Power grid transaction price prediction method for load fluctuation caused by high temperature and humidity

Abstract

The present application relates to the technical field of data processing, and more particularly to a power grid transaction price prediction method for load fluctuation caused by high temperature and humidity, comprising: collecting comprehensive operation data of the power grid, preprocessing to obtain optimized comprehensive operation data, extracting features from the optimized comprehensive operation data to obtain comprehensive operation features, and fusing to obtain fused operation features; constructing a load prediction model based on the comprehensive operation features to obtain a load prediction result; training an initial model based on historical load data and the fused operation features to obtain a deep learning model, and obtaining a power grid transaction price prediction result; obtaining a prediction representation value to determine the prediction robustness of the power grid transaction price prediction result; obtaining an effective information loss rate of the fused operation features to determine the correlation loss repair gain of the deep learning model; and obtaining an outlier rejection rate of the optimized comprehensive operation data to determine a humidity mutation sharpening coefficient of the comprehensive operation features. The present application improves the prediction robustness of power grid transaction price prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity. Background Technology

[0002] In existing technologies, power grid trading price forecasting primarily employs traditional load sensitivity analysis, linear regression, and conventional power flow calculations, which are ill-suited to the actual operating conditions of significant load fluctuations under high-temperature and high-humidity environments. High temperature and humidity directly cause rapid increases in cooling load, load distortion, and short-term abrupt changes, resulting in a clear nonlinear correlation between load and electricity prices. Traditional linear models struggle to accurately reflect the true changing patterns of both. Most existing methods lack specific feature extraction and enhancement processing for the characteristics of sudden changes in humidity and temperature, leading to untimely identification and low accuracy in judging load changes caused by abrupt temperature and humidity fluctuations. During data preprocessing, existing technologies lack a quantitative evaluation mechanism for outlier removal and fail to control the loss of effective information, making them susceptible to the influence of abnormal data, resulting in incomplete feature information and increased model calculation bias. Under operating conditions of significant humidity and frequent load changes, the prediction results exhibit large fluctuations in error and poor overall stability, leading to insufficient robustness in power grid trading price forecasting.

[0003] Chinese Patent Publication No. CN103606091A discloses a method for real-time electricity price information interaction in a smart distribution network based on load sensitivity. The system includes: based on the operating status of the distribution network, user response characteristics are obtained through user characteristic analysis, and combined with the load distribution characteristics and load consumption characteristics, real-time electricity price information interaction is performed on the distribution network. The method includes the following steps: Step 1: Obtain network parameter information of the distribution system from the distribution automation system, including: branch number of the line, first-end node and last-end node numbers, transformer ratio and branch impedance; perform a traversal search of the distribution network, i.e., from the feeder node to the end... The first step involves searching all nodes, branches, or sub-branches, performing two-dimensional depth encoding on the feeder, and forming an array (fx, nx, dx) to represent a radial distribution network. Here, fx is the parent node number of node x, nx is the node number of node x, and dx is the depth (layer number) from node x to the root node. The second step involves obtaining the load forecast data for the next day from the load forecasting system, including user name, electricity consumption category, industry classification, load nature, and the active power curve for the next day (24 points per day). Electricity price information for various load types is also obtained from the marketing system. The third step involves calculating the electricity demand elasticity values ​​for various load types at different times based on the data obtained in the second step. Therefore, the aforementioned real-time electricity price information interaction method for smart distribution networks based on load sensitivity suffers from several problems. Because it relies solely on single load data and lacks data quality management, the load sensitivity method is prone to failure in high-temperature and humid environments, leading to insufficient robustness in power grid transaction price forecasting. Summary of the Invention

[0004] Therefore, this invention provides a method for predicting power grid transaction prices that addresses load fluctuations caused by high temperature and humidity. This method overcomes the problem in existing technologies where the load sensitivity method is prone to failure in high temperature and humidity scenarios due to reliance on single load data and lack of data quality management, resulting in insufficient robustness in power grid transaction price prediction.

[0005] To achieve the above objectives, the present invention provides a method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity, comprising: Collect comprehensive operation data of the power grid, and clean, denoise and remove outliers in sequence to obtain optimized comprehensive operation data. Extract features from the optimized comprehensive operation data to obtain comprehensive operation features, and fuse the comprehensive operation features to obtain fused operation features. A load forecasting model is constructed based on the comprehensive operational characteristics, and the load forecasting model is used to perform load forecasting on the comprehensive operational data to obtain load forecasting results. The initial model is trained based on historical load data and the integrated operation characteristics to obtain a deep learning model. The deep learning model is then used to predict the load forecast results and the integrated operation data to obtain the power grid transaction price forecast results. The accuracy of the power grid transaction price forecast and the error fluctuation range of the power grid transaction price forecast are obtained to determine the forecast characterization value, and the forecast robustness of the power grid transaction price forecast is determined based on the forecast characterization value. If the prediction robustness of the power grid transaction price prediction result does not meet the requirements, the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics. If the fusion effectiveness of the fused operating features does not meet the requirements, determine whether to increase the correlation loss of the deep learning model to repair the gain; If it is not necessary to increase the correlation loss repair gain of the deep learning model, then the outlier removal rate of the optimized comprehensive operating data is used to determine the damp-heat mutation sharpening coefficient of the comprehensive operating features.

[0006] Further, determining whether the forecast robustness of the power grid transaction price forecast results meets the requirements based on the predicted characteristic values ​​includes: The predicted characteristic value is determined by the ratio of the accuracy of the power grid transaction price prediction results to the error fluctuation range of the power grid transaction price prediction results. The predicted characterization value is compared with the preset predicted characterization value; If the predicted characteristic value is greater than or equal to the preset predicted characteristic value, the prediction robustness of the power grid transaction price prediction result is determined to meet the requirements. If the predicted characteristic value is less than the preset predicted characteristic value, it is determined that the prediction robustness of the power grid transaction price prediction result does not meet the requirements.

[0007] Furthermore, if the prediction robustness of the power grid transaction price prediction result does not meet the requirements, the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics.

[0008] Furthermore, the effectiveness of the fusion of operational features is determined based on the effective information loss rate, including: The effective information loss rate of the integrated operational features is compared with the preset first loss rate; If the effective information loss rate of the fused operation feature is less than or equal to the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to meet the requirements. If the effective information loss rate of the fused operation feature is greater than the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to be unsatisfactory.

[0009] Further, determine whether it is necessary to increase the correlation loss repair gain of the deep learning model, including: The effective information loss rate of the fused operation features is compared with the preset first loss rate and the preset second loss rate, respectively; If the effective information loss rate of the fused operating features is greater than the preset first loss rate and less than the preset second loss rate, it is determined that the association loss repair gain of the deep learning model needs to be increased. If the effective information loss rate of the fused operating features is greater than or equal to the preset second loss rate, then it is determined that there is no need to increase the correlation loss repair gain of the deep learning model.

[0010] Furthermore, the increase in the correlation loss repair gain of the deep learning model is determined by the difference between the effective information loss rate of the fused running features and the preset first loss rate.

[0011] Furthermore, if the effective information loss rate based on the fused operation characteristics is greater than or equal to the preset second loss rate, it is initially determined that the processing effectiveness of the outliers in the integrated operation data does not meet the requirements, and the processing effectiveness of the outliers in the integrated operation data is determined based on the optimized outlier removal rate of the integrated operation data.

[0012] Furthermore, the sharpening coefficient of the damp-heat abrupt change in the comprehensive operating characteristics is determined, including: Compare the outlier removal rate of the optimized comprehensive operation data with the preset removal rate; If the outlier removal rate of the optimized comprehensive operation data is greater than or equal to the preset removal rate, the effectiveness of the outlier processing of the comprehensive operation data is deemed to meet the requirements. If the outlier removal rate of the optimized integrated operation data is less than the preset removal rate, it is determined that the effectiveness of the outlier processing of the integrated operation data does not meet the requirements, and the damp heat abrupt change sharpening coefficient of the integrated operation characteristics is increased.

[0013] Furthermore, the outlier removal rate of the optimized integrated operating data is the ratio of the number of outliers removed from the optimized integrated operating data to the total number of outliers in the optimized integrated operating data.

[0014] Furthermore, the increase in the damp-heat abrupt change sharpening coefficient of the comprehensive operating characteristics is determined by the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operating data.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The method of the present invention determines the prediction robustness of the power grid transaction price prediction result based on the prediction characterization value determined by the accuracy of the power grid transaction price prediction result and the error fluctuation range of the power grid transaction price prediction result. Due to insufficient correlation of the fused operation features, the deep learning model cannot fully explore the inherent coupling law between hot and humid weather, load fluctuation and electricity price, resulting in underfitting of the deep learning model. By judging the prediction robustness of the power grid transaction price prediction result, the underfitting problem of the deep learning model can be effectively identified, intuitively reflecting the feature fusion quality and model learning effect, and providing a basis for subsequent feature correlation correction and optimization of relevant model parameters. The correlation loss repair gain of the deep learning model is adjusted according to the effective information loss rate of the fused operation features. Due to insufficient outlier removal, noise and distortion interference in the comprehensive operation data cannot be effectively filtered out, resulting in the effective information of the fused operation features being masked, such as temperature and humidity, load and electricity price. The correlation between features is weakened, resulting in insufficient correlation of fused operational features. By increasing the correlation loss repair gain of the deep learning model, the repair and reconstruction of the correlation between features can be strengthened, the effective information obscured by noise and distortion interference can be restored, and the feature correlation between fused operational features can be improved. The damp heat mutation sharpening coefficient of the comprehensive operational features is adjusted according to the outlier removal rate of the optimized comprehensive operational data. Due to the long temperature and humidity sampling interval, it is impossible to capture data of short-term extreme fluctuations, resulting in insufficient time granularity of high temperature and damp heat data, which cannot effectively identify real outliers in rapid fluctuation sections, thus leading to insufficient outlier removal. By increasing the damp heat mutation sharpening coefficient of the comprehensive operational features, the expression effect of instantaneous fluctuation and mutation features under high temperature and damp heat environment can be strengthened, the feature significance of weak and short-term extreme fluctuations can be enhanced, the feature loss problem caused by insufficient time granularity can be compensated, the ability to identify real outliers in rapid fluctuation sections can be improved, and the prediction robustness of power grid transaction price prediction can be improved.

[0016] Furthermore, the method described in this invention determines the robustness of the power grid transaction price prediction results by setting preset prediction characterization values. Due to insufficient correlation of fused operating features, the deep learning model cannot fully explore the inherent coupling law between hot and humid weather, load fluctuations, and electricity prices, resulting in underfitting of the deep learning model. By determining the robustness of the power grid transaction price prediction results, the underfitting problem of the deep learning model can be effectively identified, intuitively reflecting the feature fusion quality and model learning effect, providing a basis for subsequent feature correlation correction and optimization of relevant model parameters, and further improving the robustness of power grid transaction price prediction.

[0017] Furthermore, the method of the present invention adjusts the correlation loss repair gain of the deep learning model by setting a preset first loss rate and a preset second loss rate. Due to insufficient outlier removal, noise and distortion interference in the integrated operation data cannot be effectively filtered out, resulting in the effective information of the fused operation features being obscured and the correlation between features such as temperature and humidity, load and electricity price being weakened, thus causing insufficient correlation of the fused operation features. By increasing the correlation loss repair gain of the deep learning model, the repair and reconstruction of the correlation between features can be strengthened, the effective information obscured by noise and distortion interference can be restored, the feature correlation between the fused operation features can be improved, and the predictive robustness of power grid transaction price prediction can be further improved.

[0018] Furthermore, the method described in this invention adjusts the damp-heat abrupt change sharpening coefficient of the comprehensive operating characteristics by setting a preset rejection rate. Due to the excessively long temperature and humidity sampling interval, it is impossible to capture data on short-term extreme fluctuations, resulting in insufficient time granularity of high-temperature and damp-heat data. This makes it impossible to effectively identify real anomalies in rapidly fluctuating segments, leading to insufficient outlier removal. By increasing the damp-heat abrupt change sharpening coefficient of the comprehensive operating characteristics, the expression effect of instantaneous fluctuations and abrupt changes under high-temperature and damp-heat conditions can be strengthened, the feature significance of weak and short-term extreme fluctuations can be enhanced, the feature loss problem caused by insufficient time granularity can be compensated for, the ability to identify real anomalies in rapidly fluctuating segments can be improved, and the predictive robustness of power grid transaction price forecasting can be further improved. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the power grid transaction price prediction method for load fluctuations caused by high temperature and humidity, according to an embodiment of the present invention. Figure 2 This is a logical flowchart of the process for determining the robustness of power grid transaction price prediction results in an embodiment of the present invention for a power grid transaction price prediction method for load fluctuations caused by high temperature and humidity. Figure 3 This is a flowchart illustrating the correlation loss repair and gain process of a deep learning model for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity, as described in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the process of determining the damp heat abrupt change sharpening coefficient for the comprehensive operating characteristics of a power grid transaction price prediction method for load fluctuations caused by high temperature and humidity, as described in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0021] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0022] Please see Figure 1 The diagram shown is an overall flowchart of the power grid transaction price prediction method for load fluctuations caused by high temperature and humidity, according to an embodiment of the present invention.

[0023] This invention provides a method for predicting power grid transaction prices to address load fluctuations caused by high temperature and humidity, comprising: Step S1: Collect comprehensive operation data of the power grid, and clean, denoise and remove outliers from the comprehensive operation data in sequence to obtain optimized comprehensive operation data. Extract features from the optimized comprehensive operation data to obtain comprehensive operation features, and fuse the comprehensive operation features to obtain fused operation features. Step S2: Construct a load forecasting model based on the comprehensive operating characteristics, and use the load forecasting model to perform load forecasting on the comprehensive operating data to obtain load forecasting results; Step S3: Train the initial model based on historical load data and the integrated operation characteristics to obtain a deep learning model, and use the deep learning model to predict the load forecast results and the integrated operation data to obtain the power grid transaction price forecast results. Step S4: Obtain the accuracy of the power grid transaction price prediction result and the error fluctuation range of the power grid transaction price prediction result to determine the prediction characterization value, and determine whether the prediction robustness of the power grid transaction price prediction result meets the requirements based on the prediction characterization value. Step S5: If the prediction robustness of the power grid transaction price prediction result does not meet the requirements, then the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics. Step S6: If the fusion effectiveness of the fused running features does not meet the requirements, determine whether to increase the correlation loss repair gain of the deep learning model. Step S7: If it is not necessary to increase the correlation loss repair gain of the deep learning model, then obtain the outlier removal rate of the optimized comprehensive operating data to determine the damp heat mutation sharpening coefficient of the comprehensive operating features.

[0024] Specifically, the comprehensive operational data includes the relative humidity of the power grid operating area, regional load, and historical electricity price series.

[0025] Specifically, optimizing the comprehensive operation data includes the relative humidity of the power grid operating area after cleaning, the load of different regions after noise reduction, and the historical electricity price series after outlier removal.

[0026] Specifically, the comprehensive operating characteristics include load change rate, mean ambient temperature, and variance of ambient humidity.

[0027] Specifically, the characteristics of integrated operation include humidity-load coupling characteristics, load-electricity price correlation characteristics, and humidity-electricity price coupling characteristics.

[0028] Specifically, the load forecasting model can be a long short-term memory network model, a Transformer model, or a convolutional neural network model, with the preferred embodiment being a long short-term memory network model.

[0029] Specifically, the load forecast results include daily average load, peak and valley period indicators, and regional load peak and valley differences.

[0030] Specifically, the initial model is an initial framework that has basic time-series feature extraction capabilities and electricity price pattern feature adaptation capabilities, and is applicable to the analysis of comprehensive power grid operation data and the prediction of transaction prices in high-temperature and humid scenarios.

[0031] Specifically, the deep learning model can be a long short-term memory network model, a Transformer model, or a convolutional neural network model, with a preferred embodiment being a long short-term memory network model.

[0032] Specifically, the process of training an initial model based on historical load data and integrated operation characteristics to obtain a deep learning model involves dividing the historical load data and integrated operation characteristics into a training set and a test set. The initial model is trained using the training set, and the parameters are optimized and adjusted using the validation set. Finally, the quality of the generated model is evaluated by the test set, resulting in a deep learning model that meets the requirements for predicting power grid transaction prices.

[0033] Specifically, the grid transaction price forecast results include the day-ahead market price, the predicted peak price, and the average price.

[0034] Specifically, the process of using a deep learning model to predict load forecasting results and comprehensive operation data to obtain grid transaction price prediction results involves fusing load forecasting results and comprehensive operation data, performing data correlation mining and electricity price trend fitting based on a deep learning model, and finally outputting grid transaction price prediction results.

[0035] Specifically, the accuracy rate of the power grid transaction price forecast is the ratio of the amount of accurately predicted power grid transaction price data to the total amount of predicted power grid transaction price data.

[0036] Specifically, accuracy means that the predicted electricity price for power grid transactions is consistent with the actual transaction price under actual operating conditions.

[0037] Specifically, the robustness of the power grid transaction price prediction results means that under non-ideal operating conditions such as high temperature and humidity disturbances, the prediction results will not fluctuate significantly, and the deep learning model can still output reasonable and reliable power grid transaction price prediction results.

[0038] Specifically, the effective information loss rate of the fused operational features is the ratio of the number of lost effective information in the fused operational features to the total amount of information in the fused operational features.

[0039] Specifically, effective information refers to information that can accurately reflect the inherent correlation between meteorological fluctuations, load changes, and electricity prices, and that contributes to model predictions.

[0040] Specifically, the correlation loss repair gain of deep learning models is used to strengthen the repair and reconstruction of the correlation between various dimensions in the fused running features, and restore the effective information that is obscured by noise and distortion interference.

[0041] Specifically, outliers are data in the collected comprehensive operational data that deviate from the normal data distribution range and do not have a true operational pattern.

[0042] Specifically, the damp-heat abrupt change sharpening coefficient of comprehensive operating characteristics is used to enhance the amplitude of instantaneous fluctuations and abrupt change characteristics under high temperature and damp-heat conditions.

[0043] In implementation, the method of this invention determines the robustness of the power grid transaction price prediction results based on the prediction characterization value determined by the accuracy and error fluctuation range of the power grid transaction price prediction results. Due to insufficient correlation of the fused operational features, the deep learning model cannot fully explore the inherent coupling law between hot and humid weather, load fluctuation, and electricity price, resulting in underfitting of the deep learning model. By determining the robustness of the power grid transaction price prediction results, the underfitting problem of the deep learning model can be effectively identified, intuitively reflecting the feature fusion quality and model learning effect, and providing a basis for subsequent feature correlation correction and optimization of relevant model parameters. The correlation loss repair gain of the deep learning model is adjusted according to the effective information loss rate of the fused operational features. Due to insufficient outlier removal, noise and distortion interference in the comprehensive operational data cannot be effectively filtered out, resulting in the effective information of the fused operational features being masked, and the correlation law between features such as temperature and humidity, load, and electricity price being obscured. The weakening of correlation between features leads to insufficient correlation in the integrated operation features. By increasing the correlation loss repair gain of the deep learning model, the repair and reconstruction of the correlation between features can be strengthened, the effective information obscured by noise and distortion interference can be restored, and the feature correlation between integrated operation features can be improved. The damp heat mutation sharpening coefficient of the integrated operation features is adjusted according to the outlier removal rate of the optimized integrated operation data. Due to the excessively long temperature and humidity sampling interval, it is impossible to capture data on short-term extreme fluctuations, resulting in insufficient time granularity of high temperature and humidity data. This makes it impossible to effectively identify real outliers in the rapid fluctuation section, leading to insufficient outlier removal. By increasing the damp heat mutation sharpening coefficient of the integrated operation features, the expression effect of instantaneous fluctuation and mutation features under high temperature and humidity conditions can be strengthened, the feature significance of weak and short-term extreme fluctuations can be enhanced, the feature loss problem caused by insufficient time granularity can be compensated, the ability to identify real outliers in the rapid fluctuation section can be improved, and the predictive robustness of power grid transaction price forecasting can be improved.

[0044] Please continue reading. Figure 2 As shown, it is a logical flowchart of the process of determining the prediction robustness of the power grid transaction price prediction result in the power grid transaction price prediction method for load fluctuations caused by high temperature and humidity according to an embodiment of the present invention.

[0045] Specifically, determining whether the forecast robustness of the power grid transaction price forecast results meets the requirements based on the predicted characteristic values ​​includes: The predicted characteristic value is determined by the ratio of the accuracy of the power grid transaction price prediction results to the error fluctuation range of the power grid transaction price prediction results. The predicted characterization value is compared with the preset predicted characterization value; If the predicted characteristic value is greater than or equal to the preset predicted characteristic value, the prediction robustness of the power grid transaction price prediction result is determined to meet the requirements. If the predicted characteristic value is less than the preset predicted characteristic value, it is determined that the prediction robustness of the power grid transaction price prediction result does not meet the requirements.

[0046] Understandably, in the power grid transaction price forecasting method for load fluctuations caused by high temperature and humidity, the core logic of using a preset forecasting characterization value to characterize forecast robustness is to quantify the accuracy and stability of electricity price forecasting into a numerically calculable indicator. By comparing the forecasting characterization value obtained by the forecasting result accuracy rate and the forecasting result error fluctuation range with the preset forecasting characterization value, a scientific judgment can be made on the quality of integrated power grid operation data fusion, the effectiveness of feature correlation, and the generalization ability of the electricity price forecasting model. When the predicted characteristic value is greater than or equal to the preset predicted characteristic value, it indicates that the multi-source operation data registration is sufficient, the correlation law of comprehensive characteristics is clear, the coupling characteristics of hot and humid weather and load are prominent, the model prediction is stable and reliable, and the robustness of the power grid transaction price prediction results meets the requirements of power grid market-based trading and dispatch decision-making. When the predicted characteristic value is less than the preset predicted characteristic value, it indicates that the data noise and distortion interference are significant, the correlation of comprehensive operation characteristics is weak, the model's adaptability to fluctuating operating conditions is insufficient, the deviation and fluctuation risk of the electricity price prediction results are high, and it is impossible to achieve stable, accurate, and robust electricity price trend prediction. The prediction robustness does not meet the requirements, and the correlation loss repair gain of the deep learning model and the high-frequency fluctuation activation gain coefficient of the comprehensive operation characteristics need to be adaptively adjusted. The preset predicted characteristic value can be set according to the actual operating conditions. The setting of the preset predicted characteristic value aims to ensure the prediction robustness and practicality of the power grid transaction price prediction. Optionally, the preset predicted characteristic value is determined through a limited number of experiments by evaluating the prediction effect of different predicted characteristic values ​​on the power grid transaction price prediction. The determined preset predicted characteristic value should meet the requirement that it is neither too small nor will it cause excessive interference to the power grid transaction price prediction process. For example, the preset predicted characterization value is generally selected in the range of [5% / yuan, 10% / yuan].

[0047] Preferably, the preferred embodiment of the preset predicted characterization value is 7% / yuan.

[0048] Specifically, % / yuan is the unit of the predicted value, meaning a percentage per yuan.

[0049] In practice, the method of the present invention determines the predictive robustness of the power grid transaction price prediction results by setting preset prediction characterization values. Due to insufficient correlation of fused operating features, the deep learning model cannot fully explore the inherent coupling law between hot and humid weather, load fluctuations and electricity prices, resulting in underfitting of the deep learning model. By determining the predictive robustness of the power grid transaction price prediction results, the underfitting problem of the deep learning model can be effectively identified, intuitively reflecting the feature fusion quality and model learning effect, providing a basis for subsequent feature correlation correction and optimization of relevant model parameters, and further improving the predictive robustness of power grid transaction price prediction.

[0050] Please continue reading. Figure 3 As shown, it is a logical flowchart of the correlation loss repair gain process of the deep learning model for predicting power grid transaction prices caused by high temperature and humidity in an embodiment of the present invention.

[0051] Specifically, if the prediction robustness of the power grid transaction price prediction results does not meet the requirements, the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics.

[0052] Specifically, the determination of whether the fusion effectiveness of the fusion operation characteristics meets the requirements is based on the effective information loss rate of the fusion operation characteristics, including: The effective information loss rate of the integrated operational features is compared with the preset first loss rate; If the effective information loss rate of the fused operation feature is less than or equal to the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to meet the requirements. If the effective information loss rate of the fused operation feature is greater than the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to be unsatisfactory.

[0053] Specifically, determining whether it is necessary to increase the correlation loss repair gain of the deep learning model includes: The effective information loss rate of the fused operation features is compared with the preset first loss rate and the preset second loss rate, respectively; If the effective information loss rate of the fused operating features is greater than the preset first loss rate and less than the preset second loss rate, it is determined that the association loss repair gain of the deep learning model needs to be increased. If the effective information loss rate of the fused operating features is greater than or equal to the preset second loss rate, then it is determined that there is no need to increase the correlation loss repair gain of the deep learning model.

[0054] It is understandable that the preset first loss rate is lower than the preset second loss rate. The three intervals divided by the preset first loss rate and the preset second loss rate correspond to three different scenarios: The first interval is when the effective information loss rate of the fused operation feature is less than or equal to the preset first loss rate. The corresponding situation is: the fusion effectiveness of the fused operation feature is determined to meet the requirements. The second interval is when the effective information loss rate of the integrated operation features is greater than the preset first loss rate and less than the preset second loss rate. The corresponding situation is: due to insufficient outlier removal, noise and distortion interference in the integrated operation data cannot be effectively filtered out, resulting in the effective information of the integrated operation features being masked, the correlation between features such as temperature and humidity, load and electricity price being weakened, and thus the correlation of the integrated operation features being insufficient. The third interval is when the effective information loss rate of the fused operation features is greater than or equal to the preset second loss rate. The corresponding situation is: due to the excessively long temperature and humidity sampling interval, it is impossible to capture data of short-term extreme fluctuations, resulting in insufficient time granularity of high temperature and humidity data, making it impossible to effectively identify real anomalies in the rapid fluctuation range, thus leading to insufficient outlier removal.

[0055] Understandably, in power grid transaction price forecasting methods addressing load fluctuations caused by high temperature and humidity, the introduction of preset first and second loss rates to characterize the completeness of information about integrated operating characteristics is crucial. The core logic is to transform the fusion effectiveness of integrated operating characteristics into a quantifiable range of information loss rate values. The preset first loss rate serves as the boundary between acceptable fusion effectiveness and the need to adjust related loss repair gains, while the preset second loss rate is the critical threshold between the need to adjust related loss repair gains and severe outlier removal deficiencies. This provides a quantitative basis for specifically identifying problems such as feature fusion anomalies, information loss, and insufficient forecast robustness. The preset first and second loss rates can be set according to actual operating conditions. The setting of the preset first and second loss rates aims to ensure the forecast robustness and practicality of power grid transaction price forecasting. Optionally, the preset first and second loss rates are determined through a limited number of experiments by evaluating the forecasting effect of different information loss rates on power grid transaction price forecasting. The determined preset first and second loss rates should be neither too small nor excessively disruptive to the power grid transaction price forecasting process. For example, the preset first loss rate is generally selected in the range of [3%, 7%], and the preset second loss rate is generally selected in the range of [8%, 12%].

[0056] Preferably, the first loss rate is 5% in a preferred embodiment, and the second loss rate is 10% in a preferred embodiment.

[0057] Specifically, the increase in the correlation loss repair gain of the deep learning model is determined by the difference between the effective information loss rate of the fused running features and the preset first loss rate.

[0058] Specifically, when the difference between the effective information loss rate of the fused operating features and the preset first loss rate is within 4%, the association loss repair gain of the deep learning model increases to 1.1 times the original value. When the difference between the effective information loss rate of the fused operating features and the preset first loss rate exceeds 4%, in addition to increasing to 1.1 times the original value, for every 1% exceeding 4%, the association loss repair gain of the deep learning model increases by 0.015. For example, when the difference between the effective information loss rate of the fused operating features and the preset first loss rate is 6%, the current association loss repair gain of the deep learning model is 0.3, and the increased association loss repair gain of the deep learning model is 0.3×1.1+0.015×2=0.36.

[0059] In practice, the method of the present invention adjusts the correlation loss repair gain of the deep learning model by setting a preset first loss rate and a preset second loss rate. Due to insufficient outlier removal, noise and distortion interference in the integrated operation data cannot be effectively filtered out, resulting in the effective information of the fused operation features being obscured and the correlation between features such as temperature and humidity, load and electricity price being weakened, thus causing insufficient correlation of the fused operation features. By increasing the correlation loss repair gain of the deep learning model, the repair and reconstruction of the correlation between features can be strengthened, the effective information obscured by noise and distortion interference can be restored, the feature correlation between the fused operation features can be improved, and the predictive robustness of power grid transaction price prediction can be further improved.

[0060] Please continue reading. Figure 4 As shown, it is a logical flowchart of the process of determining the damp heat abrupt change sharpening coefficient of the comprehensive operating characteristics in the power grid transaction price prediction method for load fluctuations caused by high temperature and humidity according to an embodiment of the present invention.

[0061] Specifically, if the effective information loss rate based on the fusion operation characteristics is greater than or equal to the preset second loss rate, it is initially determined that the processing effectiveness of the outliers in the integrated operation data does not meet the requirements, and the processing effectiveness of the outliers in the integrated operation data is determined based on the optimized outlier removal rate of the integrated operation data.

[0062] Specifically, determining the sharpening factor for abrupt changes in damp heat in comprehensive operational characteristics includes: Compare the outlier removal rate of the optimized comprehensive operation data with the preset removal rate; If the outlier removal rate of the optimized comprehensive operation data is greater than or equal to the preset removal rate, the effectiveness of the outlier processing of the comprehensive operation data is deemed to meet the requirements. If the outlier removal rate of the optimized integrated operation data is less than the preset removal rate, it is determined that the effectiveness of the outlier processing of the integrated operation data does not meet the requirements, and the damp heat abrupt change sharpening coefficient of the integrated operation characteristics is increased.

[0063] It is understandable that the two intervals defined by the preset rejection rate correspond to two different scenarios: The first interval is when the outlier removal rate of the optimized comprehensive operation data is greater than or equal to the preset removal rate. The corresponding situation is that the effectiveness of the processing of outliers in the comprehensive operation data meets the requirements. The second interval is when the outlier removal rate of the optimized comprehensive operation data is less than the preset removal rate. The corresponding situation is that the temperature and humidity sampling interval is too long, which makes it impossible to capture short-term extreme fluctuation data. This results in insufficient time granularity of high temperature and humidity data, making it impossible to effectively identify real outliers in the rapid fluctuation range, thus leading to insufficient outlier removal.

[0064] Understandably, using a preset rejection rate to characterize the effectiveness of outlier handling in comprehensive operational data involves transforming the effects of data cleaning and outlier removal into a quantifiable and comparable outlier rejection rate. By comparing the optimized outlier rejection rate of comprehensive operational data with the preset rejection rate, the accuracy of anomaly identification and the completeness of feature retention in the current data preprocessing stage are determined. This provides crucial quantitative evidence for subsequent enhancement of humid and hot abrupt changes, repair of effective information, compensation for associated losses, and improvement of the accuracy of power grid transaction price prediction. Optimizing the outlier rejection rate of comprehensive operational data is a key evaluation indicator in the data preprocessing stage. An excessively high rejection rate can directly lead to the accidental deletion of effective operational information, data distribution distortion, and distortion of feature patterns, resulting in problems such as missing fused feature information, increased effective information loss rate, and increased model fitting deviation. Especially in high-temperature and humid load fluctuation scenarios, outlier data is easily confused with real abrupt change data. The rationality of outlier removal directly determines the accuracy and stability of subsequent feature fusion and price prediction. The preset rejection rate can be set according to actual operating conditions. The preset rejection rate is designed to ensure the robustness and practicality of power grid transaction price prediction. Optionally, the preset rejection rate is determined through a limited number of experiments by evaluating the predictive effect of different outlier rejection rates of optimized integrated operation data on the prediction of power grid transaction prices. The determined preset rejection rate should be neither too small nor cause excessive interference to the prediction process of power grid transaction prices. For example, the preset rejection rate is generally selected in the range of [3%, 8%].

[0065] Preferably, the preset rejection rate is 5% in this preferred embodiment.

[0066] Specifically, the outlier removal rate of the optimized integrated operating data is the ratio of the number of outliers removed from the optimized integrated operating data to the total number of outliers in the optimized integrated operating data.

[0067] Specifically, the increase in the damp-heat abrupt change sharpening coefficient of the comprehensive operating characteristics is determined by the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operating data.

[0068] Specifically, when the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operation data is within 3%, the damp-heat mutation sharpening coefficient of the comprehensive operation feature increases to 1.05 times the original value. When the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operation data exceeds 3%, in addition to increasing to 1.05 times the original value, for every 1% exceeding 3%, the damp-heat mutation sharpening coefficient of the comprehensive operation feature increases by 0.018. For example, when the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operation data is 5%, the current damp-heat mutation sharpening coefficient of the comprehensive operation feature is 0.5, and the increased damp-heat mutation sharpening coefficient of the comprehensive operation feature is 0.5×1.05+0.018×2=0.561.

[0069] In practice, the method of this invention adjusts the sharpening coefficient of damp-heat abrupt changes in comprehensive operating characteristics by setting a preset rejection rate. Due to the excessively long temperature and humidity sampling interval, it is impossible to capture data on short-term extreme fluctuations, resulting in insufficient time granularity of high-temperature and damp-heat data. This makes it impossible to effectively identify real anomalies in rapidly fluctuating sections, leading to insufficient outlier removal. By increasing the sharpening coefficient of damp-heat abrupt changes in comprehensive operating characteristics, the expression effect of instantaneous fluctuations and abrupt changes under high-temperature and damp-heat conditions can be enhanced, the significance of weak and short-term extreme fluctuations can be increased, the feature loss problem caused by insufficient time granularity can be compensated for, the ability to identify real anomalies in rapidly fluctuating sections can be improved, and the predictive robustness of power grid transaction price forecasting can be further improved.

[0070] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for predicting power grid transaction prices to address load fluctuations caused by high temperature and humidity, characterized in that, include: Collect comprehensive operation data of the power grid, and clean, denoise and remove outliers in sequence to obtain optimized comprehensive operation data. Extract features from the optimized comprehensive operation data to obtain comprehensive operation features, and fuse the comprehensive operation features to obtain fused operation features. A load forecasting model is constructed based on the comprehensive operational characteristics, and the load forecasting model is used to perform load forecasting on the comprehensive operational data to obtain load forecasting results. The initial model is trained based on historical load data and the integrated operation characteristics to obtain a deep learning model. The deep learning model is then used to predict the load forecast results and the integrated operation data to obtain the power grid transaction price forecast results. The accuracy of the power grid transaction price forecast and the error fluctuation range of the power grid transaction price forecast are obtained to determine the forecast characterization value, and the forecast robustness of the power grid transaction price forecast is determined based on the forecast characterization value. If the prediction robustness of the power grid transaction price prediction result does not meet the requirements, the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics. If the fusion effectiveness of the fused operating features does not meet the requirements, determine whether to increase the correlation loss of the deep learning model to repair the gain; If it is not necessary to increase the correlation loss repair gain of the deep learning model, then the outlier removal rate of the optimized comprehensive operating data is used to determine the damp-heat mutation sharpening coefficient of the comprehensive operating features.

2. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 1, characterized in that, Determining whether the robustness of the power grid transaction price prediction results meets the requirements based on the predicted characteristic values ​​includes: The predicted characteristic value is determined by the ratio of the accuracy of the power grid transaction price prediction results to the error fluctuation range of the power grid transaction price prediction results. The predicted characterization value is compared with the preset predicted characterization value; If the predicted characteristic value is greater than or equal to the preset predicted characteristic value, the prediction robustness of the power grid transaction price prediction result is determined to meet the requirements. If the predicted characteristic value is less than the preset predicted characteristic value, it is determined that the prediction robustness of the power grid transaction price prediction result does not meet the requirements.

3. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 2, characterized in that, If the prediction robustness of the power grid transaction price prediction results does not meet the requirements, the fusion effectiveness of the fusion operation characteristics is determined based on the effective information loss rate of the fusion operation characteristics.

4. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 3, characterized in that, The effectiveness of fusion based on the effective information loss rate of fusion operation characteristics is determined, including: The effective information loss rate of the integrated operational features is compared with the preset first loss rate; If the effective information loss rate of the fused operation feature is less than or equal to the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to meet the requirements. If the effective information loss rate of the fused operation feature is greater than the preset first loss rate, the fusion effectiveness of the fused operation feature is determined to be unsatisfactory.

5. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 4, characterized in that, Determine whether the correlation loss repair gain of the deep learning model needs to be increased, including: The effective information loss rate of the fused operation features is compared with the preset first loss rate and the preset second loss rate, respectively; If the effective information loss rate of the fused operating features is greater than the preset first loss rate and less than the preset second loss rate, it is determined that the association loss repair gain of the deep learning model needs to be increased. If the effective information loss rate of the fused operating features is greater than or equal to the preset second loss rate, then it is determined that there is no need to increase the correlation loss repair gain of the deep learning model.

6. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 5, characterized in that, The increase in the correlation loss repair gain of the deep learning model is determined by the difference between the effective information loss rate of the fused running features and the preset first loss rate.

7. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 6, characterized in that, If the effective information loss rate of the fused operation features is greater than or equal to the preset second loss rate, it is initially determined that the processing effectiveness of the outliers in the integrated operation data does not meet the requirements, and the processing effectiveness of the outliers in the integrated operation data is determined based on the optimized outlier removal rate of the integrated operation data.

8. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 7, characterized in that, Determine the sharpening factor for abrupt changes in humidity and heat in comprehensive operating characteristics, including: Compare the outlier removal rate of the optimized comprehensive operation data with the preset removal rate; If the outlier removal rate of the optimized comprehensive operation data is greater than or equal to the preset removal rate, the effectiveness of the outlier processing of the comprehensive operation data is deemed to meet the requirements. If the outlier removal rate of the optimized integrated operation data is less than the preset removal rate, it is determined that the effectiveness of the outlier processing of the integrated operation data does not meet the requirements, and the damp heat abrupt change sharpening coefficient of the integrated operation characteristics is increased.

9. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity as described in claim 8, characterized in that, The outlier removal rate of the optimized integrated operation data is the ratio of the number of outliers removed from the optimized integrated operation data to the total number of outliers in the optimized integrated operation data.

10. The method for predicting power grid transaction prices for load fluctuations caused by high temperature and humidity according to claim 9, characterized in that, The increase in the damp-heat abrupt change sharpening coefficient of the comprehensive operating characteristics is determined by the difference between the preset rejection rate and the outlier rejection rate of the optimized comprehensive operating data.

Images