Power prediction method and system based on multi-source and multi-dimensional CIM model

Abstract

The invention provides an electric power prediction method and system based on a multi-source and multi-dimensional CIM model, and relates to the field of electric power system automation and communication. Comprising the steps of expanding an existing information model standard to construct a multi-source and multi-dimensional information model, integrating data of multiple energy types, including power generation power data, meteorological data and environmental data, and forming a standardized model by analyzing an existing class and inheriting or creating a new class to add attributes and relationships; historical data in the multi-source multi-dimensional information model are preprocessed, and abnormal data are cleaned by adopting a multi-dimensional verification mechanism and are intelligently corrected; constructing and training a neural network model, and processing a time sequence data capture dependency relationship; the neural network model after deployment training interacts with the multi-source multi-dimensional information model through an interface, real-time prediction of the generated power is realized, the long-term dependency relationship and complex nonlinear characteristics of the water-wind-light generated power are effectively captured, the prediction precision and reliability of the generated power are improved, and the safety and stability of power grid operation are improved.

Description

Technical Field

[0001] This invention relates to the fields of power system automation and communication, and in particular to a power prediction method and system based on a multi-source, multi-dimensional CIM model. Background Technology

[0002] As the informatization process of power enterprises continues to advance, a large number of application management systems have been put into actual operation. However, these systems lack unified standards in the design and development stages, resulting in significant differences in the data model structure and incompatibility of application interfaces among the systems. This leads to a series of key problems: on the one hand, the same data needs to be repeatedly entered into different systems, causing serious data redundancy, which not only increases the workload of data management but also wastes storage resources; on the other hand, due to different data sources, the data values ​​of the same indicators often have deviations, making it difficult to ensure data consistency. Ultimately, this plunges power enterprises into the predicament of "information silos," making it impossible to achieve data sharing and collaborative operation among multiple systems.

[0003] Currently, under the background of energy transition, the proportion of renewable energy sources such as wind power, photovoltaics, and hydropower in the power system continues to increase. However, the following shortcomings exist: First, the power generation process of renewable energy is easily affected by natural factors such as meteorology and hydrology, exhibiting significant volatility and uncertainty, posing a huge challenge to the stable operation of the power system. Second, the instability of energy supply leads to an increased risk of grid supply-demand imbalance, affecting the efficiency of resource optimization and causing energy waste such as wind and solar curtailment. It also threatens the security and stability of the power system, making it difficult to meet the response requirements of demand-side management and hindering the efficient utilization of renewable energy and the intelligent development of the power system. Third, while there are many prediction models for single energy sources in existing technologies, there is no unified standardized information model to integrate multi-source, multi-dimensional data from water, wind, and solar power. This makes it difficult to effectively characterize the complementary characteristics and coupling relationships between multiple energy sources, resulting in insufficient prediction accuracy and a need to improve resource allocation efficiency. Summary of the Invention

[0004] The main objective of this invention is to provide a power forecasting method and system based on a multi-source, multi-dimensional CIM model, which solves the technical problems of insufficient stability of power systems, lack of unified standardized information models, insufficient forecasting accuracy, and low resource allocation efficiency in the prior art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a power prediction method based on a multi-source, multi-dimensional CIM model, comprising the following steps: The existing information model standard is extended to construct a multi-source, multi-dimensional information model. The multi-source, multi-dimensional information model integrates data from multiple energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction. The historical data in the multi-source, multi-dimensional information model is preprocessed, and a multi-dimensional verification mechanism is used to clean up abnormal data and perform intelligent correction in order to obtain high-quality data. A neural network model is constructed and trained, which processes time series data to capture dependencies and enables prediction of energy generation. The deployed and trained neural network model interacts with a multi-source, multi-dimensional information model through an interface to make real-time predictions to optimize resource allocation.

[0006] In the preferred scheme, after data collection, the following steps are also included: Data mapping: Based on the extended multi-source, multi-dimensional information model, data interface services and database services are constructed. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates.

[0007] In the preferred embodiment, the construction of data interface services and database services based on the extended multi-source multi-dimensional information model includes: The data interface service occurs between business systems. It converts the extended multi-source, multi-dimensional information model into a document object model by parsing it, and accesses it in a tree structure to generate formatted data for uploading and downloading. The database service occurs between the business system and the database. It obtains data types and values ​​by parsing the extended multi-source multi-dimensional information model and uses mapping tools to generate statements to realize data loading and reading. The data interface service and database service work together to update the data in the expanded multi-source multi-dimensional information model, and the work together to update ensures the compatibility of data from multiple energy types. The data interface service processes real-time data transmission, which includes power generation data and meteorological data.

[0008] In the preferred embodiment, extending existing information model standards to construct a multi-source, multi-dimensional information model includes: Analyze whether similar classes exist in existing information model standards: If it exists, inherit from the similar class, add new attributes to build a new class and establish the relationship, then add the new class to the package; If it does not exist, create a new class, add attributes and establish relationships, and add the new class to the package; A standardized information model for component modules is constructed for various energy types. The standardized information model covers wind power modules, photovoltaic modules, hydropower modules, energy storage modules, and grid command modules. An interface model is established based on the component modules, and the interface model supports the integration of data from multiple energy types.

[0009] In the preferred embodiment, the preprocessing of historical data in the multi-source, multi-dimensional information model includes: A multi-dimensional verification mechanism is used to extract and classify target data, which includes power generation data, meteorological data and hydrological data, and is stored according to energy type and data type. The validity of data is determined by spatiotemporal correlation verification based on geographical location and time series, the deviation rate is calculated and valid data is marked. A coupling model is established through energy coupling verification, and the deviation between the actual value and the benchmark value is calculated to determine anomalies. Dynamic threshold cleaning is used to determine threshold ranges based on historical data and different scenarios, thereby eliminating extreme and abnormal data. Abnormal data is corrected using a model through multi-source cross-validation and intelligent correction.

[0010] In the preferred embodiment, the step of determining data validity based on geographical location and time series through spatiotemporal correlation verification includes: The target data and related data are merged and aligned by time, with the related data originating from surrounding sources; Calculate the deviation rate, which is based on the merged data: If the deviation rate is within the threshold, it is marked as valid data; otherwise, it is marked as abnormal. For time series data, the sampling interval is checked, and if it is missing, the preceding and following data are correlated to determine the abnormal missing data.

[0011] In the preferred embodiment, establishing the coupling model through energy coupling verification includes: For photovoltaic types, a power reference value is constructed based on irradiance and temperature data. The deviation between the actual output power and the power reference value is calculated. If the deviation exceeds the threshold, it is judged as abnormal. For wind power types, a power coupling model is constructed based on wind speed and air density data, and the deviation is calculated. For each type of hydropower, a threshold range is constructed by combining inflow and water level data to determine whether the range is exceeded. The anomaly detection input dynamic threshold cleaning is generated by the coupling model.

[0012] In the preferred embodiment, the construction and training of the neural network model includes: Initialize the neural network model by adding layers, including layers for processing time series data, layers to prevent overfitting, and layers for outputting predictions. Compiling a neural network model uses an optimizer and a loss function, which is based on the error. Fit the neural network model using the training set, set the number of rounds and batch size, and evaluate it on the validation set; The test set is used for prediction and error calculation, and then inverse normalization is performed to generate a comparison.

[0013] In the preferred embodiment, the deployed and trained neural network model interacts with the multi-source, multi-dimensional information model through an interface, including: The trained neural network model is deployed on the server via an interface that supports real-time data interaction. Real-time data is obtained from a multi-source, multi-dimensional information model, including power generation data and meteorological data. Prediction results are generated by processing real-time data obtained through a trained neural network model. Resource allocation is optimized based on forecast results, and this optimization includes scheduling multiple energy types.

[0014] Secondly, the present invention provides a power prediction system based on a multi-source, multi-dimensional CIM model, comprising: The data acquisition module is used to extend the existing information model standard to build a multi-source, multi-dimensional information model. The multi-source, multi-dimensional information model integrates data from multiple energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction. The data mapping module is used to build data interface services and database services based on the extended multi-source and multi-dimensional information model. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates. The preprocessing module is used to preprocess historical data in the multi-source multi-dimensional information model, and uses a multi-dimensional verification mechanism to clean up abnormal data and perform intelligent correction in order to obtain high-quality data. The model building module is used to build and train a neural network model that processes time series data to capture dependencies and enable prediction of energy generation. The model prediction module is used to deploy trained neural network models to interact with multi-source, multi-dimensional information models through an interface to perform real-time predictions in order to optimize resource allocation.

[0015] This invention provides a power forecasting method and system based on a multi-source, multi-dimensional CIM model. By constructing a complete power forecasting system based on a multi-source, multi-dimensional CIM model, a full-process solution from data acquisition, preprocessing, model building to real-time forecasting is formed. By extending existing CIM standards to construct a multi-source, multi-dimensional information model, a standardized and unified description of power system data is achieved, completely solving the "information silo" problem caused by incompatible data models. By constructing data interface services and database services, efficient and seamless data transmission between business systems and linked updates between the database and the CIM model are achieved. Data preprocessing, including multi-dimensional verification mechanisms to clean abnormal data and perform intelligent correction, comprehensively improves data quality, integrity, and accuracy. By constructing and training an LSTM neural network model, which interacts with the multi-source, multi-dimensional information model after training, real-time prediction of power generation is achieved, effectively capturing the long-term dependence and complex nonlinear characteristics of hydropower, wind power, and solar power generation, improving the prediction accuracy and reliability of power generation, and enhancing the safety and stability of power grid operation. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the power prediction method of the present invention; Figure 2 This is a flowchart of the extended method of the CIM standard of this invention; Figure 3 This is a comprehensive modeling structure diagram based on CIM for this invention; Figure 4 This is a flowchart illustrating the implementation of the data model service of this invention. Detailed Implementation

[0017] Example 1 like Figure 1-4 As shown, a power forecasting method based on a multi-source, multi-dimensional CIM model includes the following steps: S1. Data Acquisition: Extend the existing Information Model (CIM) standard to build a multi-source, multi-dimensional information model. The multi-source, multi-dimensional information model integrates data from various energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction.

[0018] S2. Data Mapping: Based on the extended multi-source, multi-dimensional information model, data interface services and database services are constructed. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates.

[0019] S3. Preprocessing: Preprocess the historical data in the multi-source, multi-dimensional information model, use a multi-dimensional verification mechanism to clean up abnormal data and perform intelligent correction to obtain high-quality data.

[0020] S4. Model Building: Build and train a neural network model that processes time series data to capture dependencies and enable prediction of energy generation.

[0021] S5. Model Prediction: The deployed and trained neural network model interacts with the multi-source, multi-dimensional information model through an interface to make real-time predictions to optimize resource allocation.

[0022] In this embodiment, a complete power forecasting system based on a multi-source, multi-dimensional CIM model is constructed, forming a full-process solution from data acquisition, preprocessing, model building to real-time forecasting: By extending the existing CIM standard to build a multi-source, multi-dimensional information model, a standardized and unified description of power system data is achieved, completely solving the "information silo" problem caused by incompatible data models; by building data interface services and database services, efficient and seamless data transmission between business systems and linkage updates between the database and the CIM model are achieved; by preprocessing the data, including cleaning abnormal data with a multi-dimensional verification mechanism and performing intelligent correction, the data quality is comprehensively improved, enhancing data integrity and accuracy; by constructing and training an LSTM neural network model, after model training, interaction with the multi-source, multi-dimensional information model enables real-time forecasting of power generation, effectively capturing the long-term dependence and complex nonlinear characteristics of hydropower, wind power, and solar power generation, improving the forecasting accuracy and reliability of power generation, and enhancing the safety and stability of power grid operation.

[0023] In this embodiment, step S1 extends the existing CIM standard to build an information model, eliminating the need to build a business model from scratch. The required attribute extensions can be added to an existing mature business model. When extending the CIM standard, it is necessary to analyze whether similar classes exist in the existing model, specifically as follows: Figure 2 As shown.

[0024] In the preferred embodiment, existing information model standards are extended to construct multi-source, multi-dimensional information models, including: Analyze whether similar classes exist in existing information model standards: If it exists, inherit from the similar class, add new attributes to build a new class and establish the relationship, then add the new class to the package.

[0025] If it does not exist, create a new class, add attributes and establish relationships, and add the new class to the package.

[0026] Standardized information models for component modules are constructed for various energy types, covering wind power modules, photovoltaic modules, hydropower modules, energy storage modules, and grid command modules.

[0027] An interface model is established based on the component modules, and the interface model supports the integration of data from multiple energy types.

[0028] By eliminating data redundancy and improving consistency through a standardized information model, which integrates real-time and historical meteorological data to support forecasting, this model helps to eliminate data redundancy and improve consistency.

[0029] This embodiment extends the existing CIM standard to achieve a standardized and unified description of power system data, solving the "information silo" problem caused by incompatible data models; the multi-source, multi-dimensional information model integrates power generation data, meteorological data, and environmental data from various energy sources such as hydropower, wind power, and photovoltaics, improving data compatibility and interoperability between different systems.

[0030] like Figure 3 As shown, step S1 constructs a multi-source, multi-dimensional information model, including: constructing standardized information models for each component module, and then establishing a CIM interface model. The various component modules included in the regional power grid system are identified, including wind farms, photovoltaic power stations, hydropower units, load side, energy storage power stations, and grid-side instructions. Existing CIM standards have designed generator set classes for the generation side. Based on the CIM extension method described in the above steps, standardized modeling is performed for various components such as wind, solar, and hydropower stations.

[0031] In this embodiment, the parameters of the constructed multidimensional information model are defined as follows: : A collection of public information models (including core base classes + extended classes for water, wind and light).

[0032] For a multi-source set, including (Hydropower) (Wind power) (Photovoltaics).

[0033] It is a multidimensional set of attributes, where: Topology dimension (device connection relationships, such as plant-bay-equipment-terminal association); Physical parameter dimension (inherent attributes of the equipment, such as rated power and capacity); : Operational status dimension (real-time / steady-state operating data, such as output and voltage); Time series dimension (historical / forecast data, such as wind speed-power output curves, irradiance time series); Constraint dimensions (operational restrictions, such as upper and lower limits of storage capacity, wind speed cut-out threshold).

[0034] : CIM core base classes (such as Equipment, PowerSystemResource, Terminal); : Extended device class of power supply X ( (e represents a specific device, such as a Turbine). Attribute mapping function: Maps the device class C to the corresponding multidimensional attribute D.

[0035] There are many types of regional power grid designs. The corresponding Chinese and English names of each module are shown in Table 1.

[0036] Table 1. Chinese and English names of each category

[0037] In this embodiment, as Figure 4 As shown, step S2 implements the data interface service and the database service.

[0038] In the preferred embodiment, data interface services and database services are constructed based on the extended multi-source, multi-dimensional information model, including: The data interface service occurs between business systems. It converts the extended multi-source, multi-dimensional information model into a document object model by parsing, and accesses the data in a tree structure to generate formatted data for uploading / downloading. In this embodiment, JSON format is used.

[0039] Database services occur between the business system and the database. They obtain data types and values ​​by parsing the extended multi-source, multi-dimensional information model, and use mapping tools to generate database SQL statements to load / read data.

[0040] like Figure 3 As shown, the core of the data interface service and database service implementation lies in parsing the CIM XML model file. XML is a hierarchical structure. The approach to parsing an XML file is to convert it into a Document Object Model (DOM) and load it into memory, then access the DOM in the order of the tree structure.

[0041] The data interface service and database service work together to update the data in the expanded multi-source, multi-dimensional information model, ensuring the compatibility of data from various energy types.

[0042] Real-time data transmission is handled through a data interface service. The real-time data includes power generation data and meteorological data.

[0043] The database service stores historical data, which is used for subsequent preprocessing.

[0044] In this embodiment, by constructing a data interface service and a database service, efficient and seamless data transmission between business systems and linked updates between the database and the CIM model are achieved. The data interface service ensures the standardization of data exchange between systems by parsing the CIM model and converting it into JSON format. The database service realizes automatic data loading and reading through the mapping of the CIM model to SQL statements, which greatly improves data management efficiency and system integration convenience.

[0045] In this embodiment, step S3 involves preprocessing the historical data in the multi-source, multi-dimensional CIM model of water, wind, and solar energy using a cleaning method that combines five-dimensional verification with intelligent correction.

[0046] In the preferred scheme, historical data in the multi-source, multi-dimensional information model undergoes preprocessing, including: Step S3.1 Five-dimensional intelligent cleaning steps Multi-source data extraction and classification: Extract target data such as hydropower, wind and solar power generation power, meteorological data (irradiance, temperature, wind speed, cloud cover), hydrological data (water inflow, reservoir water level), and grid-side data, and classify and store them according to energy type (wind power, photovoltaic, hydropower) and data type (real-time data, historical data, weather forecast data).

[0047] The relevant runtime code is: import pandas as pd # Extracting multi-dimensional target variables columns_needed = ['date', 'time', 'wind_power', 'wind_speed', 'solar_power', 'irradiance', 'temperature', 'hydro_power', 'water_flow', 'reservoir_level', 'grid_command'] df_filtered = df[columns_needed] # Classified by Energy Type wind_df = df_filtered[['date', 'time', 'wind_power', 'wind_speed', 'grid_command']] solar_df = df_filtered[['date', 'time', 'solar_power', 'irradiance','temperature', 'grid_command']] hydro_df = df_filtered[['date', 'time', 'hydro_power', 'water_flow','reservoir_level', 'grid_command']] Step S3.2: Determine data validity based on geographical location and time series through spatiotemporal correlation verification, calculate the deviation rate, and mark valid data. The target data and related data are merged and aligned by time, with the related data originating from surrounding sources; 1) Calculate the deviation rate, based on the merged data: If the deviation rate is within the threshold, it is marked as valid data; otherwise, it is marked as abnormal.

[0048] In this embodiment, the wind speed data deviation of adjacent wind farms in the same area must be within a reasonable threshold (such as ±15%), and the irradiance data of photovoltaic power stations must not deviate from the data of surrounding meteorological stations during the same period by more than ±10%.

[0049] 2) For time series, check the sampling interval. If it is missing, correlate the data before and after to determine the abnormal missing data.

[0050] In terms of time dimension, the data sampling interval in this embodiment must meet the preset standard: such as 1 minute for wind power and 5 minutes for photovoltaic power; missing sampling points must be marked and correlated with the data before and after them to determine whether they are abnormally missing.

[0051] Effective data supports subsequent energy coupling verification, and effective data integrates geographical location information.

[0052] This embodiment uses spatiotemporal correlation verification as an example (irradiance data of photovoltaic power plants), and the relevant running code is as follows.

[0053] def spatial_temporal_check(solar_df, weather_station_data): # Merge data from photovoltaic power plants and surrounding weather stations (aligned by time) merged_df = pd.merge(solar_df, weather_station_data[['date', 'time','irradiance_station']], on=['date', 'time'], how='left') # Calculate the deviation rate merged_df['irradiance_diff'] = abs(merged_df['irradiance']- merged_df['irradiance_station']) / merged_df['irradiance_station'] # Retain data with a deviation ≤10%, and mark out-of-tolerance data. merged_df['valid_flag'] = merged_df['irradiance_diff'].apply(lambdax: 1 if x<= 0.1 else 0) return merged_df.

[0054] Step S3.3 Energy Coupling Verification: Establish a coupling model through energy coupling verification, and calculate the deviation between the actual value and the benchmark value to determine anomalies.

[0055] In the preferred scheme, an anomaly detection model is established that couples the characteristics of hydropower, wind power, and solar power with multiple factors. For photovoltaic power, a power prediction baseline is constructed based on irradiance, temperature, and solar altitude angle; an actual output power deviation exceeding ±20% from the baseline value is considered an anomaly. For wind power, a power coupling model is constructed based on wind speed and air density; a deviation exceeding ±18% is considered an anomaly. For hydropower, a power generation threshold range is constructed by combining inflow, reservoir water level, and grid commands; values ​​exceeding this range are marked as an anomaly.

[0056] The following is an example of photovoltaic energy coupling verification in this embodiment.

[0057] def solar_coupling_check(solar_df): # Power baseline values ​​based on irradiance and temperature (simplified model) solar_df['power_base'] = 0.8 * solar_df['irradiance']- 0.02 * solar_df['temperature'] # Calculate the deviation between actual power and reference value solar_df['power_diff'] = abs(solar_df['solar_power']- solar_df['power_base']) / solar_df['power_base'] # Data with a deviation ≤20% is considered valid. solar_df = solar_df[solar_df['power_diff']<= 0.2] return solar_df.

[0058] Step S3.4 Cleans the dynamic threshold by using the anomaly judgment input generated by the coupling model.

[0059] Dynamic threshold cleaning is used to determine threshold ranges based on historical data and different scenarios, thereby eliminating extreme and abnormal data.

[0060] Instead of fixed thresholds, dynamic threshold models are trained based on historical data and categorized by scenario (season, weather type, energy type). Seasonal scenarios are divided into spring, summer, autumn, and winter, and weather scenarios are divided into sunny, cloudy, rainy, and snowy days. Dynamic threshold intervals are determined by statistically analyzing the data distribution characteristics (mean, standard deviation, quartiles) under each scenario, and extreme outliers outside the intervals are eliminated.

[0061] The following is an example of dynamic threshold calculation (wind power) in this embodiment.

[0062] def dynamic_threshold_calculate(wind_df): # Add scene tags wind_df['season'] = pd.to_datetime(wind_df['date']).dt.season wind_df['weather_type'] = get_weather_type(wind_df['date']) # Requires combining meteorological data to obtain the weather type # Calculate dynamic thresholds based on scenarios (interquartile method) dynamic_thresholds = wind_df.groupby(['season', 'weather_type'])['wind_power'].agg( q1=lambda x: x.quantile(0.05), q3=lambda x: x.quantile(0.95) ).reset_index() # Merge thresholds and filter data wind_df = pd.merge(wind_df, dynamic_thresholds, on=['season', 'weather_type'], how='left') wind_df = wind_df[(wind_df['wind_power']>= wind_df['q1'])&(wind_df['wind_power']<= wind_df['q3'])] return wind_df Step S3.5 uses multi-source cross-validation and intelligent correction to correct abnormal data using a model. Data that cannot be corrected is removed based on weights, and quality indicators are calculated to ensure data integrity and accuracy.

[0063] Multi-source cross-validation and intelligent correction: For data with an abnormal single data source but normal related data sources, it is not directly deleted, but intelligently corrected based on multi-source related data. For example, if the irradiance data of a photovoltaic power station is abnormal at a certain moment (negative value or exceeding the reasonable range), but the data of adjacent photovoltaic power stations during the same period and the data of surrounding meteorological stations are normal, the reasonable irradiance value at that moment is predicted by using a gradient boosting tree model; for abnormal data that cannot be corrected (such as data from multiple sources being verified as abnormal), the decision on whether to remove it is based on the data importance weight (data from critical periods has a higher weight than data from non-critical periods).

[0064] # Example of Multi-Source Cross-Validation and Intelligent Correction from sklearn.ensemble import GradientBoostingRegressor def cross_validate_correct(solar_df): # Mark abnormal data solar_df['abnormal_flag'] = solar_df.apply( lambda x: 1 if x['irradiance']<0 or x['irradiance']>2000 else 0, axis=1 ) # Separate normal data from abnormal data normal_data = solar_df[solar_df['abnormal_flag'] == 0] abnormal_data = solar_df[solar_df['abnormal_flag'] == 1] if not abnormal_data.empty: # Train and refine the model using normal data X_train = normal_data[['temperature', 'sun_altitude', 'irradiance_neighbor']] y_train = normal_data['irradiance'] model = GradientBoostingRegressor(n_estimators=100, random_state=42) model.fit(X_train, y_train) # Correcting abnormal data X_abnormal = abnormal_data[['temperature', 'sun_altitude', 'irradiance_neighbor']] abnormal_data['irradiance_corrected'] = model.predict(X_abnormal) # Replace outliers solar_df.loc[solar_df['abnormal_flag'] == 1, 'irradiance']= abnormal_data['irradiance_corrected'] # Remove extreme outlier data that cannot be corrected (all data from multiple sources is outlier) solar_df = solar_df[solar_df['irradiance']>= 0] return solar_df Step S3.6 Data Integration and Quality Assessment: Merge the cleaned energy data, calculate quality indicators such as data integrity rate, anomaly removal rate, and correction accuracy, thereby improving the data integrity rate and anomaly identification accuracy after cleaning. The following is an example of data quality assessment in this embodiment: def data_quality_evaluate(original_df, cleaned_df): original_count = len(original_df) cleaned_count = len(cleaned_df) complete_rate = cleaned_count / original_count # Data integrity rate abnormal_rate = len(original_df[original_df['abnormal_flag'] == 1]) / original_count # Original abnormal rate correction_acc = len(cleaned_df[cleaned_df['corrected_flag'] == 1]) / len(original_df[original_df['abnormal_flag'] == 1])# Correction accuracy return {'complete_rate': complete_rate, 'abnormal_rate': abnormal_rate, 'correction_acc': correction_acc} In this embodiment, a preprocessing mechanism combining multi-dimensional verification and intelligent correction is adopted. Through multiple dimensions of data quality control, such as spatiotemporal correlation verification, energy coupling verification, dynamic threshold cleaning, and multi-source cross-validation, the accuracy of anomaly identification is improved, the overall data quality is enhanced, abnormal data is effectively identified and corrected, and the integrity of cleaned data is improved.

[0065] Step S4: Construct an identification model for water, wind, and solar data using an LSTM model to process and identify anomalous data. LSTM is an ideal choice for processing time series data, capable of capturing long-term dependencies and complex nonlinear features. It includes an input layer, an LSTM layer, a Dropout layer (to prevent overfitting), a fully connected layer, and an output layer.

[0066] In the preferred embodiment, constructing and training the neural network model includes: Initialize the neural network model by adding layers, including layers for processing time series data, layers to prevent overfitting, and layers for outputting predictions.

[0067] Step S4.1: An example of constructing an LSTM model architecture is shown below.

[0068] from keras.models import Sequential from keras.layers import LSTM, Dense, Dropout # Initialize the model model = Sequential() # Add an LSTM layer with 50 hidden units model.add(LSTM(units=50, return_sequences=False, input_shape=(X_train.shape[1], X_train.shape[2]))) # Add a Dropout layer to prevent overfitting model.add(Dropout(0.2)) # Add a fully connected layer (Dense Layer) model.add(Dense(units=1)) # Output layer prediction power # Compilation Model model.compile(optimizer='adam', loss='mean_squared_error') LSTM layer: This layer learns the dependencies in the time series and uses 50 hidden units.

[0069] Dropout layer: Randomly disconnects some neurons during training to prevent the model from overfitting.

[0070] Fully connected layer: Maps the features extracted by the LSTM layer to the power output.

[0071] Step S4.2 Model Training and Optimization Compiling a neural network model uses an optimizer and a loss function, which is based on the error.

[0072] The neural network model is fitted using the training set, with the number of rounds and batch size set, and evaluated on the validation set.

[0073] The test set is used for prediction and error calculation, and then inverse normalization is performed to generate a comparison.

[0074] The training model example for this embodiment is as follows.

[0075] history = model.fit(X_train, y_train, epochs=50, batch_size=32,validation_split=0.2) epochs: The number of training epochs for the model, set to 50.

[0076] batch_size: Use 32 samples per training session to optimize training efficiency.

[0077] validation_split: Uses 20% of the training data as a validation set to monitor the training process.

[0078] Step S4.3 Model Evaluation and Prediction: Capture dependencies through a neural network model, which are derived from preprocessed historical data. Evaluate the model's prediction performance on the test set and compare the predicted data with the actual data.

[0079] from sklearn.metrics import mean_squared_error import matplotlib.pyplot as plt # Make predictions on the test set y_pred = model.predict(X_test) # Denormalize the predicted values y_pred_inverse = scaler.inverse_transform(np.concatenate((y_pred,np.zeros((y_pred.shape[0], scaled_data.shape[1]- 1))), axis=1))[:, 0] y_test_inverse = scaler.inverse_transform(np.concatenate((y_test.reshape(-1, 1), np.zeros((y_test.shape[0], scaled_data.shape[1]- 1))),axis=1))[:, 0] # Calculate the mean squared error (MSE) mse = mean_squared_error(y_test_inverse, y_pred_inverse) print(f'Mean Squared Error: {mse}') # Plot a comparison chart between predicted results and actual power output plt.figure(figsize=(10, 6)) plt.plot(y_test_inverse, label='Actual Power') plt.plot(y_pred_inverse, label='Predicted Power') plt.title('Predicted vs Actual Power Output') plt.xlabel('Time') plt.ylabel('Power Output') plt.legend() plt.show() In this embodiment, by constructing and training an LSTM neural network model, the advantages of its processing time series data are fully utilized to effectively capture the long-term dependence and complex nonlinear characteristics of hydropower, wind power and solar power generation, thereby improving the prediction accuracy and reliability of power generation and enhancing the accuracy of power system optimization scheduling.

[0080] Step S5: Deploy the trained LSTM model on the server via the API interface to interact with the CIM data interface in real time for real-time prediction.

[0081] In the preferred embodiment, the deployed and trained neural network model interacts with the multi-source, multi-dimensional information model through an interface, including: The trained neural network model is deployed on the server via an interface that supports real-time data interaction.

[0082] Real-time data is obtained from a multi-source, multi-dimensional information model, including power generation data and meteorological data.

[0083] Predictive results are generated by processing real-time data obtained through a trained neural network model.

[0084] Resource allocation is optimized based on forecast results, and this optimization includes scheduling multiple energy types.

[0085] The system's stable operation is ensured by predicting the results, and this stable operation is able to cope with environmental changes.

[0086] This embodiment deploys the trained neural network model through the API interface and interacts with the CIM data interface in real time, realizing real-time prediction of power generation, improving the response speed to environmental and climate change, providing decision support for real-time optimized scheduling and resource allocation of the power system, and improving the safety and stability of power grid operation.

[0087] The overall technical solution of this invention enhances interoperability by adopting a unified data model: the CIM model, as a standard data exchange model, provides standardized descriptions of hydropower, wind power, photovoltaic power generation, and other power system components; it comprehensively considers three different types of power generation methods (hydropower, wind power, and solar power) as well as meteorological factors (such as wind speed, solar radiation, and temperature) to form a multi-dimensional data model, which can effectively capture the uncertainty and complementarity of various energy generation methods and accurately reflect the overall operating status and power fluctuation characteristics of the power system; it uses a machine learning model to better capture complex dynamic features in time series, improving the prediction accuracy of power output; it optimizes resource scheduling, reduces operating costs, improves the renewable energy absorption rate of the power grid, and enhances the stability and reliability of the power grid.

[0088] Example 2 Further illustrating with reference to Example 1, a power prediction system based on a multi-source, multi-dimensional CIM model is applicable to the power prediction method based on a multi-source, multi-dimensional CIM model in Example 1, comprising: The data acquisition module is used to extend existing information model standards to build multi-source, multi-dimensional information models. These models integrate data from various energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction.

[0089] The data mapping module is used to build data interface services and database services based on the extended multi-source, multi-dimensional information model. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates.

[0090] The preprocessing module is used to preprocess historical data in the multi-source, multi-dimensional information model, and uses a multi-dimensional verification mechanism to clean up abnormal data and perform intelligent correction in order to obtain high-quality data.

[0091] The model building module is used to build and train a neural network model that processes time series data to capture dependencies and enable prediction of energy generation.

[0092] The model prediction module is used to deploy trained neural network models to interact with multi-source, multi-dimensional information models through an interface to perform real-time predictions in order to optimize resource allocation.

[0093] This embodiment provides the working process, working details and technical effects of a power prediction method based on a multi-source multidimensional CIM model, which can be referred to in Embodiment 1 and will not be repeated here.

[0094] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A power forecasting method based on a multi-source, multi-dimensional CIM model, characterized in that, Includes the following steps: The existing information model standard is extended to construct a multi-source, multi-dimensional information model. The multi-source, multi-dimensional information model integrates data from multiple energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction. The historical data in the multi-source, multi-dimensional information model is preprocessed, and a multi-dimensional verification mechanism is used to clean up abnormal data and perform intelligent correction in order to obtain high-quality data. A neural network model is constructed and trained, which processes time series data to capture dependencies and enables prediction of energy generation. The deployed and trained neural network model interacts with a multi-source, multi-dimensional information model through an interface to make real-time predictions to optimize resource allocation.

2. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 1, characterized in that, After data collection, the following is also included: Data mapping: Based on the extended multi-source, multi-dimensional information model, data interface services and database services are constructed. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates.

3. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 2, characterized in that, The data interface service and database service built based on the extended multi-source multi-dimensional information model include: The data interface service occurs between business systems. It converts the extended multi-source, multi-dimensional information model into a document object model by parsing it, and accesses it in a tree structure to generate formatted data for uploading and downloading. The database service occurs between the business system and the database. It obtains data types and values ​​by parsing the extended multi-source multi-dimensional information model and uses mapping tools to generate statements to realize data loading and reading. The data interface service and database service work together to update the data in the expanded multi-source multi-dimensional information model, and the work together to update ensures the compatibility of data from multiple energy types. The data interface service processes real-time data transmission, which includes power generation data and meteorological data.

4. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 1, characterized in that, The extension of existing information model standards to construct multi-source, multi-dimensional information models includes: Analyze whether similar classes exist in existing information model standards: If it exists, inherit from the similar class, add new attributes to build a new class and establish the relationship, then add the new class to the package; If it does not exist, create a new class, add attributes and establish relationships, and add the new class to the package; A standardized information model for component modules is constructed for various energy types. The standardized information model covers wind power modules, photovoltaic modules, hydropower modules, energy storage modules, and grid command modules. An interface model is established based on the component modules, and the interface model supports the integration of data from multiple energy types.

5. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 1, characterized in that, The preprocessing of historical data in the multi-source, multi-dimensional information model includes: A multi-dimensional verification mechanism is used to extract and classify target data, which includes power generation data, meteorological data and hydrological data, and is stored according to energy type and data type. The validity of data is determined by spatiotemporal correlation verification based on geographical location and time series, the deviation rate is calculated and valid data is marked. A coupling model is established through energy coupling verification, and the deviation between the actual value and the benchmark value is calculated to determine anomalies. Dynamic threshold cleaning is used to determine threshold ranges based on historical data and different scenarios, thereby eliminating extreme and abnormal data. Abnormal data is corrected using a model through multi-source cross-validation and intelligent correction.

6. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 5, characterized in that, The method of determining data validity based on geographical location and time series through spatiotemporal correlation verification includes: The target data and related data are merged and aligned by time, with the related data originating from surrounding sources; Calculate the deviation rate, which is based on the merged data: If the deviation rate is within the threshold, it is marked as valid data; otherwise, it is marked as abnormal. For time series data, the sampling interval is checked, and if it is missing, the preceding and following data are correlated to determine the abnormal missing data.

7. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 5, characterized in that, The establishment of the coupling model through energy coupling verification includes: For photovoltaic types, a power reference value is constructed based on irradiance and temperature data. The deviation between the actual output power and the power reference value is calculated. If the deviation exceeds the threshold, it is judged as abnormal. For wind power types, a power coupling model is constructed based on wind speed and air density data, and the deviation is calculated. For each type of hydropower, a threshold range is constructed by combining inflow and water level data to determine whether the range is exceeded. The anomaly detection input dynamic threshold cleaning is generated by the coupling model.

8. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 1, characterized in that, The construction and training of the neural network model includes: Initialize the neural network model by adding layers, including layers for processing time series data, layers to prevent overfitting, and layers for outputting predictions. Compiling a neural network model uses an optimizer and a loss function, which is based on the error. Fit the neural network model using the training set, set the number of rounds and batch size, and evaluate it on the validation set; The test set is used for prediction and error calculation, and then inverse normalization is performed to generate a comparison.

9. The power forecasting method based on a multi-source, multi-dimensional CIM model according to claim 1, characterized in that, The deployed and trained neural network model interacts with the multi-source, multi-dimensional information model through an interface, including: The trained neural network model is deployed on the server via an interface that supports real-time data interaction. Real-time data is obtained from a multi-source, multi-dimensional information model, including power generation data and meteorological data. Prediction results are generated by processing real-time data obtained through a trained neural network model. Resource allocation is optimized based on forecast results, and this optimization includes scheduling multiple energy types.

10. A power forecasting system based on a multi-source, multi-dimensional CIM model, characterized in that, include: The data acquisition module is used to extend the existing information model standard to build a multi-source, multi-dimensional information model. The multi-source, multi-dimensional information model integrates data from multiple energy types, including power generation data, meteorological data, and environmental data. By analyzing existing classes and inheriting or creating new classes to add attributes and relationships, a standardized model is formed to support data interaction. The data mapping module is used to build data interface services and database services based on the extended multi-source and multi-dimensional information model. The data interface service parses the model and converts it into a format to realize data transmission between systems, while the database service uses a mapping tool to convert the model into statements to realize data updates. The preprocessing module is used to preprocess historical data in the multi-source multi-dimensional information model, and uses a multi-dimensional verification mechanism to clean up abnormal data and perform intelligent correction in order to obtain high-quality data. The model building module is used to build and train a neural network model that processes time series data to capture dependencies and enable prediction of energy generation. The model prediction module is used to deploy trained neural network models to interact with multi-source, multi-dimensional information models through an interface to perform real-time predictions in order to optimize resource allocation.

Images