Load prediction method and device for optimizing gray neural fusion model through genetic algorithm

Abstract

The invention discloses a load prediction method and device for optimizing a gray neural fusion model through a genetic algorithm, and solves the problems that a traditional expert experience prediction method depends on manpower and is difficult to deal with load complexity caused by new energy output, and a general prediction model does not fully integrate multi-source influence factors and cannot accurately capture a load rule. The method comprises the following steps that a training data set is obtained and preprocessed, and the training data set comprises power load data and meteorological data; constructing a gray neural fusion model according to the training data set, and optimizing the gray neural fusion model by adopting a genetic algorithm to obtain a load prediction model; performing verification analysis on the load prediction model to obtain an optimized load prediction model; and inputting to-be-predicted data into the trained optimized load prediction model, and performing load power prediction.

Description

Technical Field

[0001] This invention relates to power engineering technology, specifically to power grid load forecasting technology. Background Technology

[0002] Short-term power load forecasting is a core component of power system dispatching, and its accuracy directly determines the economy and security of power grid operation. However, power load is influenced by multiple factors, including weather conditions, holiday effects, and user behavior, exhibiting significant time-series nonlinearity, stochasticity, and strong time-varying characteristics. Traditional statistical methods (such as ARIMA) and classic machine learning models (such as support vector machines) are ill-suited to efficiently handle high-dimensional features and complex nonlinear relationships, and cannot accurately capture the dynamic changes in load, thus failing to meet current power grid demands.

[0003] Against this backdrop, Chinese invention application CN119849963A discloses a patent for "a short-term load forecasting method." This method integrates multiple variables such as power load, weather, and holidays, combines multimodal analysis to mine nonlinear load characteristics, and uses hyperparameter optimization technology to improve forecast accuracy, providing effective support for power grid dispatching decisions. However, this method also has limitations: the dual-network architecture and multimodal analysis require high computing power, leading to increased hardware costs. Furthermore, its forecasting performance in special scenarios such as extreme weather and sudden social events, as well as its generalization ability across regions and power consumption structures, still need further verification. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a load prediction method and device for optimizing a grey neural network fusion model using a genetic algorithm. This addresses the issues of traditional expert experience prediction methods relying on manual labor and being unable to cope with the load complexity caused by new energy output, as well as the problems of general prediction models not fully integrating multiple influencing factors and failing to accurately capture load patterns.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A short-term load prediction method for optimizing a grey neural network fusion model using genetic algorithms includes the following steps:

[0007] Acquire and preprocess the training dataset, which includes power load data and meteorological data;

[0008] Based on the training dataset, a grey neural network fusion model is constructed, and a genetic algorithm is used to optimize the grey neural network fusion model to obtain a load prediction model;

[0009] The load forecasting model was validated and analyzed to obtain an optimized load forecasting model;

[0010] The data to be predicted is input into the trained optimized load prediction model to predict load power.

[0011] Preferably, the meteorological data includes weather, daily minimum temperature, daily maximum temperature, daily average temperature, daily humidity, daily wind speed, daily wind force, daily air pressure, daily visibility, daily total precipitation, and daily average total cloud cover.

[0012] Preferably, the power load data includes photovoltaic power generation and wind power generation.

[0013] Preferably, the preprocessing of the training dataset includes normalization of the meteorological data, using the min-max normalization formula: Where X is the original temperature or original load value, , These are the minimum and maximum values ​​in the historical data, respectively. After normalization, the data range is mapped to [0,1].

[0014] Preferably, the method for constructing a grey neural network fusion model includes constructing a differential equation based on existing load data to obtain a prediction expression, and equivalently replacing the differential equation parameters with the neural network weights to form a GNNM with specified input and output.

[0015] Preferably, the method for constructing a grey neural network fusion model further includes using the initial weight threshold of GNNM as the GA "individual", using the "reciprocal of the root mean square error between the predicted value and the actual value" as the fitness function F, and obtaining the optimal initial parameters through GA operations.

[0016] Preferably, the validation analysis of the load forecasting model includes the following steps:

[0017] S1: Using the GA optimization results, set the GNNM parameters a, b and weights ω. 21 -ω 25 ω 31 -ω 35 The value;

[0018] S2: Classify data by season and weather, select similar samples by relative distance to build a training set, and use the test set to verify the accuracy after training.

[0019] S3: Input the time series data and calculate the output of the four-layer model.

[0020] S4: Calculate the error between the predicted and actual power, and adjust the weights and thresholds of each layer by combining the learning rate η and the number of iterations n.

[0021] S5: Training ends after the maximum number of iterations is reached.

[0022] In addition, a short-term load prediction device for a genetic algorithm-optimized grey neural network fusion model is also provided, comprising:

[0023] The data preprocessing module acquires the training dataset and performs preprocessing.

[0024] The load prediction model construction module constructs a grey neural network fusion model based on the training dataset, and optimizes the grey neural network fusion model using a genetic algorithm to obtain the load prediction model.

[0025] The load forecasting model optimization module verifies and analyzes the load forecasting model to obtain an optimized load forecasting model.

[0026] The load power prediction module inputs the data to be predicted into the trained optimized load prediction model to perform load power prediction.

[0027] Furthermore, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method.

[0028] Furthermore, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of the method described above.

[0029] This invention adopts the above-mentioned technical solution, which strengthens the coverage of the driving factors of load change by comprehensively incorporating multiple influencing factors; it adopts the genetic optimization grey neural network fusion (GA-GNNM) algorithm, which optimizes the initial parameters of the model by leveraging the global optimization capability of the genetic algorithm, thereby solving the problems of unstable results and easy local optima of the grey neural network model, and at the same time combining the quantitative characteristics of the grey model with the nonlinear fitting advantages of the neural network.

[0030] Therefore, it has the following beneficial effects:

[0031] 1. By integrating multiple influencing factors such as temperature, date type, and weather conditions, and combining them with the GA-GNNM algorithm to efficiently capture nonlinear load characteristics, the model's prediction accuracy far exceeds that of traditional expert experience methods, effectively solving the prediction bias problem caused by new energy output and multi-factor coupling.

[0032] 2. It does not require excessive reliance on human experience, reducing the professional requirements for forecasting personnel. Furthermore, the model has stronger adaptability, can flexibly cope with strong time-varying loads, and is easier to implement in real-world power grid scenarios.

[0033] 3. By relying on high-precision forecast results, we can reduce the waste of power generation resources or power shortages caused by load prediction deviations, lower grid dispatching costs and operational risks, and help the new power system achieve safe, stable and economical operation.

[0034] The specific technical solutions adopted in this invention and their beneficial effects will be disclosed in detail in the following specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a schematic diagram of the load prediction model structure in this invention; Figure 2 This is a diagram illustrating the application steps of the load prediction model in this invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of the present invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of the present invention.

[0037] Existing load forecasting methods still have some limitations, mainly including the following aspects:

[0038] The limitations of traditional expert experience-based forecasting methods: The short-term load forecasting methods currently relied upon by power dispatching departments require extremely high analytical skills and experience from forecasters. Furthermore, they cannot adapt to the intermittent and uncertain loads brought about by renewable energy output, and are unable to cope with increasingly complex load change patterns. The forecasting accuracy is far from meeting the needs of new power system construction.

[0039] The existing forecasting models suffer from insufficient adaptability and accuracy: general forecasting models do not fully integrate multiple influencing factors such as temperature, date type, and weather conditions, making it difficult to accurately capture the nonlinear variation of load and failing to provide reliable short-term load forecasting data support for power grid dispatching, development, and marketing departments.

[0040] Example 1

[0041] refer to Figure 1 and Figure 2 As shown, this invention proposes a short-term load prediction method for optimizing a grey neural network fusion model using a genetic algorithm, comprising the following steps:

[0042] Acquire and preprocess the training dataset, which includes power load data and meteorological data;

[0043] Based on the training dataset, a grey neural network fusion model is constructed, and a genetic algorithm is used to optimize the grey neural network fusion model to obtain a load prediction model;

[0044] The load forecasting model was validated and analyzed to obtain an optimized load forecasting model;

[0045] The data to be predicted is input into the trained optimized load prediction model to predict load power.

[0046] Electricity load data can be obtained from the power sector's main grid dispatch automation system and electricity consumption information collection system, including corresponding load data such as photovoltaic and wind power generation. Meteorological data can be obtained from meteorological data released by the meteorological bureau, including weather (cloudy), daily minimum temperature, daily maximum temperature, daily average temperature, daily humidity, daily wind speed, daily wind level, daily air pressure, daily visibility, daily total precipitation, and daily average total cloud cover.

[0047] The meteorological data was normalized using the min-max normalization formula: Where X is the original temperature or original load value, , These are the minimum and maximum values ​​in the historical data, respectively. After normalization, the data range is mapped to [0,1], which facilitates model calculation.

[0048] Load forecasting model design

[0049] Grey neural network (GM) model basics: Construct differential equations based on existing load data to obtain prediction expressions and achieve data quantification prediction.

[0050]

[0051]

[0052] Constructing a Grey Neural Fusion Model (GNNM): Integrating the parameters of the GM differential equation with the neural network weights (input - intermediate layer ω) 21 -ω 25 Intermediate-output layer ω 31 -ω 35 Equivalent substitution, combining the advantages of both, to form a GNNM with specified input and output.

[0053]

[0054]

[0055] i=(2,3,4,5)

[0056]

[0057]

[0058] GA algorithm optimizes GNNM: using the initial weight threshold of GNNM as the GA "individual", and using the "reciprocal of the root mean square error between the predicted value and the actual value" as the fitness function F, the optimal initial parameters are obtained through GA operation, which solves the problems of unstable GNNM results and easy local optima.

[0059]

[0060] Load forecasting model validation analysis

[0061] Step 1: Use the GA optimization results to set the GNNM parameters a, b and weights ω 21 -ω 25 ω 31 -ω 35 The value;

[0062] Step 2: Classify data by season and weather, select similar samples by relative distance to build a training set, and test the accuracy using the test set after training.

[0063] Step 3: Input the time series data and calculate the output of the four-layer model;

[0064] First layer:

[0065] Second layer:

[0066] Third layer: , i=(2,3,4,5)

[0067] Fourth layer:

[0068] Step 4: Calculate the error between the predicted and actual power, and adjust the weights and thresholds of each layer by combining the learning rate η and the number of iterations n.

[0069] Fourth layer error:

[0070] Third layer error:

[0071] Second layer error:

[0072] Adjust the connection weights from layer 2 to layer 3: ,

[0073] Where η is the learning rate; i = (1, 2…5); j = (2, 3…5); n is the number of iterations.

[0074] Adjust the weights from the first layer to the second layer:

[0075] Corrected threshold:

[0076] Step 5: Training ends after the maximum number of iterations is reached.

[0077] The above method enhances the coverage of load change drivers by comprehensively incorporating multiple influencing factors. It employs a Genetic Optimized Grey Neural Network Fusion (GA-GNNM) algorithm, leveraging the global optimization capability of genetic algorithms to optimize initial model parameters, thus addressing the instability and susceptibility to local optima issues inherent in grey neural network models. Simultaneously, it integrates the quantitative characteristics of grey models with the nonlinear fitting advantages of neural networks. These innovations significantly improve model accuracy compared to expert-based methods, providing highly reliable data for power grid dispatching, development, and marketing departments, thus meeting the operational needs of new power systems.

[0078] Example 2

[0079] Based on the short-term load prediction method for the genetic algorithm-optimized grey neural network fusion model described in Embodiment 1, this embodiment provides a short-term load prediction device for the genetic algorithm-optimized grey neural network fusion model, including:

[0080] The data preprocessing module acquires the training dataset and performs preprocessing.

[0081] The load prediction model construction module constructs a grey neural network fusion model based on the training dataset, and optimizes the grey neural network fusion model using a genetic algorithm to obtain the load prediction model.

[0082] The load forecasting model optimization module verifies and analyzes the load forecasting model to obtain an optimized load forecasting model.

[0083] The load power prediction module inputs the data to be predicted into the trained optimized load prediction model to perform load power prediction.

[0084] Example 3

[0085] This embodiment provides a computer device, including at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the short-term load forecasting method described in Embodiment 1.

[0086] Example 4

[0087] This embodiment provides a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the short-term load forecasting method described in Embodiment 1.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art should understand that the present invention includes, but is not limited to, the contents described in the accompanying drawings and the specific embodiments above. Any modifications that do not depart from the functional and structural principles of the present invention will be included within the scope of the claims.

Claims

1. A short-term load prediction method for a grey neural network fusion model optimized by a genetic algorithm, characterized in that, Includes the following steps: Acquire and preprocess the training dataset, which includes power load data and meteorological data; Based on the training dataset, a grey neural network fusion model is constructed, and a genetic algorithm is used to optimize the grey neural network fusion model to obtain a load prediction model; The load forecasting model was validated and analyzed to obtain an optimized load forecasting model; The data to be predicted is input into the trained optimized load prediction model to predict load power.

2. The short-term load forecasting method according to claim 1, characterized in that, The meteorological data includes weather, daily minimum temperature, daily maximum temperature, daily average temperature, daily humidity, daily wind speed, daily wind force, daily air pressure, daily visibility, daily total precipitation, and daily average total cloud cover.

3. The short-term load forecasting method according to claim 1, characterized in that, The power load data includes photovoltaic power generation and wind power generation.

4. The short-term load forecasting method according to claim 1, characterized in that, The preprocessing of the training dataset includes the normalization of meteorological data, using the min-max normalization formula: Where X is the original temperature or original load value, , These are the minimum and maximum values ​​in the historical data, respectively. After normalization, the data range is mapped to [0,1].

5. The short-term load forecasting method according to claim 1, characterized in that, The method for constructing a grey neural network fusion model includes constructing a differential equation based on existing load data to obtain a prediction expression, and then replacing the parameters of the differential equation with the weights of the neural network to form a GNNM with specified input and output.

6. The short-term load forecasting method according to claim 5, characterized in that, Another method for constructing a grey neural network fusion model is to use the initial weight threshold of GNNM as the GA "individual", use the "reciprocal of the root mean square error between the predicted value and the actual value" as the fitness function F, and obtain the optimal initial parameters through GA operation.

7. The short-term load forecasting method according to claim 1, characterized in that, The validation analysis of the load forecasting model includes the following steps: S1: Using the GA optimization results, set the GNNM parameters a, b and weights ω. 21 -ω 25 ω 31 -ω 35 The value; S2: Classify data by season and weather, select similar samples by relative distance to build a training set, and use the test set to verify the accuracy after training. S3: Input the time series data and calculate the output of the four-layer model. S4: Calculate the error between the predicted and actual power, and adjust the weights and thresholds of each layer by combining the learning rate η and the number of iterations n. S5: Training ends after the maximum number of iterations is reached.

8. A short-term load prediction device for a grey neural network fusion model optimized by a genetic algorithm, characterized in that, include: The data preprocessing module acquires the training dataset and performs preprocessing. The load prediction model construction module constructs a grey neural network fusion model based on the training dataset, and optimizes the grey neural network fusion model using a genetic algorithm to obtain the load prediction model. The load forecasting model optimization module verifies and analyzes the load forecasting model to obtain an optimized load forecasting model. The load power prediction module inputs the data to be predicted into the trained optimized load prediction model to perform load power prediction.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images