Intelligent electricity utilization information management system

Abstract

The invention provides an intelligent electricity consumption information management system which obtains and predicts power supply related emergencies in real time through a data integration unit, a feature engineering unit preprocesses and extracts power data features, and a multi-model fusion unit selects a prediction model combination for prediction according to requirements. The adaptive learning unit constructs or optimizes a prediction model through deep learning, and the reinforcement learning unit carries out continuous training on the model, thereby achieving the precise prediction and supply optimization of power demands.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity consumption behavior visualization, and in particular to a smart electricity consumption information management system. Background Art

[0002] With the development of society and the advancement of science and technology, the power system plays an increasingly important role in people's lives. The continuous growth of electricity demand and the volatility of electricity supply have brought many challenges to the safe and stable operation of the power system. The traditional power information management system mainly relies on manual experience for power dispatching and forecasting, which has certain limitations, such as low prediction accuracy and insufficient ability to respond to emergencies. In order to solve these problems, in recent years, researchers have proposed a smart electricity information management system based on big data and artificial intelligence.

[0003] Although the existing smart electricity information management system has improved the accuracy of power forecasting and the ability to respond to emergencies to a certain extent, it still has the following problems:

[0004] Insufficient data integration capabilities: When acquiring external data, existing systems can often only process data from a single source, making it difficult to achieve real-time integration of multi-source data, resulting in inaccurate prediction results.

[0005] Single feature engineering processing method: Existing systems often use fixed feature extraction methods when processing power data, which makes it difficult to adapt to prediction needs in different scenarios.

[0006] Single prediction model: Existing systems usually use a single prediction model for power forecasting, which is difficult to cope with complex power system environments.

[0007] Insufficient model adaptability: When facing environmental changes, emergencies or nonlinear impacts, the existing system's prediction model has poor adaptability, which easily leads to increased prediction errors.

[0008] Insufficient model learning and updating capabilities: After obtaining new data samples, the existing system fails to effectively use these data to update and optimize the prediction model. Summary of the invention

[0009] The purpose of the present invention is to provide a smart electricity information management system to solve the above technical problems.

[0010] To achieve the above objectives, the present invention provides a smart electricity information management system, comprising:

[0011] A data integration unit, which is used to obtain external data from different sources in real time and predict emergencies related to power supply based on the external data;

[0012] A feature engineering unit, used for acquiring power data, preprocessing the power data to obtain a power data set, and extracting features from the power data set to obtain feature input for prediction;

[0013] A feature engineering unit, configured to select a matching prediction model combination from a model database according to a prediction requirement, perform prediction processing on a feature input for prediction, and obtain a prediction result; the prediction model combination includes at least one prediction model, and the prediction requirement includes a power demand prediction and a power supply prediction;

[0014] An adaptive learning unit, used to construct a prediction model for prediction demand through deep learning or select a prediction model to be adaptively learned from the model database, and add the constructed or selected prediction model to the model database after adaptive learning training based on environmental changes, emergencies or nonlinear effects;

[0015] The reinforcement learning unit is used to obtain incremental data samples and perform reinforcement learning training on multiple prediction models in the model database based on the incremental data samples.

[0016] Preferably, the data integration unit comprises:

[0017] A real-time data acquisition module is used to acquire external data from different sources in real time; the external data from different sources include meteorological data, market demand data and social and economic activity data;

[0018] The time prediction module is used to analyze the external data based on a time-driven prediction method to predict whether an emergency event will occur, and trigger an early warning mechanism to issue an early warning notification when an emergency event occurs.

[0019] Preferably, the feature engineering unit comprises:

[0020] A timeliness adjustment module is used to obtain power data within the most recent preset time period;

[0021] A data enhancement module, configured to generate synthetic data based on the power data acquired by the timeliness adjustment module, so as to increase data samples used for model training; the synthetic data is a data sample similar to the acquired power data generated by using a simulation model, or a data sample simulating an emergency or abnormal situation;

[0022] A preprocessing module is used to preprocess the power data obtained by the timeliness adjustment module for missing values, abnormal values ​​and duplicate data to obtain a power data set, and to extract features from the data of the power data set to obtain feature inputs for prediction; and is also used to extract features from the synthetic data generated by the data enhancement module to obtain feature inputs for training the model.

[0023] Preferably, the multi-model fusion unit comprises:

[0024] The selection module selects the best model combination based on the historical performance of the model and the data distribution of the current feature input;

[0025] A learning method fusion module, used for fusing the multiple prediction models of the best model combination by weighting or stacking;

[0026] The outlier processing module is used to automatically identify and correct anomalies or deviations in the prediction results of the prediction model.

[0027] The multi-model fusion unit comprises:

[0028] A deep neural network module, used to construct a prediction model for prediction demand through deep learning or to select a prediction model to be adaptively learned from the model database;

[0029] The adaptive algorithm module is used to perform adaptive learning and training on the constructed or selected prediction model based on environmental changes, emergencies or nonlinear effects, and then add it to the model database.

[0030] Preferably, it also includes:

[0031] The mobile terminal is used to log in, log out and operate the information management system.

[0032] Preferably, the mobile terminal is integrated with an archived work order data upload module and a query module, the archived work order data upload module is used to store and display archived work orders, and the query module is used to query the archived work orders.

[0033] Preferably, the query conditions of the query module are: power supply unit, work order number, user number, user name;

[0034] The query results output by the query module are: attachment details, work order number, user name, user number and electricity address. When the user clicks on the attachment details, a pop-up window displays the attachments of the business expansion archiving link under this work order, which corresponds to the attachment management of the business expansion archiving link in the marketing system.

[0035] The smart electricity information management system of the present invention has the following beneficial effects:

[0036] Improve forecast accuracy: The data integration unit acquires multi-source external data in real time, and the feature engineering unit is used to pre-process and extract features of power data, which can more comprehensively capture the factors affecting power demand and supply, thereby significantly improving the accuracy of forecasts.

[0037] Enhanced ability to respond to emergencies: The present invention can predict emergencies related to power supply based on external data, so that the power system can respond quickly when faced with emergencies and reduce or avoid the resulting losses.

[0038] Improve system adaptability: The multi-model fusion unit can select the appropriate model combination according to different prediction requirements, making the system more adaptable when facing complex and changing power markets and environmental conditions.

[0039] Optimize prediction models: The adaptive learning unit can perform adaptive learning training on the prediction model according to environmental changes, emergencies or nonlinear effects, ensuring that the model always maintains high prediction performance.

[0040] Continuously improve the prediction model: The reinforcement learning unit uses incremental data samples to perform reinforcement learning training on the prediction model in the model database, so that the model can continuously learn and adapt to new data features, thereby continuously improving the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0042] Figure 1 This is a framework diagram of a smart electricity information management system in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The detailed description of the accompanying drawings is intended as an illustration of the current embodiments of the present invention and is not intended to represent the only form in which the present invention can be implemented. It should be understood that the same or equivalent functions can be accomplished by different embodiments that are intended to be included in the spirit and scope of the present invention.

[0044] See also Figure 1 The embodiment of the present invention provides a smart electricity information management system, including:

[0045] A data integration unit 1, used for acquiring external data from different sources in real time and predicting emergencies related to power supply based on the external data;

[0046] Specifically, the main function of feature engineering unit 2 is to process raw power data and convert it into feature input that can be used for predictive analysis through preprocessing and feature extraction steps. The specific steps and functions are as follows:

[0047] Get power data:

[0048] Data Source: The feature engineering unit obtains real-time and historical power data from sources such as monitoring equipment, smart meters, and historical databases of the power system.

[0049] Data type: These data may include voltage, current, power, frequency, power consumption, load factor, etc.

[0050] Preprocessing power data:

[0051] Data cleaning: remove invalid data, outliers, noise, etc. to ensure data quality.

[0052] Data filling: Process missing data and fill in missing values ​​through interpolation, averaging, regression and other methods.

[0053] Data transformation: Converting data into a format suitable for analysis, such as converting date and time into a uniform format or encoding categorical data into numeric values.

[0054] Data normalization: Standardize or normalize the data so that they are at the same level to facilitate model processing.

[0055] Construct the electricity dataset:

[0056] Dataset formation: Organize the preprocessed data into a structured dataset, usually in a table format, where each row represents a sample and each column represents a feature.

[0057] Feature extraction:

[0058] Feature selection: Select features that have an important impact on the prediction target from the preprocessed data.

[0059] Feature construction: Create new features based on business knowledge and data analysis, such as extracting trends, periodicity, seasonality, and other features from time series data.

[0060] Feature transformation: Applying mathematical transformations (e.g., logarithmic transformation, square root transformation) to improve the distribution or relationship of features.

[0061] Get the feature input:

[0062] Final output: The dataset after feature engineering, i.e., feature input, will be provided as input to the subsequent prediction model for training and prediction.

[0063] For example, Feature Engineering Unit 2 might obtain hourly electricity usage data from smart meters. In the preprocessing stage, it cleans out abnormal readings due to meter failures and fills in missing data due to communication interruptions. It might then extract features such as peak electricity usage hours during the day and differences in electricity usage between weekdays and weekends, and use these features as input to the forecasting model to predict future electricity demand.

[0064] Through the processing of feature engineering unit 2, the smart electricity information management system can more accurately capture the key information in the power data, thereby improving the performance of the prediction model and the reliability of the prediction results. Feature engineering unit 2 is used to obtain power data, pre-process the power data to obtain a power data set, and extract features from the data of the power data set to obtain feature inputs for prediction;

[0065] Specifically, the main function of feature engineering unit 2 is to process raw power data and convert it into feature input that can be used for predictive analysis through preprocessing and feature extraction steps. The specific steps and functions are as follows:

[0066] Get power data:

[0067] Data Source: The feature engineering unit obtains real-time and historical power data from sources such as monitoring equipment, smart meters, and historical databases of the power system.

[0068] Data type: These data may include voltage, current, power, frequency, power consumption, load factor, etc.

[0069] Preprocessing power data:

[0070] Data cleaning: remove invalid data, outliers, noise, etc. to ensure data quality.

[0071] Data filling: Process missing data and fill in missing values ​​through interpolation, averaging, regression and other methods.

[0072] Data transformation: Converting data into a format suitable for analysis, such as converting date and time into a uniform format or encoding categorical data into numeric values.

[0073] Data normalization: Standardize or normalize the data so that they are at the same level to facilitate model processing.

[0074] Construct the electricity dataset:

[0075] Dataset formation: Organize the preprocessed data into a structured dataset, usually in a table format, where each row represents a sample and each column represents a feature.

[0076] Feature extraction:

[0077] Feature selection: Select features that have an important impact on the prediction target from the preprocessed data.

[0078] Feature construction: Create new features based on business knowledge and data analysis, such as extracting trends, periodicity, seasonality, and other features from time series data.

[0079] Feature transformation: Applying mathematical transformations (e.g., logarithmic transformation, square root transformation) to improve the distribution or relationship of features.

[0080] Get the feature input:

[0081] Final output: The dataset after feature engineering, i.e., feature input, will be provided as input to the subsequent prediction model for training and prediction.

[0082] For example, Feature Engineering Unit 2 might obtain hourly electricity usage data from smart meters. In the preprocessing stage, it cleans out abnormal readings due to meter failures and fills in missing data due to communication interruptions. It might then extract features such as peak electricity usage hours during the day and differences in electricity usage between weekdays and weekends, and use these features as input to the forecasting model to predict future electricity demand.

[0083] Through the processing of feature engineering unit 2, the smart electricity information management system can more accurately capture the key information in the power data, thereby improving the performance of the prediction model and the reliability of the prediction results.

[0084] The multi-model fusion unit 3 is used to select a matching prediction model combination from the model database according to the prediction demand to perform prediction processing on the feature input for prediction to obtain a prediction result; the prediction model combination includes at least one prediction model, and the prediction demand includes power demand prediction and power supply prediction;

[0085] Specifically, the main function of the multi-model fusion unit 3 is to select and combine multiple prediction models from the model database according to specific prediction requirements, perform prediction processing on the feature input after feature engineering, and finally output the prediction results. The specific steps and functions are as follows:

[0086] Forecasting needs identification:

[0087] Demand type: Forecasted demand may include electricity demand forecast (such as electricity consumption forecast in the future period) and electricity supply forecast (such as power generation and power supply reliability forecast).

[0088] Demand analysis: Analyze the specific goals of the forecast, such as short-term, medium-term or long-term forecasts, and the time granularity of the forecast (such as hours, days, months).

[0089] Model database:

[0090] Model storage: The model database stores a variety of prediction models, which may include time series models (such as ARIMA, SARIMA), machine learning models (such as random forest, support vector machine), deep learning models (such as neural network, recurrent neural network), etc.

[0091] Model characteristics: Each model has its own specific advantages and applicable scenarios, and can show different performance under different data characteristics and prediction requirements.

[0092] Select a forecast model combination:

[0093] Model matching: Based on the forecast requirements, the system will select the most appropriate forecast model or model combination from the model database. For example, for power demand forecasting with obvious seasonality, the SARIMA model may be selected.

[0094] Ensemble strategies: A predictive model combination can be a simple collection of models or a more complex ensemble learning method such as stacking, boosting, or bagging.

[0095] Prediction Processing:

[0096] Model training: Use feature input to train the selected model and adjust model parameters to optimize performance.

[0097] Model fusion: Fusion of prediction results from multiple models. Common fusion methods include weighted averaging, voting mechanisms, or more advanced integration techniques.

[0098] Get the prediction result:

[0099] Result output: The multi-model fusion unit 3 finally outputs the prediction result after fusion processing, which is more accurate and reliable than the prediction result of a single model.

[0100] Assume that it is necessary to predict the power demand for the next week. The multi-model fusion unit 3 selects a prediction model combination consisting of an ARIMA model, a random forest model, and a neural network model from the model database. Each model performs prediction processing on the feature input separately, and then the prediction results of the three models are fused by the weighted average method to obtain the final power demand prediction result.

[0101] Through the application of the multi-model fusion unit 3, the smart electricity information management system can make full use of the advantages of different models, improve the accuracy of prediction and the adaptability of the system, and thus better support the operation and management decisions of the power system.

[0102] An adaptive learning unit 4 is used to construct a prediction model for the prediction demand through deep learning or select a prediction model to be adaptively learned from the model database, and add the constructed or selected prediction model to the model database after adaptive learning training based on environmental changes, emergencies or nonlinear effects;

[0103] Specifically, the main function of the adaptive learning unit 4 is to build a new prediction model or select a model from an existing model database and perform adaptive learning training on it through deep learning. This enables the prediction model to better adapt to environmental changes, emergencies or nonlinear characteristics of data, thereby improving the accuracy of prediction and the flexibility of the system. The specific steps and functions are as follows:

[0104] Forecast demand analysis:

[0105] Analyze current forecasting needs to determine whether a new forecasting model needs to be built or an existing model needs to be selected for adaptive learning.

[0106] Model building or selection:

[0107] Building a new model: If the prediction requirements are special or the existing model cannot meet the requirements, the adaptive learning unit 4 will build a new prediction model.

[0108] Select existing models: If there are suitable models in the model database, unit 4 will select these models for subsequent adaptive learning.

[0109] Adaptive learning training:

[0110] Data preparation: Collect the latest data, including historical data on environmental changes, emergencies, and other variables relevant to the forecast.

[0111] Training process: The selected model is trained using a deep learning algorithm, which can learn complex patterns and relationships from large amounts of data, allowing the model to adapt to nonlinear influences.

[0112] Adjustment and optimization: Adjust model parameters based on training results and optimize model structure to improve the model’s prediction performance under new environmental conditions.

[0113] Model Update:

[0114] Update the model trained by adaptive learning back to the model database to ensure that the model in the database always maintains the latest learning and adaptation status.

[0115] Model database update:

[0116] Add new model: If a new model is built, add it to the model database.

[0117] Update the existing model: If adaptive learning is performed on an existing model, the corresponding model record in the database is updated.

[0118] Assume that the power system experiences a rare natural disaster, which causes a significant change in the power supply pattern. The adaptive learning unit 4 selects relevant prediction models from the model database and uses new data containing the impact of the disaster to perform adaptive learning training on these models. After the training is completed, the updated model can better predict the power demand and supply after the disaster, and is thus added back to the model database for future use.

[0119] Through the adaptive learning unit 4, the smart electricity information management system can continuously improve and optimize the prediction model to ensure the accuracy of the prediction results and the long-term effectiveness of the system.

[0120] A reinforcement learning unit 5, used for acquiring incremental data samples, and performing reinforcement learning training on multiple prediction models in the model database based on the incremental data samples;

[0121] Specifically, the main function of the reinforcement learning unit 5 is to use the newly collected data samples (i.e., incremental data samples) to perform reinforcement learning training on the prediction model in the model database. This training method can gradually improve the prediction ability of the model and make it better adapt to the changes and trends of the data. The specific steps and functions are as follows:

[0122] Get incremental data samples:

[0123] Incremental data samples refer to newly collected data after the last training of the model. These data may contain new market trends, changes in user behavior, technological advances, and other factors, which have a significant impact on the prediction performance of the model.

[0124] Data preparation:

[0125] Preprocess the incremental data samples to ensure data quality and consistency and prepare for enhanced learning training.

[0126] Reinforcement Learning Training:

[0127] Select model: Select the prediction model that needs reinforcement learning from the model database.

[0128] Training process: The selected model is trained using a reinforcement learning algorithm. Reinforcement learning is a method of learning optimal behavior strategies by interacting with the environment, and it is able to learn without the full data distribution.

[0129] Strategy update: Adjust the model parameters and prediction strategy based on incremental data samples to improve the model's prediction performance.

[0130] Model Evaluation:

[0131] During reinforcement learning training, the performance of the model is evaluated regularly to ensure that the training is on the right track and the model is improving.

[0132] Model Update:

[0133] The model trained by reinforcement learning is updated back to the model database, so that the model in the model database always maintains the latest learning status.

[0134] Assume that a new demand pattern emerges in the electricity market, for example, due to the popularity of electric vehicles, the demand for electricity increases at night. The reinforcement learning unit 5 collects incremental data samples during this period and uses this data to perform reinforcement learning training on the electricity demand forecasting model in the model database. Through this training, the model is able to learn the new demand pattern and take these changes into account in future forecasts.

[0135] Through the application of enhanced learning unit 5, the smart electricity information management system can not only use historical data, but also learn from the latest data, thereby continuously improving the accuracy and adaptability of the prediction model. This is crucial for the reliable operation and effective management of the power system.

[0136] Furthermore, the data integration unit includes:

[0137] A real-time data acquisition module is used to acquire external data from different sources in real time; the external data from different sources include meteorological data, market demand data and social and economic activity data;

[0138] The time prediction module is used to analyze the external data based on a time-driven prediction method to predict whether an emergency event will occur, and trigger an early warning mechanism to issue an early warning notification when an emergency event occurs.

[0139] Specifically, the data integration unit is an important part of the smart electricity information management system. It is responsible for collecting and processing various external data so that the system can conduct in-depth analysis of these data and predict possible emergencies. The following is a detailed explanation of the real-time data acquisition module and time prediction module in the data integration unit:

[0140] Real-time data acquisition module:

[0141] Function: The main task of the real-time data acquisition module is to collect external data from multiple sources in real time. This data is crucial for predicting power supply status and potential risks.

[0142] Data source:

[0143] Meteorological data: including temperature, humidity, wind speed, rainfall, weather conditions, etc. These data can affect electricity demand and supply. For example, extreme weather may cause damage to power facilities.

[0144] Market demand data: data related to changes in demand in the electricity market, such as electricity consumption, load peak, and demand data by user type (such as residential, commercial, and industrial).

[0145] Socioeconomic activity data: including information on population growth, economic development indicators, holiday schedules, large-scale events, etc., which may affect electricity demand patterns.

[0146] Function: By acquiring these data in real time, the system can promptly understand the external factors that may affect the power system, providing a basis for subsequent analysis and prediction.

[0147] Time prediction module:

[0148] Function: The time prediction module uses a time-driven prediction method to analyze external data in real time to predict whether an emergency will occur and activate the early warning mechanism when an emergency is detected.

[0149] Prediction Method:

[0150] This module may use techniques such as time series analysis, trend analysis, and periodic analysis to predict the changing trends of data.

[0151] By setting thresholds and rules, the module is able to identify unusual patterns in the data that may indicate an impending emergency.

[0152] Early warning mechanism:

[0153] When the prediction model analysis results show that an emergency may occur (such as a surge in electricity demand, supply disruption, etc.), the module will trigger the early warning mechanism.

[0154] Early warning notifications may include sending alarms to power system operators, displaying warning messages on the system interface, notifying relevant personnel via text messages or emails, etc.

[0155] Function: The function of the time prediction module is to identify potential power supply risks in advance so that power companies can take preventive measures to reduce the impact of emergencies on the power system.

[0156] In summary, the data integration unit provides the smart electricity information management system with real-time monitoring and prediction capabilities of the external environment through the collaborative work of the real-time data acquisition module and the time prediction module, thereby improving the reliability and emergency response capabilities of the power system.

[0157] Furthermore, the feature engineering unit comprises:

[0158] A timeliness adjustment module is used to obtain power data within the most recent preset time period;

[0159] A data enhancement module, configured to generate synthetic data based on the power data acquired by the timeliness adjustment module, so as to increase data samples used for model training; the synthetic data is a data sample similar to the acquired power data generated by using a simulation model, or a data sample simulating an emergency or abnormal situation;

[0160] A preprocessing module is used to preprocess the power data obtained by the timeliness adjustment module for missing values, abnormal values ​​and duplicate data to obtain a power data set, and to extract features from the data of the power data set to obtain feature inputs for prediction; and is also used to extract features from the synthetic data generated by the data enhancement module to obtain feature inputs for training the model.

[0161] Specifically, the feature engineering unit is a key component in the smart electricity information management system. It is responsible for processing and converting raw power data to provide high-quality input features for the prediction model. The following is a detailed explanation of each module in the feature engineering unit:

[0162] Timeliness adjustment module:

[0163] Function: The purpose of the timeliness adjustment module is to obtain power data within the recent preset time period from the data source. These data are usually real-time or near real-time. For the prediction model, they contain the latest information and are crucial for capturing the current state and recent trends of the power system.

[0164] Purpose: By acquiring data from the most recent time period, the module ensures the relevance and timeliness of the data, which is very important for predicting upcoming events or state changes.

[0165] Data augmentation module:

[0166] Function: The data enhancement module generates synthetic data based on the power data obtained by the timeliness adjustment module. These synthetic data are used to increase the data sample size for model training and improve the generalization ability of the model.

[0167] Synthetic Data Generation:

[0168] Simulation Model: Use simulation models to generate data samples similar to actual acquired power data, which can simulate the behavior of the power system under normal operating conditions.

[0169] Simulate emergencies or abnormal situations: Generate data samples that simulate emergencies (such as power facility failures, extreme weather events) or abnormal situations (such as sudden fluctuations in electricity demand) to train the model's predictive capabilities in these situations.

[0170] Effect: Through data augmentation, the model can be trained in a wider range of situations, thereby improving its robustness and accuracy in practical applications.

[0171] Preprocessing module:

[0172] Function: The preprocessing module is responsible for cleaning and converting the raw power data obtained by the timeliness adjustment module, as well as extracting features to generate feature inputs suitable for model training and prediction.

[0173] Data cleaning:

[0174] Missing value handling: filling or removing missing values ​​in a dataset.

[0175] Outlier handling: Identify and handle outliers in a data set, which may be caused by measurement errors or system failures.

[0176] Duplicate data processing: Remove duplicate records from a data set.

[0177] Feature extraction:

[0178] Perform feature extraction on the cleaned power data set, such as extracting statistical features of the time series (mean, variance, peak, etc.), frequency features, or other advanced features.

[0179] The same feature extraction process is performed on the synthetic data generated by the data augmentation module to ensure the consistency of the training data.

[0180] Function: The preprocessing module ensures that the data input into the model is clean, consistent, and has predictive value, which is crucial for training efficient and reliable prediction models.

[0181] Through the collaborative work of these three modules, the feature engineering unit can provide high-quality feature input for the prediction model of the smart electricity information management system, thereby improving the prediction performance and practicality of the entire system.

[0182] Furthermore, the multi-model fusion unit comprises:

[0183] The selection module selects the best model combination based on the historical performance of the model and the data distribution of the current feature input;

[0184] A learning method fusion module, used for fusing the multiple prediction models of the best model combination by weighting or stacking;

[0185] The outlier processing module is used to automatically identify and correct anomalies or deviations in the prediction results of the prediction model.

[0186] Specifically, the multi-model fusion unit is an advanced processing module in the smart power information management system, which improves the accuracy and reliability of prediction by combining the advantages of multiple prediction models. The following is a detailed explanation of each module in the multi-model fusion unit:

[0187] Select Module:

[0188] Function: The responsibility of the selection module is to select the best model combination based on the historical performance of the model and the data distribution of the current feature input. This selection process is based on the past performance of the model and the characteristics of the current data to ensure that the selected model is best suited to the current forecasting needs.

[0189] Selection basis:

[0190] Model historical performance: Consider the performance of the model on historical data, including indicators such as prediction accuracy, stability, and computational efficiency.

[0191] Data distribution of current feature input: Analyze the characteristics of the current data, such as distribution pattern, feature range, etc., to determine which models may be more suitable for processing this data.

[0192] Function: The selection module ensures that the system uses a verified model combination that is suitable for the current data characteristics, thereby improving the accuracy of the prediction.

[0193] Learning method fusion module:

[0194] Function: The learning method fusion module is responsible for fusing the selected best model combination. The purpose of fusion is to combine the advantages of different models to generate a more accurate and robust prediction result.

[0195] Fusion method:

[0196] Weighted: Assign different weights to the prediction results of each model based on its historical performance or some evaluation criteria, and then weighted average these results.

[0197] Stacking: A hierarchical model integration technique that first uses multiple different models to make predictions, and then uses another model (usually a simple linear model) to combine these predictions.

[0198] Function: The learning method fusion module improves the stability and accuracy of prediction results through effective model fusion strategies.

[0199] Outlier processing module:

[0200] Function: The outlier processing module automatically identifies and corrects anomalies or deviations in the prediction model results. These outliers may be caused by inaccurate predictions of the model, data quality issues, or rare events that actually occurred.

[0201] Treatment method:

[0202] Identify: Use statistical methods or machine learning techniques to detect outliers in the forecast results.

[0203] Correction: Correct identified outliers, perhaps by replacing them, adjusting weights, or re-running the predictive model.

[0204] Function: The outlier processing module ensures the reliability and practicality of the prediction results and reduces the impact of erroneous predictions on power system operation decisions.

[0205] In summary, the multi-model fusion unit improves the overall prediction performance of the smart electricity information management system by selecting the best model combination, fusing the prediction results of different models, and processing outliers in the prediction, providing strong support for the stable operation and optimized scheduling of the power system.

[0206] Furthermore, the multi-model fusion unit comprises:

[0207] A deep neural network module, used to construct a prediction model for prediction demand through deep learning or to select a prediction model to be adaptively learned from the model database;

[0208] The adaptive algorithm module is used to perform adaptive learning and training on the constructed or selected prediction model based on environmental changes, emergencies or nonlinear effects, and then add it to the model database.

[0209] Specifically, the deep neural network module and the adaptive algorithm module in the multi-model fusion unit are two key components that work together to improve the prediction capabilities of the smart electricity information management system. The following is a detailed explanation of these two modules:

[0210] Deep Neural Network Module:

[0211] Function: The deep neural network module uses deep learning technology to build new prediction models or select existing prediction models from the model database for further training and learning. Deep Neural Networks (DNNs) is a complex neural network structure that can process the deep features of learning data through multiple layers.

[0212] Build prediction models: This module can design and train one or more deep neural network models that can learn complex patterns and relationships from large amounts of electricity data for power demand forecasting, power supply forecasting, etc.

[0213] Model selection: The module can also select already trained deep neural network models from the model database. These models may already be suitable for specific prediction tasks, but need to be adaptively learned based on new data or conditions.

[0214] Adaptive algorithm module:

[0215] Function: The adaptive algorithm module is responsible for adaptive learning and training of the prediction model constructed or selected by the deep neural network module. This training is to enable the model to adapt to environmental changes, emergencies or nonlinear effects of data.

[0216] Adaptive learning training:

[0217] Environmental changes: When the external environment (such as weather, economic conditions, etc.) changes, the model needs to adjust its parameters to maintain the accuracy of the prediction.

[0218] Unexpected events: When unexpected events occur (such as natural disasters, power grid failures, etc.), the model needs to quickly adapt to these abnormal situations to provide accurate predictions.

[0219] Nonlinear influences: Power systems often have nonlinear characteristics, and adaptive learning can help the model better capture these complex nonlinear relationships.

[0220] Update the model database: The model trained by adaptive learning will be added back to the model database so that it can be quickly called and used in future prediction tasks.

[0221] Through the cooperation of these two modules, the multi-model fusion unit can ensure that the prediction model in the smart power information management system is always in the optimal state, can effectively deal with various complex situations, and improve the accuracy of prediction and the robustness of the system. This is of great significance for the stable operation of the power system, energy management, and emergency response.

[0222] Furthermore, it also includes:

[0223] The mobile terminal is used to log in, log out and operate the information management system.

[0224] The mobile terminal is integrated with an archived work order data upload module and a query module. The archived work order data upload module is used to store and display archived work orders, and the query module is used to query the archived work orders.

[0225] Specifically, in the smart electricity information management system, the addition of mobile terminals provides greater flexibility and convenience for system operation. The following is a detailed explanation of the mobile terminal and its integrated archived work order data upload module and query module:

[0226] Mobile terminals:

[0227] Function: Mobile terminals refer to portable devices such as smartphones and tablets that can be remotely connected to the smart electricity information management system. Users can log in to the system through mobile terminals to perform operations such as viewing data, performing forecasting tasks, managing work orders, etc., and can also log out of the system after completing the operation.

[0228] Login / Logout: Users log in to the system by entering their username and password or other authentication methods (such as QR code scanning, fingerprint recognition, etc.). Logout means that users can safely exit the system after completing the operation to protect the security of their account information.

[0229] Archived work order data upload module:

[0230] Function: The archived work order data upload module is a function on the mobile terminal that allows users to upload archived work order data of business expansion (i.e. power business expansion) to the information management system. These work orders may include records of power facility installation, maintenance, inspection, etc.

[0231] Storage and display: The uploaded work order information is stored in the system and can be displayed to authorized users. This helps maintain the integrity and traceability of the work order records and facilitates future audits and inquiries.

[0232] Query module:

[0233] Function: The query module allows users to search and query archived work orders on mobile terminals. Users can find specific work order records by entering keywords, work order numbers, user names, etc.

[0234] Query results: Query results usually display detailed information of the work order, including attachment details, work order number, user name, user number, and electricity address, etc. Users can click on the attachment details to view the attachments of the work order, such as photos, documents, etc.

[0235] By integrating these modules on mobile terminals, the smart electricity information management system provides convenient data entry and query tools for on-site staff, improving work efficiency and data real-time performance. This is of great significance to customer service, on-site management and decision support of power companies.

[0236] Furthermore, the query conditions of the query module are: power supply unit, work order number, user number, user name;

[0237] The query results output by the query module are: attachment details, work order number, user name, user number and electricity address. When the user clicks on the attachment details, a pop-up window displays the attachments of the business expansion archiving link under this work order, which corresponds to the attachment management of the business expansion archiving link in the marketing system.

[0238] Specifically, in the smart electricity information management system, the query module is a tool for retrieving and displaying work order information. The following is a detailed explanation of the query conditions and output query results of the query module:

[0239] Query conditions:

[0240] Power supply unit: One of the query conditions, allowing users to filter work order records based on the power supply unit (such as a substation or power supply area).

[0241] Work Order Number: Another search condition, users can search for a specific work order by entering a specific work order number.

[0242] User ID: Users can use the user ID as a query to find all tickets related to a specific user.

[0243] Username: Users can also query tickets by entering a user name, which is useful when the user knows the name but is not sure of the number.

[0244] These query conditions can be used individually or in combination to accurately locate the work order information required by the user.

[0245] Query results:

[0246] Attachment details: An item in the query results that users can click to view attachments related to the work order, such as documents, pictures, tables, etc.

[0247] Work order number: The query results display the unique number of the work order, which makes it easier for users to confirm the queried work order information.

[0248] User Name: Displays the user name corresponding to the work order to help users identify the owner of the work order.

[0249] User ID: The query results contain the user ID, which provides another identification information of the user.

[0250] Electricity usage address: Displays the user's electricity usage address, which is very important for on-site staff to perform field operations.

[0251] Attachment details display:

[0252] When the user clicks "Attachment Details" in the query results, the system will pop up a window to display all the attachments in the business expansion and archiving link under this work order. These attachments may include engineering drawings, acceptance reports, maintenance records, etc., which are important document evidence in the work order execution process.

[0253] Attachment management corresponding to the business expansion archiving link in the marketing system: This means that the query module is tightly integrated with the business expansion archiving link in the marketing system, and users can directly access and manage the attachments of work orders in the information management system without switching to other systems.

[0254] In this way, the query module not only provides powerful search functions, but also ensures the integrity and accessibility of work order information, thereby improving the work efficiency and customer service level of power companies.

[0255] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A smart electricity information management system, characterized in that: include: A data integration unit, which is used to obtain external data from different sources in real time and predict emergencies related to power supply based on the external data; A feature engineering unit, used for acquiring power data, preprocessing the power data to obtain a power data set, and extracting features from the power data set to obtain feature input for prediction; A multi-model fusion unit is used to select a matching prediction model combination from a model database according to a prediction requirement to perform prediction processing on a feature input for prediction to obtain a prediction result; the prediction model combination includes at least one prediction model, and the prediction requirement includes a power demand prediction and a power supply prediction; An adaptive learning unit, used to construct a prediction model for prediction demand through deep learning or select a prediction model to be adaptively learned from the model database, and add the constructed or selected prediction model to the model database after adaptive learning training based on environmental changes, emergencies or nonlinear effects; The reinforcement learning unit is used to obtain incremental data samples and perform reinforcement learning training on multiple prediction models in the model database based on the incremental data samples.

2. The system according to claim 1, characterized in that The data integration unit comprises: A real-time data acquisition module is used to acquire external data from different sources in real time; the external data from different sources include meteorological data, market demand data and social and economic activity data; The time prediction module is used to analyze the external data based on a time-driven prediction method to predict whether an emergency event will occur, and trigger an early warning mechanism to issue an early warning notification when an emergency event occurs.

3. The system according to claim 2, characterized in that The feature engineering unit comprises: A timeliness adjustment module is used to obtain power data within the most recent preset time period; A data enhancement module, configured to generate synthetic data based on the power data acquired by the timeliness adjustment module, so as to increase data samples used for model training; the synthetic data is a data sample similar to the acquired power data generated by using a simulation model, or a data sample simulating an emergency or abnormal situation; A preprocessing module is used to preprocess the power data obtained by the timeliness adjustment module for missing values, abnormal values ​​and duplicate data to obtain a power data set, and to extract features from the data of the power data set to obtain feature inputs for prediction; and is also used to extract features from the synthetic data generated by the data enhancement module to obtain feature inputs for training the model.

4. The system according to claim 3, characterized in that The multi-model fusion unit comprises: The selection module selects the best model combination based on the historical performance of the model and the data distribution of the current feature input; A learning method fusion module, used for fusing the multiple prediction models of the best model combination by weighting or stacking; The outlier processing module is used to automatically identify and correct anomalies or deviations in the prediction results of the prediction model.

5. The system according to claim 4, characterized in that The multi-model fusion unit comprises: A deep neural network module, used to construct a prediction model for prediction demand through deep learning or to select a prediction model to be adaptively learned from the model database; The adaptive algorithm module is used to perform adaptive learning and training on the constructed or selected prediction model based on environmental changes, emergencies or nonlinear effects, and then add it to the model database.

6. The system according to claim 5, characterized in that Also includes: The mobile terminal is used to log in, log out and operate the information management system.

7. The system according to claim 6, characterized in that The mobile terminal is integrated with an archived work order data upload module and a query module. The archived work order data upload module is used to store and display archived work orders, and the query module is used to query the archived work orders.

8. The system according to claim 7, characterized in that The query conditions of the query module are: power supply unit, work order number, user number, user name; The query results output by the query module are: attachment details, work order number, user name, user number and electricity address. When the user clicks on the attachment details, a pop-up window displays the attachments of the business expansion archiving link under this work order, which corresponds to the attachment management of the business expansion archiving link in the marketing system.

Images