Power marketing management information platform daily power fitting method and related equipment

Abstract

The invention discloses an electric power marketing management information platform daily electric power fitting method and related equipment, and relates to the technical field of electric power data management, and the method comprises the steps: obtaining historical load data, external environment data and equipment operation state data of a target region; constructing an initial load fitting model based on the historical load data; extracting a date characteristic factor according to a preset time classification rule, and dynamically correcting the initial load fitting model based on the date characteristic factor to generate a corrected load model; generating a multi-source fusion feature based on the external environment data and the equipment operation state data; and inputting the multi-source fusion features into the corrected load model, and outputting a target load prediction result.

Description

Technical Field

[0001] The present application relates to the technical field of power data management, and in particular to a daily power fitting method and related equipment for a power marketing management information platform. Background Art

[0002] Daily power fitting for power marketing and management information platforms is a key technology for optimizing power system scheduling. Traditional fitting methods typically perform time-series modeling based on historical load data and establish prediction models through statistical learning or machine learning algorithms. With the development of smart grids, the dimensionality of power system data is becoming increasingly rich, but existing methods still have shortcomings in the fusion and processing of multi-source heterogeneous data. Most solutions only achieve feature integration through simple splicing or weighted averaging, which causes traditional prediction methods to easily accumulate prediction errors when faced with sudden changes, affecting the accuracy and timeliness of power grid scheduling decisions. Therefore, a daily power fitting method for power marketing and management information platforms is urgently needed to address the technical problems mentioned above. Summary of the Invention

[0003] The Summary of the Invention introduces a series of simplified concepts that will be further described in the Detailed Description of the Invention. The Summary of the Invention of this application is not intended to limit the key features and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0004] In a first aspect, the present application provides a daily power fitting method for a power marketing management information platform, comprising: Obtain historical load data, external environment data, and equipment operating status data for the target area; Based on historical load data, build an initial load fitting model; Extract date characteristic factors according to preset time classification rules, dynamically modify the initial load fitting model based on the date characteristic factors, and generate a modified load model; Generate multi-source fusion features based on external environment data and equipment operation status data; The multi-source fusion features are input into the modified load model and the target load prediction results are output.

[0005] In some embodiments, obtaining historical load data, external environment data, and equipment operating status data of a target area includes: Based on a preset collection frequency, extracting load time series data within a first preset time range of the target area from the power monitoring system to generate historical load data; Based on the geographical distribution of meteorological monitoring stations, obtain external environmental data aligned with the timestamps of historical load data. The external environmental data includes weather type, temperature series, and humidity series. Based on the Industrial Internet of Things protocol, the operating status data of the power equipment in the target area is analyzed. The operating status data includes transformer temperature, line load rate and equipment start and stop status.

[0006] In some embodiments, constructing an initial load fitting model based on historical load data includes: Based on the time series of historical load data, the load fluctuation amplitude parameters within the preset time window are extracted; Based on the load fluctuation amplitude parameter, the separation threshold of the trend item component and the random item component of the historical load data is determined; Decompose the historical load data into components based on the separation threshold to generate trend item component data and random item component data; Based on the fitting residuals of the trend component data and the fluctuation characteristics of the random component data, a candidate model structure is selected from the preset model library; The trend item component data and the random item component data are jointly trained based on the candidate model structure to generate an initial load fitting model, wherein the initial load fitting model includes a trend capturing module and a fluctuation compensation module.

[0007] In some embodiments, extracting date characteristic factors according to preset time classification rules, dynamically revising the initial load fitting model based on the date characteristic factors, and generating a revised load model includes: Based on preset time classification rules, the timestamps of historical load data are divided into weekday type, weekend type and holiday type, wherein the preset time classification rules include a date tag matching algorithm and a statutory holiday database; Based on the historical load data corresponding to each date type after classification, the date characteristic factors are extracted, where the date characteristic factors include the load fluctuation baseline on weekdays, the load peak period on weekends, and the load decay rate on holidays; Based on the date characteristic factors, determine the timing weight parameters of the trend capture module and the sensitivity coefficient of the fluctuation compensation module; Based on the time series weight parameters and sensitivity coefficients, the long-term trend fitting function of the trend capture module and the random fluctuation compensation function of the fluctuation compensation module are recalibrated. The initial load fitting model is updated based on the recalibrated parameters to generate a revised load model, wherein the output result of the revised load model includes a load forecast curve matching the date type.

[0008] In some embodiments, generating multi-source fusion features based on external environment data and device operating status data includes: Based on the weather type, a weather feature vector is generated through a classification coding algorithm; Based on the temperature series and humidity series, standardized environmental parameters are generated through normalization processing; Based on transformer temperature and line load factor, a dynamic threshold algorithm is used to generate equipment status fluctuation characteristics. The dynamic threshold algorithm calculates the real-time deviation rate based on the preset safe operation threshold range in historical equipment data. Generate a timestamp sequence of device start and stop events based on the device start and stop status; Construct a multidimensional feature input matrix based on weather feature vectors, standardized environmental parameters, equipment state fluctuation characteristics, and timestamp sequences of equipment start and stop events; Based on the multi-dimensional feature input matrix, the first association weight between the weather feature vector and the device state fluctuation feature, as well as the second association weight between the standardized environmental parameters and the timestamp sequence of the device start and stop events, are calculated through the cross-attention mechanism. Based on the weighted fusion result of the first association weight and the second association weight, a multi-source fusion feature containing spatiotemporal correlation is generated.

[0009] In some embodiments, the multi-source fusion features are input into the modified load model to output the target load prediction result, including: Based on the long-term trend fitting function, the spatiotemporal correlation in the multi-source fusion characteristics is analyzed to generate the base load forecast value; Based on the random fluctuation compensation function, the device state fluctuation characteristics in the multi-source fusion characteristics are analyzed to generate a dynamic compensation coefficient; The basic load prediction value and the dynamic compensation coefficient are linearly superimposed to generate the initial predicted load curve; Based on the timestamp sequence of equipment start and stop events, the initial forecast load curve is modified by event-driven to generate intermediate forecast results; Based on the modified load model, calculate the residual sequence between the intermediate prediction results and the real-time collected load data; Based on the sliding window statistics of the residual sequence, the time series weight parameters and sensitivity coefficients are updated; Based on the updated time series weight parameters and sensitivity coefficients, the intermediate prediction results are recalibrated to generate the target load forecast results.

[0010] In some embodiments, further comprising: Based on the target load forecast results, a probability distribution is generated through the Monte Carlo simulation algorithm; Calculate dynamic confidence intervals based on probability distributions; Based on the dynamic confidence interval and target load forecast results, a visualization chart is generated, which includes the target load forecast curve, the real-time collected load data curve and the confidence interval boundary.

[0011] In a second aspect, the present application proposes a daily power fitting device for a power marketing management information platform, comprising: Power data acquisition unit, used to obtain historical load data, external environment data and equipment operating status data of the target area; An initial model building unit builds an initial load fitting model based on historical load data; A load model correction unit is used to extract date characteristic factors according to preset time classification rules, dynamically correct the initial load fitting model based on the date characteristic factors, and generate a corrected load model; The multi-source feature fusion unit generates multi-source fusion features based on external environment data and equipment operation status data; The load forecast output unit is used to input multi-source fusion features into the modified load model and output the target load forecast result.

[0012] In a third aspect, an electronic device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to implement the steps of the daily power fitting method of the power marketing management information platform of any one of the first aspects when executing the computer program stored in the memory.

[0013] In a fourth aspect, the present application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the daily power fitting method of the power marketing management information platform of any one of the first aspects.

[0014] In summary, this application constructs multi-source dynamic features by integrating historical load data, external environmental data, and equipment operating status data, and dynamically corrects the initial model based on date characteristic factors, thereby improving prediction accuracy and scenario adaptability. First, the multi-source data fusion mechanism effectively captures the synergistic impact of environmental factors and equipment operating status on load, enhancing the model's ability to analyze complex working conditions; second, the dynamic correction strategy automatically adjusts the model parameters according to the date type, accurately matching the load change patterns under different time scenarios; finally, the residual feedback mechanism optimizes the model weights in real time to ensure that the prediction results can quickly respond to real-time load fluctuations, providing high-precision and high-robustness decision support for power grid dispatching. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present description. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 A flowchart of a daily power fitting method for a power marketing management information platform provided in an embodiment of the present application; Figure 2A schematic diagram of the structure of a daily power fitting device for a power marketing management information platform provided in an embodiment of the present application; Figure 3 This is a schematic diagram of the structure of an electronic device for daily power fitting of a power marketing management information platform provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments.

[0017] See also Figure 1 , which is a flowchart of a daily power fitting method for a power marketing management information platform provided in an embodiment of the present application, and may specifically include: S110, obtaining historical load data, external environment data, and equipment operating status data of the target area; For example, through the integration of multi-dimensional data sources, basic data support is provided for the daily power fitting of the power marketing management information platform. First, historical load data is extracted by the power monitoring system at a preset collection frequency, covering the load time series changes in the target area within the first preset time range, forming a core data set that reflects the long-term trend and short-term fluctuations in power demand. Secondly, based on the geographical distribution characteristics of meteorological monitoring stations, external environmental data obtains weather type, temperature and humidity sequences that are strictly aligned with the load data timestamps to ensure the spatiotemporal correlation between environmental factors and load changes. Equipment operating status data is analyzed in real time through the industrial Internet of Things protocol, collecting key parameters such as transformer temperature and line load rate, providing the model with dynamic observation indicators of the impact of equipment operating conditions on power load.

[0018] The above steps aim to achieve the collaborative acquisition and alignment of heterogeneous data from multiple sources. Data granularity is constrained by presetting the acquisition frequency, and a timestamp alignment mechanism is used to eliminate temporal deviations in multi-source data. Industrial IoT protocols are also leveraged to ensure the real-time and integrity of device status data. This data integration strategy provides highly consistent and comprehensive input features for subsequent model construction, effectively overcoming the prediction bias caused by a single data dimension in traditional methods and laying the foundation for capturing the multi-factor driving mechanisms of load changes.

[0019] S120, constructing an initial load fitting model based on historical load data; For example, the construction of the initial load fitting model focuses on the inherent pattern mining of historical load data. By extracting the load fluctuation amplitude parameters within a preset time window, the historical load data is decomposed into trend terms and random terms, separating the stable component reflecting the long-term trend and the random component representing short-term fluctuations. Based on the fitting residuals and fluctuation characteristics of the components, matching candidate model structures are adaptively selected from the preset model library to achieve differentiated modeling of trend terms and random terms. Through a joint training strategy, the trend capture module and the fluctuation compensation module are integrated into the initial model, forming a dual-module prediction framework that takes into account both long-term trend analysis and short-term fluctuation compensation.

[0020] The above steps aim to establish the basic model's ability to analyze load time series characteristics. The trend capture module captures macroscopic variations in power demand by fitting long-term trend components, while the fluctuation compensation module provides a dynamic compensation mechanism for the high-frequency fluctuations of random components. The synergy between these two modules provides a stable initial forecast baseline for subsequent dynamic model adjustments, while also reserving feature interfaces for multi-source data fusion, ensuring the scalability of the model architecture.

[0021] S130, extracting date characteristic factors according to preset time classification rules, dynamically revising the initial load fitting model based on the date characteristic factors, and generating a revised load model; For example, a date characteristic factor is extracted through preset time classification rules and used to dynamically modify the initial load fitting model. This aims to address the load pattern deviation problem caused by differences in date types in traditional forecasting methods. The date characteristic factor analyzes the load fluctuation characteristics corresponding to weekdays, weekends, and holidays in historical load data to extract key parameters such as the weekday load fluctuation baseline, weekend load peak period, and holiday load decay rate. These parameters reflect the inherent differences in power demand patterns under different date types and provide a basis for dynamic adjustment of model parameters.

[0022] The dynamic correction process maps date-specific factors to the model's trend capture and fluctuation compensation modules, enabling differentiated configuration of model parameters. For example, the weekday load fluctuation baseline is used to adjust the trend capture module's timing weights to enhance its ability to fit stable trends; the holiday load decay rate is used to calibrate the fluctuation compensation module's sensitivity coefficient to enhance its response to sudden load drops. This dynamic adjustment mechanism enables the model to adaptively match load fluctuation patterns under different date scenarios, thereby improving the scenario-wide generalization of the forecast results.

[0023] S140, generating multi-source fusion features based on external environment data and device operation status data; For example, by integrating external environmental data with equipment operating status data, multidimensional correlation features are constructed to enhance the model's ability to analyze complex influencing factors. External environmental data, including weather type, temperature, and humidity series, are standardized to eliminate dimensional differences and transform into quantifiable input features. Equipment operating status data, including dynamic parameters such as transformer temperature and line load factor, directly characterizes the operating efficiency of power grid equipment and the potential risk of load fluctuations. These two types of data are structured to form an initial feature set.

[0024] A deep fusion of multi-source features is achieved through a spatiotemporal correlation analysis mechanism. A cross-attention algorithm is used to explore the coupling relationship between weather characteristics and equipment status fluctuations. The temporal correlation between environmental parameters and equipment start and stop events is also established, ultimately generating multi-source fusion features with spatiotemporal consistency. These features can characterize the synergistic effects of external environmental changes and equipment operating status on power load, providing the model with multi-dimensional dynamic input information and supporting subsequent high-precision load forecasting.

[0025] S150: Input the multi-source fusion features into the modified load model and output the target load prediction result.

[0026] For example, a modified load model analyzes multi-source fusion features to generate a final forecast. Based on dynamically adjusted time-series weight parameters and sensitivity coefficients, the modified load model analyzes multidimensional input features, including environmental factors, equipment status, and date characteristics. The trend capture module extracts long-term load variation patterns, and the fluctuation compensation module dynamically corrects short-term fluctuations caused by abnormal equipment conditions. The forecast curve is then event-drivenly modified using the timestamp sequence of equipment start and stop events, ultimately outputting a target load forecast that matches the real-time operating conditions.

[0027] The above steps, through the synergistic effect of multi-level feature fusion and dynamic model correction, ensure that the forecast results not only reflect long-term trends in load changes, but also accurately capture short-term disturbances caused by sudden changes in the external environment and equipment status. The output results are used to update model parameters in real time through a residual feedback mechanism, forming a closed-loop optimization system, providing highly robust and adaptable load forecast data support for power grid dispatch.

[0028] In summary, the embodiment of the present application constructs multi-source dynamic features by integrating historical load data, external environmental data and equipment operating status data, and dynamically corrects the initial load fitting model in combination with date characteristic factors, thereby improving the accuracy and scenario adaptability of daily power fitting of the power marketing management information platform. The multi-source data fusion mechanism effectively integrates environmental parameters such as weather type, temperature, and humidity with real-time equipment operating status such as transformer temperature and line load rate, and uses the cross-attention algorithm to mine the spatiotemporal correlation between environmental factors and equipment status, thereby enhancing the model's ability to analyze the synergistic impact of complex working conditions. The dynamic correction strategy is based on date type characteristics such as weekdays, weekends and holidays, extracts key parameters such as load fluctuation baseline, peak period and attenuation rate, adaptively adjusts the model's time series weight and sensitivity coefficient, and accurately matches the load change pattern under different time scenarios. The residual feedback mechanism dynamically optimizes model parameters and recalibrates the prediction curve through deviation analysis between real-time collected data and prediction results, ensuring that the prediction results quickly respond to sudden load fluctuations. The final output target load forecast results, combined with confidence interval visualization, provide a highly robust and timely decision-making basis for power grid dispatching, effectively solving the technical defects of traditional methods such as insufficient data integration, model rigidity and delayed response.

[0029] In some instances, historical load data, external environment data, and equipment operating status data of the target area are obtained, including: Based on a preset collection frequency, extracting load time series data within a first preset time range of the target area from the power monitoring system to generate historical load data; Based on the geographical distribution of meteorological monitoring stations, obtain external environmental data aligned with the timestamps of historical load data. The external environmental data includes weather type, temperature series, and humidity series. Based on the Industrial Internet of Things protocol, the operating status data of the power equipment in the target area is analyzed. The operating status data includes transformer temperature, line load rate and equipment start and stop status.

[0030] Exemplarily, based on the preset acquisition frequency, the load time series data of the target area within the first preset time range is extracted from the power monitoring system to generate historical load data. The preset acquisition frequency is set to a minute or hourly sampling interval according to the grid dispatching requirements, for example, once every 15 minutes or every hour, to ensure the time series continuity and integrity of the load data. The first preset time range is dynamically adjusted according to the forecast demand, and usually covers at least one complete seasonal cycle (such as 12 months) to fully reflect the long-term trend and periodic characteristics of the load changes. In specific implementation, the original load data within the time range is batch extracted according to the preset acquisition frequency through the data interface of the power monitoring system, and the data is cleaned (such as removing outliers, filling in missing values) and the format is standardized to generate a load time series data set with a unified timestamp. The above data set contains the power consumption of the target area at different time points, which is used for subsequent modeling and analysis of the long-term trend and short-term fluctuation pattern of load changes.

[0031] Based on the geographical distribution of meteorological monitoring stations, external environmental data that is aligned with the timestamps of historical load data is obtained. The external environmental data includes weather types, temperature series, and humidity series. Specifically, according to the geographical scope covered by the target area, the location coordinates of neighboring meteorological monitoring stations are matched to ensure the spatial consistency of meteorological data and load data; at the same time, through the timestamp alignment algorithm, the time resolution of meteorological data (such as hourly updates) is synchronized with the collection frequency of load data to generate an external environmental data set with a completely aligned time dimension. For example, if the load data is collected hourly, the temperature and humidity series in the meteorological data are interpolated or aggregated at the same time granularity, and the weather type is labeled according to the main meteorological state within the hour. The above steps eliminate the spatiotemporal deviations between data sources through the dual spatiotemporal alignment mechanism to ensure that the correlation analysis between environmental factors and load changes has a high-precision basis.

[0032] Based on the Industrial Internet of Things protocol, the operating status data of the power equipment in the target area is parsed. The operating status data includes transformer temperature, line load rate, and equipment start and stop status. In specific implementation, the IoT sensors deployed on the power equipment collect real-time temperature values ​​of the transformer windings, the current load rate of the transmission line (the ratio of actual load to rated capacity expressed as a percentage), and the switch signals of the equipment start and stop status. The Industrial Internet of Things protocol is a standardized communication protocol such as Modbus, OPC UA, or MQTT, which is used to convert the raw data collected by the sensors into structured data and transmit it to the central data processing platform. During the parsing process, the equipment start and stop events are timestamped to generate an event sequence aligned with the load data timeline. For example, when a transformer is detected to be shut down due to overload protection, the precise time point and duration of the shutdown event are recorded, providing key input for subsequent analysis of the impact of sudden changes in equipment status on the load.

[0033] In some examples, an initial load fitting model is constructed based on historical load data, including: Based on the time series of historical load data, the load fluctuation amplitude parameters within the preset time window are extracted; Based on the load fluctuation amplitude parameter, the separation threshold of the trend item component and the random item component of the historical load data is determined; Decompose the historical load data into components based on the separation threshold to generate trend item component data and random item component data; Based on the fitting residuals of the trend component data and the fluctuation characteristics of the random component data, a candidate model structure is selected from the preset model library; The trend item component data and the random item component data are jointly trained based on the candidate model structure to generate an initial load fitting model, wherein the initial load fitting model includes a trend capturing module and a fluctuation compensation module.

[0034] Exemplarily, based on the time series of historical load data, the load fluctuation amplitude parameters are extracted within a preset time window through a sliding window algorithm. The preset time window is set to a fixed length (for example, 24 hours, 7 days, or 30 days) according to the periodic characteristics of the load changes, and is used to capture the load fluctuation characteristics at different time scales. In specific implementation, the maximum value, minimum value, and standard deviation of the load data in each time window are calculated to generate an amplitude parameter reflecting the intensity of the load fluctuation within the window. For example, for hourly load data, 24 hours is set as the time window, and the standard deviation of the load in each window is calculated in sequence by sliding to generate a 24-dimensional fluctuation amplitude parameter sequence. The above steps provide a data basis for subsequent component decomposition by quantitatively analyzing the load fluctuation characteristics at different time scales.

[0035] Based on the load fluctuation amplitude parameter, the separation threshold of the trend item component and the random item component is determined by cluster analysis or statistical distribution fitting. Specifically, the probability density estimation of the fluctuation amplitude parameter is performed to identify its distribution characteristics (such as bimodal distribution), and the distribution valley point or preset quantile (such as 15% quantile) is set as the separation threshold. When the fluctuation amplitude is lower than the threshold, it is determined to be a random item component that characterizes short-term random fluctuations; when it is higher than the threshold, it is determined to be a trend item component that reflects the long-term change trend. For example, if the 15% quantile value of the fluctuation amplitude parameter is 20kW, the data with a load fluctuation amplitude less than or equal to 20kW is classified as a random item, and the data greater than 20kW is classified as a trend item. The above steps achieve preliminary classification of load data by quantifying the differences in fluctuation intensity.

[0036] Based on the separation threshold, the empirical mode decomposition (EMD) or wavelet transform algorithm is used to decompose the historical load data into components. For the trend term component, the smoothed long-term trend curve is extracted by low-pass filtering or moving average method; for the random term component, the high-frequency fluctuation sequence is separated by high-pass filtering or residual calculation. During the decomposition process, it is ensured that the trend term and the random term are orthogonal in the time domain, and the error between the superposition result of the two and the original data is less than the preset tolerance (such as 1%). For example, EMD is used to decompose the original load data into multiple intrinsic mode functions (IMFs), and the low-frequency IMFs are superimposed as trend terms, and the high-frequency IMFs are superimposed as random terms. The above steps realize the explicit expression of the inherent laws of the load through mathematical decomposition, ensuring the effectiveness of the separation of trend terms and random terms in the time domain and frequency domain.

[0037] Based on the fitted residuals of the trend component data (i.e., the deviation between the fitted trend component value and the actual value) and the volatility characteristics of the random component data (such as variance and autocorrelation coefficient), candidate model structures are adaptively selected from a preset model library. This preset model library contains a variety of predefined forecasting models. Candidate models for the trend capture module include linear regression, autoregressive integrated moving average (ARIMA), and long short-term memory (LSTM), which are suitable for capturing long-term patterns in trend components. Candidate models for the volatility compensation module include GARCH models, random forests, and GRU networks, which are suitable for different volatility characteristics. During the selection process, the mean squared error (MSE) of the trend component fitted residuals and the explained variance (EV) of the random component fluctuations are calculated for each model. The model combination with the lowest MSE and highest EV is selected. For example, if the MSE of the trend component residuals is lowest in the LSTM model, and the EV of the random component is highest in the GARCH model, the LSTM+GARCH model is selected as the candidate model structure. These steps quantitatively evaluate the fit between the model and the components, achieving adaptive optimization of the model structure.

[0038] Based on the selected candidate model structure, the trend term component data and the random term component data are jointly trained. During training, the trend capture module uses the trend term component as input to fit the long-term load variation function; the fluctuation compensation module uses the random term component as input to learn a short-term fluctuation compensation strategy. The parameters of these two modules are synchronously updated using an alternating optimization algorithm (such as coordinate descent) to ensure the synergy between the trend and fluctuation models. The resulting initial load fitting model consists of a trend capture module and a fluctuation compensation module. The trend capture module outputs a base load forecast value, reflecting the long-term load variation trend; the fluctuation compensation module outputs a dynamic compensation coefficient to correct for the impact of short-term random fluctuations. For example, an LSTM is used to fit the trend term, and a GRU is used to fit the random term, with the final forecast value generated through weighted summation. This step, through differentiated modeling and joint training, achieves a multi-level analysis of load variation patterns, providing a high-precision initial model benchmark for subsequent dynamic correction and multi-source fusion.

[0039] In some examples, date characteristic factors are extracted according to preset time classification rules, and the initial load fitting model is dynamically modified based on the date characteristic factors to generate a modified load model, including: Based on preset time classification rules, the timestamps of historical load data are divided into weekday type, weekend type and holiday type, wherein the preset time classification rules include a date tag matching algorithm and a statutory holiday database; Based on the historical load data corresponding to each date type after classification, the date characteristic factors are extracted, where the date characteristic factors include the load fluctuation baseline on weekdays, the load peak period on weekends, and the load decay rate on holidays; Based on the date characteristic factors, determine the timing weight parameters of the trend capture module and the sensitivity coefficient of the fluctuation compensation module; Based on the time series weight parameters and sensitivity coefficients, the long-term trend fitting function of the trend capture module and the random fluctuation compensation function of the fluctuation compensation module are recalibrated. The initial load fitting model is updated based on the recalibrated parameters to generate a revised load model, wherein the output result of the revised load model includes a load forecast curve matching the date type.

[0040] For example, based on preset time classification rules, the timestamps of historical load data are classified into weekday, weekend, and holiday types. The preset time classification rules include a date label matching algorithm and a statutory holiday database. The date label matching algorithm parses the day of the week information in the timestamp based on the International Standard Date Format (ISO 8601), marking Monday through Friday as weekdays and Saturday and Sunday as weekends, completing the initial classification. The statutory holiday database stores statutory holidays and offset dates announced by the country or region. By accurately matching timestamps, holidays and offset days are separated from the weekday or weekend types and marked as holiday types. For example, if a timestamp is October 1, 2023, the date label matching algorithm identifies it as a Sunday (weekend type), but the statutory holiday database determines it to be a National Day holiday, and the final classification is the holiday type. If a Saturday becomes a weekday due to an offset, the offset record in the database is used to classify it as a weekday. The above steps ensure that the date type classification conforms to the conventional calendar rules and accurately reflects the actual holiday schedule through a double verification mechanism, providing a reliable date label foundation for the subsequent extraction of date feature factors.

[0041] Based on the historical load data corresponding to each date type after division, date characteristic factors are extracted, including: weekday load fluctuation baseline, statistical mean and standard deviation of hourly load data on weekdays, and generation of a time series characteristic curve reflecting the typical load fluctuation range on weekdays; weekend load peak period, through cluster analysis to identify the peak period of weekend load data (such as 12:00-14:00 and 18:00-20:00), and calculate the start time, duration and peak intensity of the load surge during the weekend; holiday load decay rate: fit an exponential decay function to the load data during holidays, and calculate the decay rate parameter of the load from the peak value to the stable value (such as the percentage of decrease per hour). For example, for holiday types, the load decay curve is fitted by the least squares method, and the decay coefficient is extracted as the characteristic factor. The above characteristic factors provide a data-driven basis for the dynamic correction of model parameters by quantifying the load change patterns of different date types.

[0042] Based on the date characteristic factors, the timing weight parameters of the trend capture module and the sensitivity coefficient of the fluctuation compensation module are determined, including: The time series weight parameter adjusts the trend capture module's fit strength for long-term trends based on the stability of the workday load fluctuation baseline. For example, if the workday baseline fluctuation is small (the standard deviation is below the preset threshold), the trend capture module's weight is increased to strengthen the fit of the stable trend; The sensitivity coefficient sets the fluctuation compensation module's response to sudden fluctuations based on weekend load peaks and holiday attenuation rates. For example, if the sudden increase in load during the weekend peak period exceeds the historical average by 20%, the sensitivity coefficient is increased to enhance peak load compensation. During the parameter mapping process, dynamic parameter adjustment is achieved through a mapping function (such as a piecewise linear function or a sigmoid function) between characteristic factors and model parameters, ensuring that the model parameters accurately match the load characteristics of the date type.

[0043] Based on the time series weight parameters and sensitivity coefficients, the long-term trend fitting function of the trend capture module and the random fluctuation compensation function of the fluctuation compensation module are recalibrated, including: The trend capture module adjusts the long-term trend fitting function (such as the autoregressive coefficient in the ARIMA model or the hidden layer weight in the LSTM network) based on the time series weight parameters. For example, it updates the trend term fitting parameters through weighted least squares to make it more consistent with the weekday baseline characteristics. The fluctuation compensation module reconfigures the random fluctuation compensation function (such as the volatility equation of the GARCH model or the gating weight of the GRU network) based on the sensitivity coefficient, for example, by scaling the output gain of the compensation function so that it can quickly respond to load surges during weekend peak periods.

[0044] During parameter recalibration, a gradient descent algorithm is used to simultaneously optimize the parameters of both modules to ensure synergy between trend capture and fluctuation compensation. For example, during the decline phase of holidays, the weight of the trend capture module is reduced, while the sensitivity of the fluctuation compensation module is increased to accurately capture rapid load declines.

[0045] The initial load fitting model is updated based on the recalibrated parameters to generate a revised load model, including: The recalibrated time series weight parameters and sensitivity coefficients are integrated into the initial model architecture to form a modified load model that is adaptive to the date type. When making predictions, the modified model automatically matches the date type according to the timestamp of the input data and calls the corresponding parameter configuration to generate the load forecast curve. For example, when the forecast date is a weekday, a parameter combination of high trend weight and standard sensitivity coefficient is used to enhance the trend capture module's fitting of stable fluctuations and output a smooth forecast result that conforms to the weekday baseline; when the forecast date is a holiday, a low trend weight and high sensitivity coefficient configuration is enabled to accurately reflect the holiday load attenuation characteristics. The above steps enable the model to adapt to the load change patterns under different date scenarios in real time through a dynamic parameter switching mechanism, and ultimately output a high-precision forecast result that matches the date type.

[0046] In some instances, multi-source fusion features are generated based on external environment data and device operating status data, including: Based on the weather type, a weather feature vector is generated through a classification coding algorithm; Based on the temperature series and humidity series, standardized environmental parameters are generated through normalization processing; Based on transformer temperature and line load factor, a dynamic threshold algorithm is used to generate equipment status fluctuation characteristics. The dynamic threshold algorithm calculates the real-time deviation rate based on the preset safe operation threshold range in historical equipment data. Generate a timestamp sequence of device start and stop events based on the device start and stop status; Construct a multidimensional feature input matrix based on weather feature vectors, standardized environmental parameters, equipment state fluctuation characteristics, and timestamp sequences of equipment start and stop events; Based on the multi-dimensional feature input matrix, the first association weight between the weather feature vector and the device state fluctuation feature, as well as the second association weight between the standardized environmental parameters and the timestamp sequence of the device start and stop events, are calculated through the cross-attention mechanism. Based on the weighted fusion result of the first association weight and the second association weight, a multi-source fusion feature containing spatiotemporal correlation is generated.

[0047] Exemplarily, based on the weather type in the external environment data (such as sunny, rainy, snowy, etc.), a weather feature vector is generated using a one-hot encoding or embedding layer algorithm. Specifically, each weather type is mapped to a fixed-length numerical vector, for example, sunny is encoded as [1,0,0], rainy is [0,1,0], and snowy is [0,0,1]. If the weather type contains multiple labels (such as cloudy to light rain), a sparse vector is generated through multi-label encoding. The classification coding algorithm achieves accurate mapping between types and vectors through a preset weather type dictionary to ensure that the weather feature vector Quantifiable input to the model.

[0048] Based on the temperature series and humidity series, the dimension difference is eliminated through normalization processing, and the temperature data is normalized by Z-score. The calculation formula is: in, is the standardized temperature series; is the historical temperature average, is the standard deviation of historical temperature; the processed temperature series is converted into standardized environmental parameters with a mean of 0 and a variance of 1 to ensure that different environmental variables are comparable in model training.

[0049] The humidity data is normalized by the maximum and minimum values, and the calculation formula is: in, is the standardized humidity series; This is the lowest humidity in history; This is the highest humidity in history.

[0050] Based on the transformer temperature and line load rate, the dynamic threshold algorithm is used to generate the equipment status fluctuation characteristics. The dynamic threshold algorithm is based on the preset safe operation threshold range in the historical equipment data (such as the transformer temperature threshold , , Line load rate threshold , ), calculate the real-time deviation rate, expressed as: in, 、 is the rated operating parameter of the equipment; and Weighted summation to generate equipment status fluctuation characteristics , weight 、 The above steps quantify the abnormality of the equipment status and generate fluctuation characteristics that represent the risk of equipment operation.

[0051] Based on the start and stop status of the equipment, the precise timestamps of equipment state switching events (such as transformer start, line disconnection) are recorded to generate a discrete event sequence. Specifically, when the equipment state changes from "running" to "stopping" or vice versa, the event type (start / stop) and the time of occurrence (accurate to milliseconds) are recorded and sorted in chronological order to generate a timestamp sequence. , where each Associated event type tags. The timestamp sequence is strictly aligned with the time axis of historical load data, providing a temporal correlation basis for subsequent event-driven corrections.

[0052] Based on weather feature vector , standardized environmental parameters and , Equipment status fluctuation characteristics , equipment start and stop event timestamp sequence , construct a time-aligned multi-dimensional feature input matrix. Specifically, each feature is aligned along the time axis and concatenated into a matrix form, expressed as: Among them, the timestamp sequence The matrix is ​​represented by a binary mask (1 if the event occurs and 0 if it does not occur) to ensure the consistency of the matrix dimensions.

[0053] Based on multi-dimensional feature input matrix , through the cross-attention mechanism, the association weights between weather features and device state fluctuation features, as well as the temporal association weights between environmental parameters and device start and stop events are calculated, including: For weather feature vector Fluctuation characteristics of device status , through linear transformation, the weather characteristic vector Mapping to query matrix , the device state fluctuation characteristics Mapping to a key matrix , where the mapping weight matrix is ​​learned through model training; based on the query matrix With the key matrix , calculate the attention score matrix by scaling the dot product attention formula; perform Softmax normalization on the attention score matrix to generate the first association weight matrix ,The element value of the first association weight matrix represents the ,association strength of different weather types on the fluctuation of the ,device state.

[0054] For standardized environmental parameters 、 Sequence of device start and stop event timestamps , standardize environmental parameters 、 Sequence of events related to device start and stop Aligned by time axis, the input is fed into a temporal convolutional network (TCN). TCN contains multiple layers of causal dilation convolution kernels to capture the long-term temporal dependencies between environmental parameters and device events. The temporal correlation matrix between environmental parameters and device events is calculated using the high-level features output by TCN. , whose dimensions are ,in, is the time step, is the environmental parameter dimension, is the number of event types; for the time correlation matrix Perform global average pooling and map it into the second association weight matrix through the fully connected layer ,The element values ​​of the second association weight matrix reflect the ,temporal coupling strength between the environmental parameter changes and the ,startup and shutdown events of the equipment.

[0055] The first association weight matrix The sensitivity of equipment status fluctuations under different weather conditions is quantified. For example, rainy days may cause abnormal temperature rise of transformers. The second correlation weight matrix The model reveals the temporal causal relationship between environmental parameters (such as sudden temperature rise) and equipment start-up and shutdown events (such as line overload protection triggering). Through the cross-attention mechanism, the model dynamically captures the complex correlation patterns between multi-source features, providing a precise basis for weight allocation for subsequent load forecasting.

[0056] Based on the first association weight matrix With the second associated weight matrix , perform weighted fusion on the original features, expressed as: in, Represents matrix multiplication; It is the feature splicing operation; 、 is the preset fusion coefficient. The multi-source fusion feature finally generated It includes the spatiotemporal correlation of weather, environment, equipment status and events, providing high-information-density input features for the load forecasting model.

[0057] In summary, the embodiments of the present application achieve deep fusion of multi-source heterogeneous data through structured processing and correlation weight calculation. The correlation weights between weather feature vectors and equipment state fluctuation characteristics reflect the impact of meteorological conditions on equipment operating conditions; the temporal weights between environmental parameters and equipment events capture the potential patterns of environmental changes triggering sudden changes in equipment states. The multi-source fusion features ultimately generated effectively integrate the synergy between the external environment and the equipment operating status, providing input for the model to analyze complex load driving mechanisms, thereby improving prediction accuracy and robustness.

[0058] In some examples, multi-source fusion features are input into a modified load model to output target load forecast results, including: Based on the long-term trend fitting function, the spatiotemporal correlation in the multi-source fusion characteristics is analyzed to generate the base load forecast value; Based on the random fluctuation compensation function, the device state fluctuation characteristics in the multi-source fusion characteristics are analyzed to generate a dynamic compensation coefficient; The basic load prediction value and the dynamic compensation coefficient are linearly superimposed to generate the initial predicted load curve; Based on the timestamp sequence of equipment start and stop events, the initial forecast load curve is modified by event-driven to generate intermediate forecast results; Based on the modified load model, calculate the residual sequence between the intermediate prediction results and the real-time collected load data; Based on the sliding window statistics of the residual sequence, the time series weight parameters and sensitivity coefficients are updated; Based on the updated time series weight parameters and sensitivity coefficients, the intermediate prediction results are recalibrated to generate the target load forecast results.

[0059] Exemplarily, based on the long-term trend fitting function (such as LSTM network or ARIMA model) of the trend capture module in the modified load model, the spatiotemporal correlation in the multi-source fusion features is analyzed. The long-term trend fitting function captures the macroscopic variation pattern of electricity demand by analyzing the long-term dependency between the trend item components of the historical load data and the weather feature vectors and standardized environmental parameters in the multi-source fusion features. Specifically, the multi-source fusion features are input into the input layer of the trend capture module, and the stable trend item components of the load changes are extracted through a multi-layer neural network or a time series regression algorithm to generate a basic load forecast value. The basic load forecast value reflects the long-term load trend of the target area under ideal environment and normal equipment operation, and its output is in the form of a continuous load curve corresponding to the forecast time range.

[0060] The fluctuation compensation module of the modified load model analyzes the equipment state fluctuation characteristics in the multi-source fusion features based on a random fluctuation compensation function (such as a GRU network or GARCH model). The equipment state fluctuation characteristics are generated by calculating the real-time deviation rate of transformer temperature and line load factor using a dynamic threshold algorithm, representing the short-term disturbance intensity of abnormal equipment operation on the load. The fluctuation compensation module learns the nonlinear mapping relationship between the equipment state fluctuation characteristics and the random term components through a gated recurrent unit or a volatility equation, and outputs a dynamic compensation coefficient. The dynamic compensation coefficient is a scalar value that corresponds one-to-one to the prediction time point and is used to quantify the instantaneous impact of abnormal equipment operating conditions on the load forecast value.

[0061] The basic load forecast value generated above is linearly superimposed with the dynamic compensation coefficient. The calculation formula is: initial forecast load value = basic load forecast value + dynamic compensation coefficient × fluctuation intensity factor; where the fluctuation intensity factor is preset as a fixed weight based on the historical statistical value of the equipment status fluctuation characteristics (such as standard deviation). For example, if the dynamic compensation coefficient is 0.8 and the fluctuation intensity factor is 10kW, the initial forecast value is increased by 8kW on the basic value. The initial forecast load value generated after superposition forms a preliminary forecast result by integrating the influence of long-term trends and short-term fluctuations. Its output curve includes the load fluctuation characteristics caused by sudden changes in equipment status.

[0062] Based on the timestamp sequence of equipment start and stop events, the initial predicted load curve is event-drivenly corrected. Specifically, when the timestamp of a detected equipment start and stop event (such as a transformer shutdown or line disconnection) overlaps with the prediction period, the predicted value at the corresponding time point is adjusted based on the event type (start or shutdown) and the load change pattern associated with the historical event (such as the percentage of instantaneous load drop caused by the shutdown event). For example, if there is a transformer shutdown event in the timestamp sequence, the predicted value for that period is corrected based on the preset event impact coefficient (such as a 15% load drop). The corrected intermediate prediction result reflects the direct impact of the equipment start and stop event on the load, and its output is in the form of a segmented adjusted load curve.

[0063] The intermediate prediction results are compared with real-time load data at each time point to calculate a residual sequence, expressed as: Residual sequence = Real-time collected value - Intermediate prediction result. Real-time load data is acquired through the power monitoring system at a preset collection frequency and strictly aligned with the timestamp of the prediction results. The residual sequence is used to quantify the model prediction deviation and provide input for subsequent parameter updates.

[0064] Perform sliding window statistical analysis on the residual sequence and calculate the mean, variance and autocorrelation coefficient of the residuals in the window. The preset sliding window length is preset to a fixed time length according to the forecast requirements. When the statistic exceeds the preset threshold (such as the variance threshold is set to 1.5 times the historical residual variance), the model parameter update mechanism is triggered. Parameter update includes: adjusting the timing weight parameters of the trend capture module based on the residual mean, and adjusting the sensitivity coefficient of the volatility compensation module based on the residual variance. For example, when the residual mean is continuously positive, the weight of the trend capture module is reduced to suppress overfitting; when the residual variance increases suddenly, the sensitivity of the volatility compensation module is increased to enhance the response to abnormal fluctuations. Parameter update is achieved through the gradient descent algorithm to ensure the coordinated optimization of the trend capture and volatility compensation modules.

[0065] The updated time series weight parameters and sensitivity coefficients are loaded into the modified load model, and the intermediate prediction results are recalibrated. Specifically, the intermediate prediction results are input into the feedback optimization layer of the modified load model, and the parameters of the trend capture module and the fluctuation compensation module are iteratively optimized through the gradient descent algorithm until the statistics of the residual sequence regress to the preset tolerance range (such as the absolute value of the residual mean is less than 2%, and the variance is less than 1.2 times the historical mean). The optimized model outputs the target load forecast result, and its prediction curve simultaneously reflects the combined influence of long-term trends, equipment status fluctuations and real-time residual feedback, and ultimately generates high-precision load forecast data that is highly matched with real-time operating conditions.

[0066] In some instances, this also includes: Based on the target load forecast results, a probability distribution is generated through the Monte Carlo simulation algorithm; Calculate dynamic confidence intervals based on probability distributions; Based on the dynamic confidence interval and target load forecast results, a visualization chart is generated, which includes the target load forecast curve, the real-time collected load data curve and the confidence interval boundary.

[0067] Exemplarily, a probability distribution is generated by a Monte Carlo simulation algorithm based on the target load forecast result. In specific implementation, an error probability model is first constructed based on the statistical characteristics of the forecast error of the target load forecast result. The error probability model fits the mean and variance of the error by analyzing the residual distribution of historical forecast results and real-time collected load data, and assumes that the error obeys a Gaussian distribution or a non-parametric distribution based on kernel density estimation. During the Monte Carlo simulation process, a preset number (for example, 1000 times) of random perturbation samples are generated for each forecast time point in the target load forecast result. The perturbation samples are generated by random sampling from the error probability model and superimposing them with the original forecast value. Through this method, a set of perturbation samples covering all time points in the forecast period is generated to form a probability distribution cloud map of the forecast result. The probability distribution characterizes the uncertainty range of the target load forecast result.

[0068] Statistical analysis is performed on the probability distribution set of predicted values ​​generated by the Monte Carlo simulation to calculate the dynamic confidence interval. Specifically, according to the preset confidence level (for example, 95%), the upper and lower quantiles of the probability distribution are determined as the confidence interval boundaries. For example, for each prediction time point, the 2.5% quantile is extracted as the lower limit and the 97.5% quantile is extracted as the upper limit from the set of predicted values ​​corresponding to that time point to generate a dynamic confidence interval curve that changes over time. The calculation of the confidence interval boundaries is achieved through the quantile regression algorithm to ensure that the interval range is dynamically adjusted as the prediction time progresses, accurately reflecting the reliability differences of the prediction results in different time periods. During the calculation process, the width of the confidence interval is determined by the variance and distribution shape of the prediction error. When the error variance increases or the distribution presents a heavy-tailed feature, the confidence interval automatically expands to cover a larger range of uncertainty. The dynamic nature of the confidence interval is reflected in two aspects. First, based on the residual sequence between the real-time collected load data and the prediction results, the parameters of the error probability model are dynamically updated so that the width of the confidence interval can be adaptively adjusted as the prediction error changes. Second, when a sudden change in equipment status or abnormal environmental parameters is detected, the confidence interval's coverage of sudden operating conditions is improved by increasing the number of sampling times or adjusting the distribution type.

[0069] The target load forecast curve, real-time load data curve, and dynamic confidence interval boundaries are integrated into the same visualization chart. Specifically, the following operations are performed through the visualization rendering engine: the target load forecast curve is drawn with a first line type (such as a solid line) to reflect the final forecast result; the real-time load data curve is drawn with a second line type (such as a dotted line) to achieve an intuitive comparison between the forecast value and the actual value; the confidence interval is annotated, and a semi-transparent color domain is filled around the forecast curve, and the upper and lower boundaries of the color domain correspond to the upper and lower limit curves of the dynamic confidence interval respectively; a time axis zoom control is integrated into the chart, allowing users to adjust the time range by dragging or scrolling; and a manual feedback input window is added to support users to submit correction suggestions for abnormal intervals (such as actual values ​​that continue to exceed the confidence interval). The sampling strategy or confidence level parameters of the Monte Carlo simulation are dynamically adjusted based on the feedback instructions to achieve controllable optimization of the forecast results.

[0070] See also Figure 2 , which is a schematic structural diagram of a daily power fitting device for a power marketing management information platform provided in an embodiment of the present application, including: The power data acquisition unit 21 is used to acquire historical load data, external environment data and equipment operation status data of the target area; An initial model building unit 22 builds an initial load fitting model based on historical load data; The load model correction unit 23 is used to extract date characteristic factors according to a preset time classification rule, dynamically correct the initial load fitting model based on the date characteristic factors, and generate a corrected load model; The multi-source feature fusion unit 24 generates multi-source fusion features based on the external environment data and the device operation status data; The load prediction output unit 25 is used to input the multi-source fusion features into the modified load model and output the target load prediction result.

[0071] See also Figure 3 An embodiment of the present application also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, the steps of any method of daily power fitting of the power marketing management information platform are implemented.

[0072] Since the electronic device introduced in this embodiment is the device used to implement a daily power fitting device of an electric power marketing management information platform in the embodiment of this application, based on the method introduced in the embodiment of this application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of this application will not be introduced in detail here. As long as the equipment used by technical personnel in this field to implement the method in the embodiment of this application falls within the scope of protection of this application.

[0073] During the specific implementation process, when the computer program 311 is executed by the processor, any implementation method of the embodiments corresponding to the first aspect can be implemented.

[0074] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0075] Those skilled in the art will appreciate that the embodiments of the present application may provide methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-readable program code.

[0076] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0077] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0079] The present application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device executes Figure 1 The process of a daily power fitting method of a power marketing management information platform in the corresponding embodiment.

[0080] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium can be a magnetic medium, an optical medium or a semiconductor medium, etc.

[0081] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0083] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0084] In addition, the functional units in the various embodiments of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The above-mentioned integrated units may be implemented in the form of hardware and / or software functional units.

[0085] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

[0086] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

[0087] Although the preferred embodiments of this specification have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of this specification.

[0088] Obviously, those skilled in the art may make various changes and modifications to this specification without departing from the spirit and scope of this specification. Thus, if such changes and modifications fall within the scope of the claims of this specification and their equivalents, this specification is intended to include such changes and modifications.

Claims

1. A daily power fitting method for a power marketing management information platform, characterized in that: include: Obtain historical load data, external environment data, and equipment operating status data for the target area; constructing an initial load fitting model based on the historical load data; Extracting date characteristic factors according to preset time classification rules, dynamically revising the initial load fitting model based on the date characteristic factors, and generating a revised load model; generating a multi-source fusion feature based on the external environment data and the device operation status data; The multi-source fusion features are input into the modified load model, and a target load prediction result is output.

2. The method according to claim 1, characterized in that The acquisition of historical load data, external environment data, and equipment operating status data of the target area includes: Based on a preset collection frequency, extracting load time series data within a first preset time range of the target area from the power monitoring system to generate historical load data; Based on the geographical location distribution of meteorological monitoring stations, obtaining external environmental data aligned with the timestamp of the historical load data, the external environmental data including weather type, temperature sequence, and humidity sequence; Based on the industrial Internet of Things protocol, the operating status data of the power equipment in the target area is analyzed, and the operating status data includes transformer temperature, line load rate and equipment start and stop status.

3. The method according to claim 2, characterized in that The constructing of an initial load fitting model based on the historical load data includes: Based on the time series of historical load data, the load fluctuation amplitude parameters within the preset time window are extracted; Determining a separation threshold between a trend item component and a random item component of the historical load data based on the load fluctuation amplitude parameter; Decomposing the historical load data into components based on the separation threshold to generate trend item component data and random item component data; Selecting a candidate model structure from a preset model library based on the fitting residual of the trend item component data and the fluctuation characteristics of the random item component data; The trend item component data and the random item component data are jointly trained based on the candidate model structure to generate an initial load fitting model, wherein the initial load fitting model includes a trend capturing module and a fluctuation compensation module.

4. The method according to claim 3, characterized in that The extracting of date characteristic factors according to a preset time classification rule, and dynamically revising the initial load fitting model based on the date characteristic factors to generate a revised load model include: Based on a preset time classification rule, the timestamps of the historical load data are divided into a weekday type, a weekend type, and a holiday type, wherein the preset time classification rule includes a date tag matching algorithm and a statutory holiday database; Extracting date characteristic factors based on the historical load data corresponding to each date type after classification, wherein the date characteristic factors include the load fluctuation baseline on weekdays, the load peak period on weekends, and the load decay rate on holidays; Determining a timing weight parameter of the trend capture module and a sensitivity coefficient of the fluctuation compensation module based on the date characteristic factor; recalibrating the long-term trend fitting function of the trend capture module and the random fluctuation compensation function of the fluctuation compensation module based on the time series weight parameter and the sensitivity coefficient; The initial load fitting model is updated based on the recalibrated parameters to generate the revised load model, wherein the output result of the revised load model includes a load forecast curve matching the date type.

5. The method according to claim 4, characterized in that The generating of multi-source fusion features based on the external environment data and the device operation status data includes: Based on the weather type, generating a weather feature vector through a classification coding algorithm; Based on the temperature sequence and the humidity sequence, generating standardized environmental parameters through normalization processing; Based on the transformer temperature and the line load rate, generating a device state fluctuation characteristic using a dynamic threshold algorithm, wherein the dynamic threshold algorithm calculates a real-time deviation rate based on a preset safe operation threshold range in historical device data; Based on the device start and stop status, generate a timestamp sequence of device start and stop events; constructing a multidimensional feature input matrix based on the weather feature vector, the standardized environmental parameters, the device state fluctuation characteristics, and the timestamp sequence of the device start and stop events; Based on the multidimensional feature input matrix, calculating a first association weight between the weather feature vector and the device state fluctuation feature, and a second association weight between the standardized environmental parameter and the timestamp sequence of the device start and stop events through a cross-attention mechanism; Based on the weighted fusion result of the first association weight and the second association weight, a multi-source fusion feature containing spatiotemporal correlation is generated.

6. The method according to claim 5, characterized in that The step of inputting the multi-source fusion features into the modified load model and outputting a target load prediction result includes: Based on the long-term trend fitting function, the spatiotemporal correlation in the multi-source fusion features is analyzed to generate a base load forecast value; Based on the random fluctuation compensation function, analyzing the device state fluctuation characteristics in the multi-source fusion characteristics to generate a dynamic compensation coefficient; The basic load prediction value and the dynamic compensation coefficient are linearly superimposed to generate the initial predicted load curve; Based on the timestamp sequence of the equipment start and stop events, the initial predicted load curve is modified by an event-driven method to generate an intermediate prediction result; Based on the modified load model, calculate the residual sequence between the intermediate prediction results and the real-time collected load data; Based on the sliding window statistics of the residual sequence, updating the time series weight parameters and sensitivity coefficients; Based on the updated time series weight parameter and the sensitivity coefficient, the intermediate prediction result is recalibrated to generate a target load prediction result.

7. The method according to claim 1, characterized in that Also includes: Based on the target load prediction result, generating a probability distribution through a Monte Carlo simulation algorithm; Calculating a dynamic confidence interval based on the probability distribution; Based on the dynamic confidence interval and the target load prediction result, a visualization chart is generated, where the visualization chart includes a target load prediction curve, a real-time collected load data curve, and a confidence interval boundary.

8. A daily power fitting device for a power marketing management information platform, characterized in that: include: Power data acquisition unit, used to obtain historical load data, external environment data and equipment operating status data of the target area; An initial model building unit, which builds an initial load fitting model based on the historical load data; A load model correction unit, configured to extract date characteristic factors according to a preset time classification rule, and dynamically correct the initial load fitting model based on the date characteristic factors to generate a corrected load model; a multi-source feature fusion unit, which generates a multi-source fusion feature based on the external environment data and the device operation status data; The load prediction output unit is used to input the multi-source fusion features into the modified load model and output the target load prediction result.

9. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to implement the steps of the daily power fitting method for the power marketing management information platform as described in any one of claims 1 to 7 when executing the computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the daily power fitting method of the power marketing management information platform according to any one of claims 1 to 7 is implemented.

Images