CN117578440A - Power grid load prediction method based on neural network - Google Patents
Power grid load prediction method based on neural network Download PDFInfo
- Publication number
- CN117578440A CN117578440A CN202311593363.6A CN202311593363A CN117578440A CN 117578440 A CN117578440 A CN 117578440A CN 202311593363 A CN202311593363 A CN 202311593363A CN 117578440 A CN117578440 A CN 117578440A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- load
- prediction
- load prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000011156 evaluation Methods 0.000 claims abstract description 34
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 238000003062 neural network model Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000013480 data collection Methods 0.000 claims abstract description 4
- 238000010606 normalization Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 13
- 238000005457 optimization Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 238000004140 cleaning Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 8
- 238000010200 validation analysis Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000009499 grossing Methods 0.000 claims description 6
- 238000002360 preparation method Methods 0.000 claims description 6
- 238000012800 visualization Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000013500 data storage Methods 0.000 claims description 4
- 238000002474 experimental method Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000003066 decision tree Methods 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 238000012417 linear regression Methods 0.000 claims description 2
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 239000013598 vector Substances 0.000 claims description 2
- 238000013136 deep learning model Methods 0.000 abstract description 6
- 238000013461 design Methods 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000000354 decomposition reaction Methods 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 238000007726 management method Methods 0.000 description 5
- 238000013459 approach Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2123/00—Data types
- G06F2123/02—Data types in the time domain, e.g. time-series data
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Power Engineering (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a power grid load prediction method based on a neural network, which comprises the following 6 steps in sequence: data collection and preprocessing, feature selection and extraction, neural network model design, training models, model testing and evaluation, and model application. Compared with the traditional method, the invention has the following steps: the adaptability is strong: the deep learning model can automatically extract and combine features, does not need manual selection and design of features, and has the advantages of strong adaptability, high prediction precision, strong interpretability, good expansibility, low data quality requirement and the like.
Description
Technical Field
The invention belongs to the technical field of power, and particularly relates to a power grid load prediction method based on a neural network.
The background technology is as follows:
the traditional power grid load prediction method comprises methods based on statistics, time series, machine learning and the like. These approaches predict future load demands based primarily on historical data to guide the scheduling and management of the power system. These conventional prediction methods have the advantage that an appropriate method needs to be selected according to specific scenes and application requirements. For example, statistical-based methods can simply and efficiently predict load, but have high demands on data stationarity and randomness. The time series based approach has good predictive power for seasonal features and periodic variations, but poor adaptability to emergencies and nonlinear factors. The machine learning based approach can effectively mine complex relationships and nonlinear features in the load data, but requires a significant amount of data and model training time.
Disclosure of Invention
Aiming at the defects in the background art, the invention provides a power grid load prediction method based on a neural network. The method comprises the following steps:
a power grid load prediction method based on a neural network is implemented by a computer,
step 1: data collection and preprocessing: and collecting historical power grid load data, and performing data cleaning, denoising, interpolation, normalization and the like.
Step 2: feature selection and extraction: selecting a feature related to load prediction: load data of several days or weeks in the past, weather conditions, time, etc. are characterized. Preprocessing these features, including: smoothing the temperature of the weather, feature crossing and switching, etc. to better represent its relationship to the load.
Step 3: designing a neural network model: parameters such as a topological structure, the number of hidden layers, the number of hidden units, an activation function and the like of the neural network are designed, and adjustment and optimization are carried out according to the actual situation of the load prediction problem.
Step 4: training a model: the data set is divided into a training set and a validation set, the data of the training set is used to train the neural network model, and the validation set is used to evaluate the performance of the model. And continuously adjusting the weight and the offset of the model by using a back propagation algorithm, an optimizer and the like until the model reaches a convergence state.
Step 5: model test and evaluation: the test set is used to verify and evaluate the performance and behavior of the model. Various metrics may be used to measure the accuracy, robustness, and generalization ability of the model, such as Root Mean Square Error (RMSE), mean Absolute Error (MAE), mean Absolute Percent Error (MAPE), and the like.
Step 6: model application: and applying the trained neural network model to an actual scene to perform load prediction. The prediction time step and the prediction period can be set according to actual needs to realize real-time or near real-time load prediction. Advantageous technical effects
The method has the following characteristics:
the adaptability is strong: the deep learning model can automatically extract and combine the features, does not need to manually select and design the features, and has strong adaptability.
The prediction precision is high: the deep learning model can process large-scale and complex data, has better fitting capability on nonlinear and long-term dependence and has relatively higher prediction precision.
The interpretability is strong: the deep learning model can display the activation condition and the feature importance distribution of each layer of neurons through the visualization tool, and is helpful for understanding the prediction process and the result of the model.
The expansibility is good: the deep learning model can be expanded by increasing the depth and width of the network or introducing new neuron types, convolution kernels and the like, and has good expandability.
The data quality requirement is low: the deep learning model has strong fault tolerance to noise, missing and abnormal data, and can rapidly process original and incomplete data.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a block flow diagram of step 1.
Fig. 3 is a block flow diagram of step 2.
Fig. 4 is a block flow diagram of step 3.
Fig. 5 is a block flow diagram of step 4.
Detailed Description
The structural features of the present invention will now be described in detail with reference to the accompanying drawings.
Referring to fig. 1, a power grid load prediction method based on a neural network, by a computer,
step 1: data collection and preprocessing: and collecting historical power grid load data, and performing data cleaning, denoising, interpolation, normalization and the like.
Step 2: feature selection and extraction: selecting a feature related to load prediction: load data of several days or weeks in the past, weather conditions, time, etc. are characterized. Preprocessing these features, including: smoothing the temperature of the weather, feature crossing and switching, etc. to better represent its relationship to the load.
Step 3: designing a neural network model: parameters such as a topological structure, the number of hidden layers, the number of hidden units, an activation function and the like of the neural network are designed, and adjustment and optimization are carried out according to the actual situation of the load prediction problem.
Step 4: training a model: the data set is divided into a training set and a validation set, the data of the training set is used to train the neural network model, and the validation set is used to evaluate the performance of the model. And continuously adjusting the weight and the offset of the model by using a back propagation algorithm, an optimizer and the like until the model reaches a convergence state.
Step 5: model test and evaluation: the test set is used to verify and evaluate the performance and behavior of the model. Various metrics may be used to measure the accuracy, robustness, and generalization ability of the model, such as Root Mean Square Error (RMSE), mean Absolute Error (MAE), mean Absolute Percent Error (MAPE), and the like.
Step 6: model application: and applying the trained neural network model to an actual scene to perform load prediction. The prediction time step and the prediction period can be set according to actual needs to realize real-time or near real-time load prediction.
Further, the step 1 specifically includes 7 steps, which are sequentially: data source selection, data acquisition, data cleaning and denoising, data interpolation, data normalization, data division and data storage.
Referring to fig. 2, further, the contents of the 7 sub-steps in step 1 are respectively:
step 1.1 data source selection: a data source for load prediction is determined. The data source should be real load data collected by the grid operator, energy company or other relevant institutions. The reliability of the data source and the time frame of the data coverage are as comprehensive as possible in view of the accuracy of the predictions.
Step 1.2 data acquisition: historical load data is obtained from the selected data sources. These data are typically in the form of a time series, including a date and a corresponding load value. The data can be obtained through an API interface, database query, file download or other approaches.
Step 1.3, data cleaning and denoising: and performing data cleaning operations such as abnormal value removal, missing value restoration, outlier processing and the like on the acquired data. Outliers may be due to data acquisition errors or other disturbances that need to be rejected or corrected.
Repair missing values: the missing value may be data that has not been captured for some reason during the data acquisition process, or may be data that has been lost or erroneous for various reasons during the data processing process. The missing values need to be repaired before the load prediction model training. The method for repairing the missing value comprises the following steps: deletion of missing values, interpolation, data modeling.
Wherein the deletion value is deleted: if the number of missing values is small, the sample or the feature column where the missing values are located can be deleted directly.
Interpolation method: common interpolation methods include linear interpolation, polynomial interpolation, lagrangian interpolation, and the like.
Modeling data: according to the existing data, a model is built, missing values are predicted, and common methods comprise KNN, decision trees, random forests and the like.
Processing outliers: outliers refer to individual observations that do not fit a certain statistical law or probability model, and may be caused by data input errors, outlier data points, unreasonable data values, and the like. Outliers mislead the construction and training of the model for load prediction, and reduce the accuracy and generalization capability of the model. The method for processing the outlier comprises the following steps: deleting outliers, replacing outliers, clipping and modeling.
Wherein, delete outliers: outliers are directly deleted from the dataset.
Replacing outliers: the outliers are replaced with the mean, median, mode, etc. of the dataset, or with interpolation.
Clipping: and setting upper and lower numerical limits to limit outliers within a reasonable range.
Modeling: and (5) establishing a model according to the existing data, screening out outliers and correcting.
Step 1.4 data interpolation: if there is a missing value in the data, the missing value is filled using an interpolation method. The interpolation method is linear interpolation, lagrange interpolation or spline interpolation. The interpolation method should be selected according to the nature of the data and the requirements of the load prediction problem.
Step 1.5 data normalization: and carrying out normalization processing on the load data to eliminate data differences under different scales. Preferably, the following steps are adopted: min-max normalization or normalization. Normalization can scale the load data to a specific range, helping to improve the stability and convergence speed of the model.
Step 1.6 data partitioning: the complete historical load data set is divided into a training set, a validation set and a test set. A time series division method is generally adopted to ensure that the test set contains a future period to be predicted, and the training set and the verification set are divided according to a certain proportion.
Step 1.7 data storage: the preprocessed data is saved to an appropriate data structure, such as a data table, CSV file, or database, for subsequent model training and predictive use.
Further, the purpose of feature selection and extraction described in step 2 is to determine features related to load prediction, reduce noise and redundant information, thereby improving the predictive power of the model. The characteristic selection and extraction specifically comprises 5 substeps, which are sequentially as follows: feature selection, feature pretreatment, feature extraction, feature normalization, feature selection and extraction experiments.
Referring to fig. 3, further, the details of each sub-step of step 2 are:
step 2.1, feature selection: features related to load prediction are selected based on knowledge and experience with the problem area. Common characteristics include load data, weather conditions, time over the last days or weeks. Feature selection needs to be considered in combination with practical problems to ensure that the selected features are sufficiently representative and informative, while avoiding overfitting due to an excessive number of features.
Step 2.2, pretreatment of characteristics: preprocessing the selected characteristics, specifically: smoothing, feature crossing and/or conversion. The smoothing process may smooth out successive values of temperature, load, etc. to reduce the effects of random variations thereof. Feature crossing and conversion may combine the original features into new features, e.g., converting a timestamp into a day of the week, time of day, etc., to better represent the link between it and the load.
Step 2.3, feature extraction: features are extracted from the original data by using a specific algorithm to reduce noise and redundant information and improve the prediction capability of the model. The feature extraction algorithm is wavelet transformation, decomposition-reconstruction and principal component analysis.
The specific method for extracting the wavelet transformation characteristics comprises the following steps:
and 2.3.1, carrying out wavelet decomposition on the original load data. And selecting proper wavelet basis functions and decomposition layers, and carrying out wavelet decomposition on the original load data. The wavelet basis function is haar or db4.
When wavelet decomposition is carried out, the original load data is subjected to multi-scale frequency domain transformation to obtain detail coefficients and approximate coefficients of different frequencies.
Step 2.3.2 extracting wavelet coefficients as features
Appropriate wavelet coefficients are selected as features, preferably detail coefficients or approximation coefficients of the scale.
The wavelet coefficients may be further feature extracted in combination with statistics, frequency domain indicators, etc.
The specific method for extracting the decomposition-reconstruction characteristics comprises the following steps:
and 2.3.1, decomposing the original load data: the raw load data is frequency domain decomposed using a Fast Fourier Transform (FFT) or wavelet transform. Different frequency bands (such as low frequency, high frequency, etc.) or scale levels can be selected for decomposition.
Step 2.3.2, extracting the decomposed signal as a feature: extracting features from the decomposed signals of the respective frequency bands or scale levels, the features being: mean, variance, spectral ratio statistics.
The method for extracting the principal component analysis features comprises the following steps:
step 2.3.1, data standardization: the raw load data is normalized so that the individual features have the same units and dimensions.
Step 2.3.2, calculating a characteristic covariance matrix: based on the normalized load data, a covariance matrix between features is calculated.
Step 2.3.3, calculating main components of the characteristics: and carrying out eigenvalue decomposition on the covariance matrix to obtain eigenvalues and corresponding eigenvectors, namely principal components.
And 2.3.4, extracting main components as characteristics: according to actual requirements, main components with larger characteristic values are selected as characteristics. In general, principal components with higher cumulative contribution rates are retained to preserve as much information as possible of the original load data.
Step 2.4, feature normalization: and normalizing the selected features to eliminate data differences under different scales. The common normalization methods include min-max normalization and normalization, which are similar to the normalization of the data in step 1.
Step 2.5 feature selection and extraction experiments: and evaluating the effect of feature selection and extraction by using the evaluation index, and repeatedly adjusting according to the experimental result to obtain better feature representation.
Further, the model construction and training described in step 3 includes 5 substeps, in order: model selection, model construction, model training, model evaluation and model optimization.
The model construction of power load prediction is the most core link in the whole prediction process. In the step, a model suitable for predicting load is constructed, and the model is trained, so that the model learns rules and features on a training set, and a foundation is provided for later prediction. The following is the details of step 3:
referring to fig. 4, further, the model construction and training described in step 3 is specifically:
step 3.1 model selection: a model is selected that is appropriate for the load prediction problem. The load prediction model is as follows: linear regression, support Vector Machines (SVMs), decision trees, neural networks, or time series models. An appropriate model needs to be selected according to the actual situation of the problem so as to ensure the prediction accuracy, complexity and stability of the model.
Step 3.2, constructing a model: and constructing a load prediction model by adopting a fully connected neural network, a convolutional neural network, a cyclic neural network or an autoregressive moving average model, and setting parameters and super parameters of the model. Different model building methods correspond to different prediction effects and computational resource costs.
Step 3.3 model training: training and fitting the constructed model using historical load data: during the training process, the model needs to be optimized according to the loss function. The aforementioned loss functions include gradient descent, adam or adagard functions. In order to prevent the overfitting, the training set is divided into a training set and a verification set, and the methods of cross verification and the like are adopted for model selection and parameter adjustment.
Step 3.4 model evaluation: after model training is complete, the model needs to be evaluated to determine its predictive power and accuracy. The evaluation index may select Mean Square Error (MSE), root Mean Square Error (RMSE), and/or Mean Absolute Error (MAE). The smaller the evaluation index, the stronger the predictive power of the model.
Step 3.5, model optimization: and adjusting and optimizing the model according to the evaluation result and the actual demand. The optimization mode in the step comprises increasing the data quantity, selecting more proper characteristics and/or integer model parameters and the like, and the prediction results of different models can be tried to be integrated so as to improve the prediction accuracy of the whole system.
Further, the load prediction and result evaluation in step 4 includes data preparation, feature engineering, load prediction, result evaluation, and model updating and iteration. And step 4, applying the trained model to an actual load prediction task, and evaluating a prediction result.
Further, the specific steps of load prediction and result evaluation in the step 4 are as follows:
step 4.1 data preparation: preparing load data to be predicted: the historical load data or the real-time load data is selected for use as input. It is important to ensure the accuracy and integrity of the data.
Step 4.2, feature engineering: the conversion of the original load data into the feature vector suitable for the model input is a key step in the load prediction process. In the step, the processing such as feature extraction, data smoothing and data standardization can be performed so as to better reflect the rules and features of the load data.
Step 4.3 load prediction: and inputting the preprocessed load data into a trained model to perform load prediction. The model learns the change trend and mode of the load according to the historical data, and the load prediction result in a future period of time is obtained through analysis and calculation of new data.
Step 4.4 evaluation of results: and evaluating the prediction result to measure the accuracy and the prediction capability of the model. The evaluation index such as Mean Square Error (MSE), root Mean Square Error (RMSE), mean Absolute Error (MAE) can be calculated in comparison with the actual observed data. The smaller the evaluation index, the stronger the predictive power of the model.
Step 4.5, updating and iterating the model: over time, load data and demand may change, thus requiring periodic updating of the model and iteration. And (3) carrying out data preparation, model training and prediction evaluation again according to new data and problem demands, and continuously optimizing and improving the load prediction system.
Further, the load prediction and result evaluation in step 4 further includes result visualization: the prediction results are visually displayed so as to be more intuitively understood and analyzed. The prediction result and the actual observation data can be displayed in a line graph, a bar graph and the like, and compared and analyzed. The visualization can help the decision maker to better understand the load prediction results and make corresponding decisions and adjustments.
Further, the application and deployment in step 5 includes model integration, system testing, operation monitoring, and system maintenance.
Further, the details of step 5 are:
5.1 model integration: the load prediction model is integrated into the actual energy management system. According to the system architecture and interface specifications, the input, output and control interfaces of the model are integrated and adjusted with the existing system, so that the model can smoothly cooperate with other systems.
5.2 System test: and performing integrated test and system test. Firstly, unit testing is carried out to check whether the functions of the modules are normal. And then, carrying out an integration test to check whether the overall functions and performances of the system meet the requirements. And finally, performing system testing to ensure the stability, reliability, safety and the like of the system.
5.3 operation monitoring: and performing operation monitoring and fault removal on the system. By means of real-time monitoring and analysis of the load prediction result, performance optimization and parameter adjustment are performed on the system, so that accuracy and stability of load prediction are improved.
5.4 maintenance of the system: and performing system maintenance and updating. According to actual conditions and system requirements, the system is maintained and updated regularly, including data acquisition, model training, parameter adjustment, system safety and the like, and the accuracy and the practicability of load prediction are continuously improved on the premise of ensuring stable operation of the system.
Further, step 6 continuous improvement and optimization includes: monitoring and evaluation, data updating, model adjustment and effect verification.
Further, the specific steps of continuous improvement and optimization in the step 6 are as follows:
6.1 monitoring and evaluation: the performance and accuracy of the load prediction system is continuously monitored and evaluated.
6.2 data update the training data set is updated periodically. The latest load data, weather data, etc. are collected and added to the training data set to maintain the adaptability and accuracy of the model.
6.3 model adjustment: the prediction capability and generalization capability of the model can be improved by changing the model architecture, adjusting the superparameter, introducing new features and the like.
6.4 effect verification: and verifying the effect of the improved load prediction model.
The power grid load prediction application scene based on deep learning comprises the aspects of power supply and demand balance management, power grid operation optimization, power grid planning, energy market prediction and the like. For example, in power supply and demand balance management, accurate load prediction may help power system dispatchers optimize power generation plans and scheduling strategies, ensure power systems operate stably and meet ever-increasing energy demands. In grid operation and planning, load prediction can help power enterprises to better formulate investment plans and maintenance strategies, and system cost and risk are reduced. In terms of energy market prediction, load prediction can help market participants make scientific energy trading decisions and risk management.
The following is an example of simulation data for grid load prediction:
| time stamp | Load value (MW) |
| 2023-01-0100:00:00 | 235 |
| 2023-01-0101:00:00 | 234 |
| 2023-01-0102:00:00 | 245 |
| 2023-01-0103:00:00 | 253 |
| 2023-01-0104:00:00 | 266 |
| 2023-01-0105:00:00 | 281 |
| 2023-01-0106:00:00 | 310 |
| 2023-01-0107:00:00 | 325 |
| 2023-01-0108:00:00 | 345 |
| 2023-01-0109:00:00 | 360 |
| 2023-01-0110:00:00 | 375 |
The above table is a data table containing a time stamp indicating time per hour and a corresponding load value indicating the load condition at that time. The data is collected from historical load data of a certain power system and can be used as a training, verifying and testing data set of a power grid load prediction model. In practical application, more data are required to be collected to obtain more accurate prediction results, and steps such as data processing and feature extraction are performed according to actual requirements to obtain better prediction effects.
Claims (10)
1. A power grid load prediction method based on a neural network is characterized by comprising the following steps of: the following operations are performed by the computer:
step 1: data collection and preprocessing: collecting historical power grid load data, and performing data cleaning, denoising, interpolation, normalization and the like;
step 2: feature selection and extraction: selecting a feature related to load prediction: load data of several days or weeks in the past, weather conditions, time and the like are characterized;
step 3: designing a neural network model: designing parameters such as a topological structure, the number of hidden layers, the number of hidden units, an activation function and the like of the neural network, and adjusting and optimizing according to the actual situation of the load prediction problem;
step 4: training a model: dividing the data set into a training set and a validation set, training the neural network model using the data of the training set, and evaluating the performance of the model using the validation set; continuously adjusting the weight and the offset of the model by a back propagation algorithm, an optimizer and the like until the model reaches a convergence state;
step 5: model test and evaluation: using the test set to verify and evaluate the performance and behavior of the model;
step 6: model application: and applying the trained neural network model to an actual scene to perform load prediction.
2. The neural network-based power grid load prediction method according to claim 1, wherein: step 1 specifically includes 7 steps, which are in turn: data source selection, data acquisition, data cleaning and denoising, data interpolation, data normalization, data division and data storage.
3. The neural network-based power grid load prediction method according to claim 2, wherein: the contents of the 7 sub-steps in the step 1 are respectively as follows:
step 1.1 data source selection: determining a data source for load prediction; the data source is real load data collected by a power grid operator, an energy company or other related institutions; the reliability of the data source and the time range of the data coverage are as comprehensive as possible in consideration of the accuracy of prediction;
step 1.2 data acquisition: acquiring historical load data from the selected data source; these data are typically in the form of a time series, including a date and a corresponding load value;
step 1.3, data cleaning and denoising: performing data cleaning operations such as abnormal value removal, missing value restoration, outlier processing and the like on the acquired data;
step 1.4 data interpolation: if the missing value exists in the data, filling the missing value by using an interpolation method; the interpolation method is linear interpolation, lagrange interpolation or spline interpolation
Step 1.5 data normalization: carrying out normalization processing on the load data to eliminate data differences under different scales; preferably, the following steps are adopted: min-max normalization or normalization;
step 1.6 data partitioning: dividing a complete historical load data set into a training set, a verification set and a test set;
step 1.7 data storage: the preprocessed data is saved to the appropriate data structure for subsequent model training and predictive use.
4. The neural network-based power grid load prediction method according to claim 1, wherein: the purpose of feature selection and extraction in step 2 is to determine features related to load prediction, reduce noise and redundant information, thereby improving the prediction capability of the model; the characteristic selection and extraction specifically comprises 5 substeps, which are sequentially as follows: feature selection, feature pretreatment, feature extraction, feature normalization, feature selection and extraction experiments.
5. The neural network-based power grid load prediction method according to claim 4, wherein: the details of each sub-step of the step 2 are as follows:
step 2.1, feature selection: features include load data over the past days or weeks, weather conditions, time;
step 2.2, pretreatment of characteristics: preprocessing the selected characteristics, specifically: smoothing, feature crossing and/or conversion;
step 2.3, feature extraction: extracting features from the original data by using a specific algorithm to reduce noise and redundant information and improve the prediction capability of the model;
step 2.4, feature normalization: normalizing the selected features to eliminate data differences under different scales; the common normalization method comprises minimum-maximum normalization, normalization and the like, and is similar to the data normalization in the step 1;
step 2.5 feature selection and extraction experiments: and evaluating the effect of feature selection and extraction by using the evaluation index, and repeatedly adjusting according to the experimental result to obtain better feature representation.
6. The neural network-based power grid load prediction method according to claim 1, wherein: the model construction and training in the step 3 comprises 5 substeps, which are sequentially as follows: model selection, model construction, model training, model evaluation and model optimization;
the model construction of power load prediction is the most core link in the whole prediction process; in the step, a model suitable for predicting load is constructed, and the model is trained, so that the model learns rules and features on a training set, and a foundation is provided for later prediction.
7. The neural network-based power grid load prediction method of claim 6, wherein: the model construction and training described in the step 3 specifically comprises:
step 3.1 model selection: selecting a model suitable for the load prediction problem; the load prediction model is as follows: linear regression, support vector machines, decision trees, neural networks, or time series models;
step 3.2, constructing a model: a fully connected neural network, a convolutional neural network, a cyclic neural network or an autoregressive moving average model is adopted to construct a load prediction model, and parameters and super parameters of the model are set;
step 3.3 model training: training and fitting the constructed model using historical load data: in the training process, the model needs to be optimized according to the loss function; the aforementioned loss functions include gradient descent, adam or adagard functions; in order to prevent overfitting, the training set is divided into a training set and a verification set, and the methods of cross verification and the like are adopted to perform model selection and parameter adjustment;
step 3.4 model evaluation: after model training is completed, the model needs to be evaluated to determine its predictive power and accuracy; the evaluation index may select a mean square error, a root mean square error and/or a mean absolute error; the smaller the evaluation index is, the stronger the prediction capability of the model is;
step 3.5, model optimization: according to the evaluation result and the actual demand, the model is adjusted and optimized; the optimization mode in the step comprises increasing the data quantity, selecting more proper characteristics and/or integer model parameters and the like, and the prediction results of different models can be tried to be integrated so as to improve the prediction accuracy of the whole system.
8. The neural network-based power grid load prediction method according to claim 1, wherein: the load prediction and result evaluation in the step 4 comprises data preparation, feature engineering, load prediction, result evaluation, model updating and iteration; and step 4, applying the trained model to an actual load prediction task, and evaluating a prediction result.
9. The neural network-based power grid load prediction method of claim 8, wherein: the specific steps of load prediction and result evaluation in the step 4 are as follows:
step 4.1 data preparation: preparing load data to be predicted: selecting as input either historical load data or real-time load data;
step 4.2, feature engineering: converting the raw load data into feature vectors suitable for model input,
step 4.3 load prediction: inputting the preprocessed load data into a trained model to perform load prediction; the model learns the change trend and mode of the load according to the historical data, and obtains a load prediction result in a future period of time through analysis and calculation of new data;
step 4.4 evaluation of results: evaluating the prediction result to measure the accuracy and the prediction capability of the model; the method can be compared with actual observation data, and evaluation indexes such as mean square error, root mean square error and average absolute error are calculated; the smaller the evaluation index is, the stronger the prediction capability of the model is;
step 4.5, updating and iterating the model: load data and demand may change over time, thus requiring periodic updating of the model and iteration; and (3) carrying out data preparation, model training and prediction evaluation again according to new data and problem demands, and continuously optimizing and improving the load prediction system.
10. The neural network-based power grid load prediction method of claim 8, wherein: the load prediction and result evaluation of step 4 further comprises a result visualization: visually displaying the prediction result so as to be more intuitively understood and analyzed; the prediction result and the actual observation data can be displayed in a line graph, a bar graph and other modes, and are compared and analyzed; the visualization can help the decision maker to better understand the load prediction results and make corresponding decisions and adjustments.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311593363.6A CN117578440A (en) | 2023-11-27 | 2023-11-27 | Power grid load prediction method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311593363.6A CN117578440A (en) | 2023-11-27 | 2023-11-27 | Power grid load prediction method based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117578440A true CN117578440A (en) | 2024-02-20 |
Family
ID=89893415
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311593363.6A Pending CN117578440A (en) | 2023-11-27 | 2023-11-27 | Power grid load prediction method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117578440A (en) |
-
2023
- 2023-11-27 CN CN202311593363.6A patent/CN117578440A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110766212B (en) | Ultra-short-term photovoltaic power prediction method for historical data missing electric field | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| CN107358366B (en) | Distribution transformer fault risk monitoring method and system | |
| CN108388962B (en) | Wind power prediction system and method | |
| CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
| CN113298297B (en) | Wind power output power prediction method based on isolated forest and WGAN network | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN112734128A (en) | 7-day power load peak value prediction method based on optimized RBF | |
| CN110659672B (en) | Method and device for predicting step-by-step uncertainty of output of wind turbine generator | |
| CN114035468B (en) | Method and system for predictively monitoring overhaul flow of fan based on XGBoost algorithm | |
| CN115293326A (en) | Training method and device of power load prediction model and power load prediction method | |
| CN112085285A (en) | Bus load prediction method and device, computer equipment and storage medium | |
| CN115438726A (en) | Device life and fault type prediction method and system based on digital twin technology | |
| CN115062864A (en) | Method and system for decomposing-integrating prediction of medium-term load of power distribution network | |
| CN113011680A (en) | Power load prediction method and system | |
| CN113569462A (en) | Distribution network fault level prediction method and system considering weather factors | |
| CN117013527A (en) | Distributed photovoltaic power generation power prediction method | |
| CN115238854A (en) | Short-term load prediction method based on TCN-LSTM-AM | |
| CN114155072A (en) | Financial prediction model construction method and system based on big data analysis | |
| CN117077532A (en) | Multi-model fusion method for life prediction of wind turbine generator | |
| CN108345996B (en) | System and method for reducing wind power assessment electric quantity | |
| CN111949640A (en) | Intelligent parameter adjusting method and system based on industrial big data | |
| CN112783865A (en) | Semi-supervised man-machine combined operation and maintenance fault library generation method and system | |
| CN117578440A (en) | Power grid load prediction method based on neural network | |
| CN111091292A (en) | Real-time risk dynamic modeling analysis system for nuclear power station |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |