learning model is constructed and trained. The model includes a shared feature encoding layer, a main task branch for load prediction, and an auxiliary task branch for meteorological error prediction. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors. Using the trained multi-task learning model, the predicted power load and corresponding uncertainty quantification information are output based on real-time data.

2. The method according to claim 1, characterized in that, The construction of the multi-source collaborative dataset containing the propagation sequence of weather forecast errors includes: Acquire historical power load data, numerical weather forecast data, and meteorological observation data for the corresponding time periods; Based on the meteorological forecast data and the actual meteorological observation data, a dynamic meteorological forecast error propagation sequence is constructed.

3. The method according to claim 2, characterized in that, The step of constructing a dynamic weather forecast error propagation sequence based on the weather forecast data and the actual weather observation data includes: The difference between the forecast and actual values ​​of each meteorological station at different time steps is calculated to form the original error sequence; A graph structure representing the spatial relationships between meteorological stations is constructed, and a spatial propagation operator of error is learned from the original error sequence using a graph convolutional network; Based on the physical mechanism equation of error propagation, the spatial propagation operator is used to discretize the physical mechanism equation to generate the dynamic weather forecast error propagation sequence.

4. The method according to claim 2 or 3, characterized in that, The method further includes: Based on the dynamic weather forecast error propagation sequence, a weather-load demand coupling feature is constructed; The meteorological-load demand coupling features are input into the multi-task learning model.

5. The method according to claim 1, characterized in that, The construction and training of the multi-task learning model includes training the model using a phased strategy. In the first stage, the shared feature encoding layer and the main load prediction task branch are trained using the load prediction task; In the second stage, the main task branch for load forecasting and the auxiliary task branch for meteorological error forecasting are trained collaboratively.

6. The method according to claim 5, characterized in that, During the second phase of training, the weights of the main load forecasting task branch and the auxiliary meteorological error forecasting task branch in the joint loss function are dynamically adjusted based on the meteorological uncertainty index assessed in the historical training cycle.

7. The method according to claim 1, characterized in that, The method of outputting the predicted power load and the corresponding uncertainty quantification information based on real-time data includes: Based on the error results predicted by the auxiliary task branch, calculate the meteorological uncertainty index; Based on the meteorological uncertainty index and the preset mapping relationship, the disturbance magnitude of the load forecast is determined; Based on the load forecast value output by the main load forecasting task branch and the disturbance amplitude, a forecast confidence interval is generated.

8. A device for collaborative prediction of electricity demand and weather forecast error, characterized in that, include: The data construction module is used to build a multi-source collaborative dataset containing the propagation sequence of weather forecast errors; The model processing module is used to build and train a multi-task learning model. The model includes a shared feature encoding layer, a main task branch for load forecasting, and an auxiliary task branch for meteorological error forecasting. The auxiliary task branch is used to predict the propagation sequence of meteorological forecast errors. The prediction output module is used to output the prediction results of power load and the corresponding uncertainty quantification information based on real-time data using the trained model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for collaborative prediction of electricity demand and weather forecast error as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for collaborative prediction of electricity demand and weather forecast error as described in any one of claims 1 to 7.

Images