A smart energy system optimization scheduling method
Through the smart energy system optimization scheduling method, load prediction and optimization scheduling are used using multi-source data and machine learning models, the problems of low prediction accuracy and poor scheduling flexibility in scheduling management of new energy system are solved, and efficient resource utilization and system stability are achieved.
Patent Information
- Application Number
- CN202411830617.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-12
- Publication Date
- 2025-04-04
AI Technical Summary
In the scheduling management, existing new energy systems have problems such as low prediction accuracy, poor scheduling flexibility, low resource utilization rate and poor system stability, and it is difficult to effectively match the power supply and demand.
A smart energy system optimization scheduling method is adopted to perform load prediction through preprocessing and feature engineering of multi-source data, combined with machine learning models, and define the optimization objective function to solve optimization problems through linear planning or mixed integer planning methods, and adjust the power scheduling strategy in real time.
It improves the accuracy of load prediction and scheduling flexibility, enhances resource utilization and system stability, significantly reduces supply and demand differences, and improves the reliability and user satisfaction of the power system.
Smart Images
Figure CN119761726A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of energy dispatching and relates to a method for optimizing dispatching of a smart energy system. Background Art
[0002] With the continuous growth of global energy demand and the improvement of environmental protection awareness, the proportion of new energy systems (such as solar energy, wind energy, etc.) in electricity supply is increasing. However, the power generation characteristics of new energy are highly random and uncertain, which brings great challenges to the dispatching and management of power systems. Traditional load forecasting methods are usually based on historical data and statistical models, and the prediction accuracy is not high, which makes it difficult to accurately match supply and demand in the actual dispatching process, and it is easy to have a situation of insufficient supply or oversupply.
[0003] At present, the dispatching management of smart energy systems mainly faces the following problems:
[0004] 1. Low prediction accuracy: Traditional load forecasting methods rely on historical data and statistical models and cannot accurately capture the randomness and uncertainty of power demand.
[0005] 2. Poor scheduling flexibility: Existing scheduling methods are usually based on fixed scheduling plans, which make it difficult to quickly respond to real-time changes in electricity demand.
[0006] 3. Low resource utilization: Inaccurate prediction and scheduling lead to waste of energy resources and inefficient utilization.
[0007] 4. Poor system stability: The situation of insufficient supply or oversupply can easily lead to instability in the power system and affect the quality of power supply.
[0008] In summary, the existing new energy system has many problems in scheduling and management, and an efficient optimization scheduling method is urgently needed to improve the accuracy of load forecasting and the flexibility of scheduling, so as to achieve efficient utilization of resources and stable operation of the system. Summary of the invention
[0009] The purpose of the present invention is to provide a smart energy system optimization scheduling method, aiming to solve the problems of low prediction accuracy and poor scheduling flexibility in the prior art, thereby realizing intelligent optimization scheduling of new energy systems.
[0010] The technical solution adopted by the present invention is a smart energy system optimization scheduling method, which is specifically implemented according to the following steps:
[0011] S1. Data preprocessing:
[0012] Collect multi-source data and pre-process them to ensure that the multi-source data are within the same dimension range and form a standardized multi-source data set;
[0013] S2. Feature Engineering:
[0014] Extract features from multi-source data sets, reduce the dimensionality of the features, and obtain feature vectors;
[0015] S3. Load forecast:
[0016] Carry out load testing through machine learning models, combine feature vectors, and finally generate load forecast values;
[0017] S4. Optimization scheduling model:
[0018] Define the optimization objective function to solve the optimization, combine the load forecast value, and obtain the minimized supply and demand difference;
[0019] S5. Real-time adjustment:
[0020] Real-time monitoring of power demand and resource status, combined with minimizing the gap between supply and demand, and making timely adjustments;
[0021] S6. Evaluation and Feedback:
[0022] Adjust the model parameters according to the adjustment results and continue to perform the above steps.
[0023] Furthermore, in S1, the multi-source data includes historical power demand data, weather data, and holiday data.
[0024] Furthermore, in S1, the preprocessing of multi-source data includes cleaning processing and normalization processing. The cleaning processing includes processing of missing values, outliers and duplicate values. The normalization processing is used to ensure that each data is within the same dimensional range.
[0025] Furthermore, the normalization process is to scale the data to the [0, 1] interval, and the formula is:
[0026]
[0027] In the formula, x represents the original data point, min(x) represents the minimum value of all values in the data set, max(x) represents the maximum value of all values in the data set, and x' represents the normalized data point;
[0028] Standardization is to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is:
[0029]
[0030] In the formula, μ represents the mean value of all values in the data set, and σ represents the standard deviation of the data set.
[0031] Furthermore, the features extracted from the data set in S2 include time features, weather features and holiday features; wherein the time features include hours, days, weeks and months, the weather features include temperature, humidity and wind speed, and the holiday features include days and peak periods.
[0032] Furthermore, principal component analysis is used in S2 to reduce the dimension of features and reduce the data dimension.
[0033] Furthermore, the machine learning model in S3 uses a long short-term memory network (LSTM) or a support vector machine (SVM) for load forecasting.
[0034] Furthermore, the load forecast value is obtained in S3 by the following formula:
[0035]
[0036] Where X represents the feature vector, f represents the machine learning model, Represents the load forecast value.
[0037] Furthermore, in S4, a linear programming or mixed integer programming (MIP) method is used to solve the optimization problem.
[0038] Furthermore, the following formula is used in S4 to optimize the objective function:
[0039]
[0040] In the formula, S represents the minimum supply-demand difference, y t Represents the actual load value, w t represents the amount of resource waste, α represents the weight parameter, and its value range is 0.1-1.0.
[0041] The present invention improves the accuracy of load forecasting through the fusion and feature processing of multi-source data; enhances the flexibility of scheduling through real-time monitoring and dynamic adjustment; improves resource utilization and system stability through optimizing the scheduling model, and effectively solves the problems of low prediction accuracy and poor scheduling flexibility in the prior art. This method not only improves the accuracy of load forecasting and the flexibility of scheduling, but also improves the resource utilization and system stability, and has significant economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a flow chart of a smart energy system optimization scheduling method of the present invention. DETAILED DESCRIPTION
[0043] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0044] like Figure 1As shown, a smart energy system optimization scheduling method of the present invention specifically includes the following steps:
[0045] S1. Data preprocessing
[0046] The key to this step is to collect and integrate data from different sources, such as historical electricity demand, weather forecasts, holiday schedules, etc. These data are cleaned (removing or correcting missing values, outliers, and duplicate records) and normalized (converting data to the same dimensional range, such as 0-1) to ensure consistency and accuracy in subsequent processing. The preprocessed data set becomes a standardized multi-source data set, which is convenient for subsequent feature engineering processing. The details are as follows:
[0047] S1.1 Data collection
[0048] Collect multiple types of data, including historical electricity demand data, weather data, and holiday data. This data can be obtained from multiple sources, such as sensors, weather stations, and power grid monitoring systems.
[0049] Historical electricity demand data: Obtain electricity consumption records from the power grid company over the past few years and store them by timestamp (minute or hour level).
[0050] Weather data: including temperature, humidity, wind speed, precipitation, etc., which can be obtained from the Meteorological Bureau API interface.
[0051] Holiday data: national holidays, local holidays, school holidays, etc., which can be obtained through public datasets or government websites.
[0052] S1.2 Data cleaning
[0053] Missing value processing: For a small number of missing values, interpolation methods (such as linear interpolation and spline interpolation) can be used to fill them; for a large number of missing values, consider deleting related records or using statistical methods (such as mean and median) to replace them.
[0054] Outlier detection: Use statistical methods (such as Zscore and IQR) to identify and process outliers to prevent extreme values from affecting model training.
[0055] Duplicate value processing: Delete identical records to avoid data redundancy.
[0056] S1.3 Data normalization
[0057] MinMax normalization: Scale the data to the [0, 1] interval, the formula is:
[0058]
[0059] Where x represents the original data point, min(x) represents the minimum value of all values in the data set, max(x) represents the maximum value of all values in the data set, and x' represents the normalized data point.
[0060] Zscore standardization: Convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is:
[0061]
[0062] In the formula, μ represents the mean value of all values in the data set, and σ represents the standard deviation of the data set.
[0063] MinMax normalization: It is suitable for scenarios where the data range is known and the data needs to be compressed to a specific interval (such as [0, 1]). It is very sensitive to extreme values of the data, and if there are outliers in the data set, it may cause distortion of the normalized data.
[0064] Z-score standardization: It is suitable for scenarios where the data distribution is close to normal distribution and the dimension effect needs to be eliminated. It can handle outliers in the data and make the data have zero mean and unit variance, which is suitable for many machine learning algorithms.
[0065] Through these two normalization and standardization methods, the dimensional differences and distribution characteristics of the data can be effectively processed, and the training effect and prediction accuracy of the model can be improved.
[0066] S2. Feature Engineering
[0067] Feature engineering is to extract features from preprocessed multi-source data sets and reduce the dimensionality of the features to obtain more effective feature vectors. Feature engineering is one of the key factors in the performance of machine learning models. Useful information is mined from preprocessed multi-source data sets to construct feature vectors. Time features, weather features, and holiday features are extracted. Then, dimensionality reduction techniques such as principal component analysis (PCA) are applied to remove redundant features and retain a few principal components that best reflect the essence of the data, effectively reducing the computational complexity while maintaining model performance. The details are as follows:
[0068] S2.1 Feature extraction
[0069] Time features: Extract time information such as year, month, day, hour, day of the week, and holiday identifiers to capture periodic changes in time.
[0070] Weather characteristics: temperature, humidity, wind speed, precipitation, etc., so that the model can understand the impact of weather conditions on electricity demand.
[0071] Holiday characteristics: whether it is a holiday, whether it is a peak period, the number of holidays, etc., help the model identify demand changes on special days.
[0072] S2.2 Feature Coding
[0073] One-hot encoding: Convert categorical variables (such as day of the week, type of holiday) into binary vectors.
[0074] Numerical processing: continuous variables (such as temperature and humidity) are directly used as feature inputs.
[0075] S2.3 Feature Dimensionality Reduction
[0076] In order to reduce data dimensions and improve model training efficiency and generalization ability, principal component analysis (PCA) is usually used for dimensionality reduction. PCA is an unsupervised linear dimensionality reduction method that retains the main variation information of the data by projecting high-dimensional data into a low-dimensional space.
[0077] Principal Component Analysis (PCA): Projects high-dimensional data into a low-dimensional space through linear transformation, retaining the direction of maximum variance and reducing the number of features.
[0078] tSNE: Suitable for nonlinear dimensionality reduction, mainly used for visualizing high-dimensional data, and can also be used for feature dimensionality reduction in some scenarios.
[0079] S3. Load Forecasting
[0080] Load forecasting is one of the core steps of the smart energy system optimization dispatching method. At this stage, load forecasting is performed through machine learning models, combined with feature vectors, and finally the load forecast value is generated.
[0081] The prediction model is trained based on feature vectors using advanced machine learning algorithms such as LSTM (Long Short-Term Memory Network) or SVM (Support Vector Machine). LSTM is particularly suitable for processing time series data and can capture long-term dependencies, while SVM is good at solving nonlinear classification problems. The prediction model estimates future power demand by learning historical patterns, and the output is the load forecast value. The details are as follows:
[0082] S3.1 Model selection
[0083] Long Short-Term Memory Network (LSTM): Suitable for processing time series data and can capture long-term dependencies. It is a special recurrent neural network (RNN) that can effectively process long-term dependencies in time series data. LSTM controls the flow of information through a gating mechanism (input gate, forget gate, and output gate), avoiding the gradient vanishing and gradient exploding problems in traditional RNNs. In load forecasting, LSTM can capture the historical trends and periodic changes in electricity demand and improve forecasting accuracy.
[0084] Support Vector Machine (SVM): Applicable to small sample data, it uses kernel techniques to handle nonlinear relationships. VM is a machine learning method based on statistical learning theory, which is applicable to classification and regression problems of small sample data. In load forecasting, SVM can map data to high-dimensional space through kernel techniques and find the optimal hyperplane, thereby achieving accurate prediction of power demand.
[0085] S3.2 Model Training
[0086] Data partitioning: Divide the dataset into training set, validation set, and test set, usually in a ratio of 70%:15%:15%.
[0087] Hyperparameter tuning: Use grid search or random search methods to find the best combination of model parameters.
[0088] Model training: Use gradient descent method (such as Adam, SGD) to optimize model parameters. The loss function usually chooses mean square error (MSE) or mean absolute error (MAE).
[0089] S3.3 Prediction output
[0090] Model reasoning: Input the feature vector into the trained model to obtain the load forecast value.
[0091] The formula means:
[0092]
[0093] Where X represents the feature vector, f represents the machine learning model, Represents the load forecast value.
[0094] S4. Optimized scheduling model
[0095] The purpose of establishing an optimization objective function is to minimize the difference between supply and demand, improve resource utilization, minimize the gap between actual load and predicted load, and consider resource waste while meeting the power demand. Linear programming (LP) or mixed integer programming (MIP) methods can be used to solve the problem. By solving this optimization problem, the optimal scheduling strategy can be obtained. The details are as follows:
[0096] S4.1 Define optimization goals
[0097] Objective function: Minimize the gap between actual load and predicted load, while taking resource waste into consideration. The objective function is optimized using the following formula:
[0098]
[0099] In the formula, S represents the minimum supply-demand difference, y t Represents the actual load value, w trepresents the amount of resource waste, α represents the weight parameter, and its value range is 0.1-1.0.
[0100] S4.2 Solution method
[0101] Linear Programming (LP): Applicable to linear objective functions and constraints, using simplex method or interior point method to solve. Linear programming is a mathematical optimization method that is applicable to linear objective functions and linear constraints. In the optimal scheduling model, the objective function can be defined as minimizing the difference between supply and demand, and the constraints include resource availability, grid capacity, etc.
[0102] Mixed Integer Programming (MIP): Applicable to optimization problems involving integer variables, solved using branch and bound or cutting plane methods. Mixed integer programming is an extended linear programming method that allows some variables to be integers. In the optimization scheduling model, MIP can better handle discrete decision problems, such as the start and stop status of generators.
[0103] S4.3 Constraints
[0104] Resource constraints: Ensure that the total power generation does not exceed the total available resources.
[0105] Supply and demand balance: Ensure that the power generation in each period is equal to or greater than the load demand.
[0106] Equipment capacity: Ensure that the output power of each power generation equipment does not exceed its rated capacity.
[0107] S5. Real-time adjustment
[0108] Real-time adjustment is an important part of the optimization dispatch method of smart energy systems. It aims to make timely adjustments by monitoring power demand and resource status in real time and minimizing the difference between supply and demand. The purpose of real-time adjustment is to improve the flexibility and response speed of the system and ensure the stability and reliability of power supply.
[0109] The system continuously monitors the current power demand and available resource status. When any deviation is detected, it immediately makes corresponding adjustments based on the results of the optimization scheduling model to ensure that power supply and demand are always balanced. This process emphasizes the system's dynamic adaptability and ability to respond quickly to emergencies. The details are as follows:
[0110] S5.1 Monitoring System
[0111] Data collection: Real-time collection of power demand and resource status data, usually at a frequency of minutes or seconds.
[0112] Anomaly detection: Use statistical methods or machine learning models to detect abnormal fluctuations and issue timely alarms.
[0113] S5.2 Dynamic Adjustment
[0114] Feedback mechanism: Adjust and optimize the parameters of the scheduling model based on real-time data to ensure that the scheduling strategy is always optimal.
[0115] Automatic execution: Through the automated control system, dispatch instructions are executed in real time to adjust the output power of power generation equipment.
[0116] S6. Evaluation and Feedback
[0117] Evaluation and feedback is the last step of the smart energy system optimization dispatch method, which aims to adjust the model parameters according to the results of the adjustment and continuously optimize the dispatch performance. After each adjustment, the results are evaluated and the model parameters are updated accordingly. This is an iterative cycle. Over time, the model will gradually become more accurate and the dispatch decisions will become more efficient. Through continuous self-learning and improvement, the entire system can better cope with various challenges that may arise in the future. The details are as follows:
[0118] S6.1 Performance evaluation
[0119] Forecast accuracy: Use indicators such as mean square error (MSE) and mean absolute error (MAE) to evaluate the accuracy of the load forecasting model.
[0120] Scheduling effect: Evaluate indicators such as supply-demand gap, resource utilization, and system stability to ensure the effectiveness of the scheduling strategy.
[0121] S6.2 Parameter Adjustment
[0122] Model update: Adjust model parameters based on evaluation results, retrain the model, and improve prediction accuracy and scheduling effects.
[0123] Continuous optimization: Regularly review and optimize the entire scheduling process, introduce new data sources and technical means, and continuously improve system performance.
[0124] In order to verify the effect of the technical solution of the present invention, the following test is carried out
[0125] Experimental design: Select a city's power grid as the experimental object and collect one year's power demand data, weather data and holiday data.
[0126] The data is divided into a training set and a test set. The training set is used to train the load forecasting model, and the test set is used to verify the prediction accuracy of the model.
[0127] Experimental method: Use LSTM model to forecast load and generate load forecast value. Define optimization objective function, use linear programming method to solve optimization problem and generate optimized dispatching scheme. Monitor power demand and resource status in real time, adjust load forecasting model and optimize dispatching scheme, and generate real-time dispatching scheme. Evaluate dispatching scheme, adjust model parameters and generate optimized dispatching scheme.
[0128] Test results:
[0129] Forecast accuracy: On the test set, the mean absolute error (MAE) of load forecasting is 2.5% and the mean square error (MSE) is 0.05.
[0130] Supply-demand difference: The optimized scheduling scheme significantly reduced the supply-demand difference, with an average supply-demand difference of 1.5%.
[0131] Resource Utilization: Resource utilization increased by 15% and energy costs decreased by 10%.
[0132] System stability: The number of power outages has been reduced by 50%, and user satisfaction has been significantly improved.
[0133] Conclusion: The test results show that the smart energy system optimization scheduling method proposed in this invention can significantly improve the accuracy of load forecasting, enhance the flexibility of scheduling, improve resource utilization, and enhance the stability of the system. This method has high feasibility and effectiveness in practical applications. The details are as follows:
[0134] 1. Improve load forecasting accuracy
[0135] Through the fusion and feature processing of multi-source data, more comprehensive and accurate features are extracted, and the accuracy of the load forecasting model is improved. Compared with the traditional load forecasting method, the prediction error of the present invention is reduced by more than 20%. High-precision load forecasting helps to match supply and demand more accurately, reduce the difference between supply and demand, and improve the stability and reliability of the system.
[0136] 2. Enhance scheduling flexibility
[0137] Through real-time monitoring and adjustment, the present invention can quickly respond to changes in power demand and improve the flexibility of scheduling. In actual operation, the difference between supply and demand has been reduced by more than 30%. The flexible scheduling mechanism helps to deal with emergencies such as extreme weather and equipment failures, ensuring the continuity and reliability of power supply.
[0138] 3. Improve resource utilization
[0139] By optimizing the scheduling model, the present invention can allocate energy resources more reasonably and reduce resource waste. Resource utilization rate is increased by more than 15%, and energy cost is reduced by more than 10%. Efficient resource utilization not only saves energy, but also reduces environmental pollution, which meets the requirements of sustainable development.
[0140] 4. Improve system stability
[0141] By reducing the difference between supply and demand, the present invention significantly improves the stability of the power system and improves the quality of power supply. The number of power outages has been reduced by more than 50%, and user satisfaction has been significantly improved. Stable power supply is of great significance to ensuring social production and residents' lives, especially in key industries and important places.
Claims
1. A smart energy system optimization scheduling method, characterized in that: Follow the steps below to implement it: S1. Data preprocessing: Collect multi-source data and pre-process them to ensure that the multi-source data are within the same dimension range and form a standardized multi-source data set; S2. Feature Engineering: Extract features from multi-source data sets, reduce the dimensionality of the features, and obtain feature vectors; S3. Load forecast: Carry out load testing through machine learning models, combine feature vectors, and finally generate load forecast values; S4. Optimization scheduling model: Define the optimization objective function to solve the optimization, combine the load forecast value, and obtain the minimized supply and demand difference; S5. Real-time adjustment: Real-time monitoring of power demand and resource status, combined with minimizing the gap between supply and demand, and making timely adjustments; S6. Evaluation and Feedback: Adjust the model parameters according to the adjustment results and continue to perform the above steps.
2. A smart energy system optimization scheduling method according to claim 1, characterized in that: In S1, the multi-source data includes historical power demand data, weather data, and holiday data.
3. The method for optimizing and dispatching a smart energy system according to claim 1, characterized in that: In S1, the preprocessing of the multi-source data includes cleaning processing and normalization processing, the cleaning processing includes processing of missing values, abnormal values and duplicate values, and the normalization processing is used to ensure that each data is within the same dimensional range.
4. A smart energy system optimization scheduling method according to claim 3, characterized in that: The normalization process is to scale the data to the interval [0, 1], and the formula is: In the formula, x represents the original data point, min(x) represents the minimum value of all values in the data set, max(x) represents the maximum value of all values in the data set, and x' represents the normalized data point; The standardization process is to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is: In the formula, μ represents the mean value of all values in the data set, and σ represents the standard deviation of the data set.
5. The method for optimizing and dispatching a smart energy system according to claim 1, characterized in that: The features extracted from the data set in S2 include time features, weather features and holiday features; wherein the time features include hours, days, weeks and months, the weather features include temperature, humidity and wind speed, and the holiday features include days and peak periods.
6. A smart energy system optimization scheduling method according to claim 1, characterized in that: In S2, principal component analysis is used to reduce the dimension of features and reduce the data dimension.
7. A smart energy system optimization scheduling method according to claim 1, characterized in that: The machine learning model in S3 uses a long short-term memory network LSTM or a support vector machine SVM for load forecasting.
8. The method for optimizing and dispatching a smart energy system according to claim 1, characterized in that: In S3, the load forecast value is obtained by the following formula: Where X represents the feature vector, f represents the machine learning model, Represents the load forecast value.
9. The method for optimizing and dispatching a smart energy system according to claim 1, characterized in that: In S4, a linear programming or mixed integer programming method is used to solve the optimization problem.
10. The method for optimizing and dispatching a smart energy system according to claim 1, characterized in that: The following formula is used in S4 to optimize the objective function: In the formula, S represents the minimum supply-demand difference, y t Represents the actual load value, w t represents the amount of resource waste, α represents the weight parameter, and its value range is 0.1-1.0.