Photovoltaic power generation output prediction method based on GRA-CNN-LSTM
Through gray correlation analysis and GRA-CNN-LSTM model, influencing factors are screened and photovoltaic output prediction is carried out, which solves the problem of low prediction accuracy in the existing technology, and achieves higher prediction accuracy and better evaluation indicators.
Patent Information
- Application Number
- CN202411391825.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-08
- Publication Date
- 2024-12-31
AI Technical Summary
The existing photovoltaic output prediction methods mainly focus on the changes in their own characteristics, ignore the influence of relevant factors, resulting in low prediction accuracy and difficult to meet the needs of accurate prediction.
The gray correlation analysis method was used to screen influencing factors, and a photovoltaic output prediction model based on GRA-CNN-LSTM was constructed, and data features were extracted through convolutional neural networks and predicted in combination with long and short-term memory networks.
It improves the accuracy of photovoltaic output prediction, makes up for the shortcomings of traditional methods, and provides higher prediction accuracy and better evaluation indicators.
Smart Images
Figure CN119231516A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power prediction, and more specifically to a photovoltaic power generation output prediction method based on GRA-CNN-LSTM. Background Art
[0002] The new energy industry has ushered in a great development in recent years. Against this background, photovoltaic power generation technology has ushered in a new breakthrough. By the end of 2023, the cumulative installed capacity of photovoltaic power generation in China will be about 610 million kilowatts, a year-on-year increase of 55.2%. Due to the high volatility and randomness of photovoltaic power generation, large-scale photovoltaic intervention has a great impact on the security of the power grid. Therefore, accurate prediction of photovoltaic output is crucial to ensure the safe and stable operation of the power system, improve the rationality of energy dispatch and energy utilization. It helps to optimize the operating efficiency of photovoltaic power stations and reduce energy waste. At the same time, it provides important data support for power grid dispatching and realizes the rational allocation and economic operation of power resources. In addition, accurate prediction can also help manage risks, prevent the adverse effects of uncertain factors, and promote the healthy development of the photovoltaic industry and technological progress.
[0003] However, at present, the traditional method for photovoltaic output prediction mainly focuses on its own characteristic change law and ignores the influence of related factors. The prediction accuracy is low and it is difficult to meet the needs of accurate prediction. Therefore, from the perspective of combination optimization, this patent takes the output power of photovoltaic stations as the prediction object, and uses gray correlation analysis to screen various possible influencing factors to obtain influencing factors with high correlation. On this basis, a photovoltaic output prediction model of CNN-LSTM is constructed to obtain a photovoltaic output prediction method based on GRA-CNN-LSTM. Summary of the invention
[0004] In view of the problems existing in photovoltaic output prediction mentioned in the background technology, the present invention proposes a photovoltaic station output prediction method based on GRA-CNN-LSTM, and the method specifically includes the following steps:
[0005] (1) Data collection and screening. Collect data on meteorological factors related to PV output, and use the grey correlation analysis method to calculate the grey correlation between the influencing factors and PV output, and select factors with higher influence as input data for the prediction model.
[0006] (2) Prediction model construction: Convolutional neural network (CNN) is used to extract data features, and then the processed data is input into long short-term neural network (LSTM), divided into training set and test set, and the training model is trained to achieve prediction accuracy.
[0007] (3) Data prediction: Input the data required for prediction into the trained prediction model to obtain the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Figure 1 This is a flow chart of photovoltaic output prediction based on GRA-CNN-LSTM;
[0009] Figure 2 This is a comparison chart of the test set effect;
[0010] Figure 3 This is a regression effect analysis chart. DETAILED DESCRIPTION
[0011] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0012] The preferred embodiments are described in detail below in conjunction with the accompanying drawings. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.
[0013] The specific steps of the present invention are as follows:
[0014] (3) Collection and screening of influencing factors
[0015] After a thorough analysis of the characteristics of photovoltaic power generation output, the factors that affect its efficiency and output can be summarized into three main categories:
[0016] Category 1: Environmental factors. Environmental factors are the primary factors affecting photovoltaic power generation, mainly including meteorological elements, such as total radiation irradiance, cloud cover, atmospheric temperature, relative humidity, diffuse irradiance, wind speed, and atmospheric pressure. These meteorological conditions have a direct and significant impact on the intensity and quality of light received by solar panels.
[0017] The second category: equipment factors. Equipment factors involve the physical and performance characteristics of PV modules, such as the tilt angle of the PV panels and the degradation of the modules over time. These factors determine the energy efficiency and performance of PV systems under different conditions.
[0018] The third category: human factors. Human factors include the site selection of power stations, the layout design of power generation equipment, and the accumulation of dust on the surface of batteries due to untimely cleaning and maintenance. Although these factors do not have a significant impact on photovoltaic power generation output as meteorological factors, they play a key role in the long-term operation and maintenance of power stations.
[0019] Since the changes in equipment and human factors are relatively limited, and it is difficult to achieve real-time monitoring and quantification, their impact on photovoltaic power generation output usually does not cause drastic fluctuations. Therefore, the research focus of this patented method on photovoltaic station output prediction is on meteorological factors. In the process of constructing a prediction model, if all meteorological factors are included in the model as input variables, it will undoubtedly greatly increase the computational burden of the model and may have a negative impact on the accuracy of the prediction results. In order to solve this problem, this study adopted the grey correlation analysis method to analyze and screen the influencing factors. The grey correlation calculation process is as follows:
[0020] The first step is to determine the analysis sequence. The model is divided into a reference sequence and a comparison sequence. The reference sequence is a data sequence that reflects the main behavioral characteristics of the system under study, and the comparison sequence is a data sequence composed of factors that will affect the behavioral characteristics of the system, represented by x0 and x1 respectively. i express.
[0021] The second step is dimensionless processing, the formula is:
[0022]
[0023] Among them, x represents an influencing factor, x min Indicates the minimum value in this influencing factor data, x max Indicates the maximum value of the influencing factor data, x′ i Represents the influencing factor data after dimensionless processing.
[0024] The third step is to calculate the grey correlation coefficient. i (k) is the correlation degree between the reference sequence and the comparison sequence at each moment, ρ is the resolution coefficient, and the general value range is (0,1). For the convenience of calculation, ρ=0.5 is usually taken.
[0025]
[0026] Δr=|x0(k)-x i (k)∣
[0027] In order to reflect the correlation between the comparison sequence and the reference sequence, the average value of all correlation coefficients of the comparison sequence is called the correlation coefficient r. 0i . Correlation r 0i The larger it is, the stronger the correlation is. The calculation formula is as follows:
[0028]
[0029] (4) Prediction model construction
[0030] Convolutional neural networks have the characteristics of local connection, weight sharing and aggregation. They can also deeply mine the characteristics of time series data while reducing calculations. Convolutional neural networks are mainly composed of convolutional layers, pooling layers, excitation layers and fully connected layers. The convolutional layer extracts local features, and the pooling layer performs feature selection and reduces the feature dimension to reduce the number of parameters. The main construction process of the convolutional neural network is as follows:
[0031] 1) The convolutional layer is used to process time series data. Suppose the meteorological feature generates a power data x at each time t t , whose data attenuation rate is ω k , after k-1 time steps, the data is the original ω k The convolution result of meteorological characteristic time series is:
[0032]
[0033] In the formula, ω k (k=1,2,···,K) is the filter (convolution kernel).
[0034] Based on convolution, the convolution kernel sliding step and zero padding are introduced to increase the diversity of convolution for flexible feature extraction.
[0035] 2) The result after equal-width convolution is called feature map. Although the convolution layer uses local connection to reduce the number of connections in the network, the number of neurons in the input feature map group of the pooling layer is not significantly reduced, and overfitting is prone to occur. Therefore, a pooling layer is added after the convolution layer to reduce the feature dimension and avoid overfitting. The input feature map group of the pooling layer divides each feature map into many regions. The pooling process is to downsample each region and obtain a value as a summary of the region. Commonly used pooling functions are maximum pooling and average pooling.
[0036] 3) The excitation layer performs nonlinear mapping on the output of the convolutional layer. In convolutional neural networks, the excitation is generally the ReLU function, which has the characteristics of fast convergence and simple gradient calculation.
[0037] 4) The fully connected layer maps the output feature sequence of the pooling layer into a column vector of a certain length to facilitate subsequent regression operations.
[0038] LSTM controls the transmission state through the gate structure, which is divided into forget gate, selective memory gate and output gate.
[0039] The forget gate determines the degree to which the previous unit state is forgotten. It uses the sigmoid function to determine the discarded useless information by reading and:
[0040] F t =σ(W f·[h t-1 ,x t ]+b f )
[0041] Where: F t is the state matrix of the forget gate at time t; h t-1 is the state matrix of the state unit at the previous moment; x t is the input value at time t; σ is the activation function; W f is the weight matrix of the forget gate; b f is the constant parameter matrix of the forget gate.
[0042] In the input determination stage, the sigmoid function is used to obtain the new information that needs to be determined. and input c t The ratio is:
[0043] i t =σ(W i ·[h t-1 ,x t ]+b i )
[0044]
[0045] Where: W i and b i ——weight matrix and bias vector of sigmoid layer; tanh——hyperbolic tangent function; W c and b c ——The weight matrix and bias vector of the tanh layer.
[0046] Output stage: c processed by tanh function t And the data classified by sigmoid function o t , passed to the hidden layer data value h at the next moment t , generate the output result.
[0047] o t =σ(W0·[h t-1 ,x t ]+b0)
[0048] h t =o t ·tanh(c t )
[0049] (3) Prediction based on the optimized model
[0050] GRA is used to analyze the factors affecting electricity load and find out the main meteorological factors that have a significant impact on photovoltaic output. On this basis, the historical data of the screened meteorological factors are used as independent variables, and the output power of photovoltaic stations is used as the dependent variable. The data is input into the CNN-LSTM model constructed in the previous article for training to obtain a load accuracy prediction model. Finally, the meteorological data to be predicted is input into the trained prediction model to obtain the prediction results.
[0051] The above are only preferred specific embodiments of the present invention. However, the protection scope of the present invention is not limited to these. Within the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by ordinary technicians familiar with the field are included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
[0052] Specific examples:
[0053] The example data comes from the "Jingyun" energy meteorological scientific research data service platform of the Power System Economic Operation Team of Shandong University. The meteorological and photovoltaic output data of a photovoltaic station from June 10 to June 19, 2023 were selected to construct an example to verify the effectiveness of the model.
[0054] There are six meteorological factors in the data set, including date, atmospheric pressure, humidity, cloudiness, temperature, wind speed, and irradiation intensity. The grey correlation method is used to measure the degree of influence of these meteorological factors on photovoltaic output. The grey correlation calculation results are as follows:
[0055]
[0056]
[0057] According to the calculation results, the top three factors that have a greater impact on photovoltaic output are selected as model input data.
[0058] The influencing factors and photovoltaic output data mentioned above are input into the CNN-LSTM model for training. The training results are as follows: Figure 2 , Figure 3 As shown:
[0059] To further demonstrate the superiority of the CNN-LSTM model constructed in this patent in terms of overall prediction accuracy, the following three models are compared:
[0060]
[0061] As can be seen from the table, the MAE of the GRA-CNN-LSTM model is significantly lower than that of the LSTM model and the CNN-LSTM model, which makes up for the shortcomings of the traditional LSTM model. For long input sequences, the prediction accuracy of the GRA-CNN-LSTM is higher than that of the traditional LSTM prediction model and the CNN-LSTM prediction model, and the evaluation index is better, which shows that the algorithm has good accuracy.
[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A photovoltaic station output prediction method based on GRA-CNN-LSTM, characterized by: The method specifically comprises the following steps: (1) Data collection and screening: Collect meteorological factor data related to PV output, and use the grey correlation analysis method to calculate the grey correlation between the influencing factors and PV output, and select factors with higher influence as input data for the prediction model; (2) Prediction model construction: Convolutional neural network (CNN) is used to extract data features, and then the processed data is input into long short-term neural network (LSTM), divided into training set and test set, and the training model is trained to achieve prediction accuracy; (3) Data prediction: Input the data required for prediction into the trained prediction model to obtain the prediction results.
2. The photovoltaic station output prediction method based on GRA-CNN-LSTM according to claim 1 is characterized in that: The collection and screening of influencing factors include: After a thorough analysis of the characteristics of photovoltaic power generation output, the factors that affect its efficiency and output can be summarized into three main categories: Category 1: Environmental factors; Environmental factors are the primary factors affecting photovoltaic power generation, mainly including meteorological elements, such as total radiation irradiance, cloud cover, atmospheric temperature, relative humidity, diffuse irradiance, wind speed and atmospheric pressure; these meteorological conditions have a direct and significant impact on the intensity and quality of light received by solar panels; The second category: equipment factors; equipment factors involve the physical and performance characteristics of PV modules, such as the tilt angle of the PV panels and the degradation of the modules over time; these factors determine the energy efficiency and performance of the PV system under different conditions; Category 3: Human factors; Human factors include the site selection of power stations, the layout design of power generation equipment, and the accumulation of dust on the surface of batteries due to untimely cleaning and maintenance. Although these factors do not have a significant impact on photovoltaic power generation output as meteorological factors, they play a key role in the long-term operation and maintenance of power stations. Since the changes of equipment and human factors are relatively limited and difficult to monitor and quantify in real time, their impact on photovoltaic power generation output usually does not cause drastic fluctuations; therefore, the research focus of this patent method on photovoltaic station output prediction is on meteorological factors; in the process of building a prediction model, if all meteorological factors are included in the model as input variables, it will undoubtedly greatly increase the calculation burden of the model and may have a negative impact on the accuracy of the prediction results; in order to solve this problem, this study uses the grey correlation analysis method to analyze and screen the influencing factors; the grey correlation calculation process is as follows: The first step is to determine the analysis sequence; the model is divided into a reference sequence and a comparison sequence. The reference sequence is a data sequence that reflects the main behavioral characteristics of the system under study, and the comparison sequence is a data sequence composed of factors that will affect the behavioral characteristics of the system, represented by x0 and x1 respectively. i express; The second step is dimensionless processing, the formula is: Among them, x represents an influencing factor, x min Indicates the minimum value in this influencing factor data, x max Indicates the maximum value of the influencing factor data, x′ i Represents the influencing factor data after dimensionless processing; The third step is to calculate the grey correlation coefficient ζ i (k) is the correlation between the reference sequence and the comparison sequence at each moment, ρ is the resolution coefficient, and the general value range is (0,1). For the convenience of calculation, ρ = 0.5 is usually taken; Δr=∣x0(k)-x i (k)∣ In order to reflect the correlation between the comparison sequence and the reference sequence, the average value of all correlation coefficients of the comparison sequence is called the correlation coefficient r. 0i ; Correlation r 0i The larger it is, the stronger the correlation is. The calculation formula is as follows:
3. The photovoltaic station output prediction method based on GRA-CNN-LSTM according to claim 1 is characterized in that: Prediction model building includes: Convolutional neural networks have the characteristics of local connection, weight sharing and aggregation. They can also deeply mine the characteristics of time series data while reducing calculations. Convolutional neural networks are mainly composed of convolutional layers, pooling layers, excitation layers and fully connected layers. The convolutional layer extracts local features, and the pooling layer performs feature selection and reduces the feature dimension to reduce the number of parameters. The main construction process of convolutional neural networks is as follows: 1) The convolutional layer is used to process time series data problems; suppose that the meteorological characteristics generate a power data x at each time t t , whose data attenuation rate is ω k , after k-1 time steps, the data is the original ω k times; the convolution result of meteorological characteristic time series is: In the formula, ω k (k=1,2,…,K) is the filter (convolution kernel); Based on convolution, the convolution kernel sliding step and zero padding are introduced to increase the diversity of convolution so as to facilitate flexible feature extraction; 2) The result after equal-width convolution is called feature map. Although the convolution layer uses local connection to reduce the number of connections in the network, the number of neurons in the input feature map group of the pooling layer is not significantly reduced, and overfitting is prone to occur. Therefore, a pooling layer is added after the convolution layer to reduce the feature dimension and avoid overfitting. The input feature map group of the pooling layer divides each feature map into many regions. The pooling process is to downsample each region and obtain a value as a summary of the region. Commonly used pooling functions include maximum pooling and average pooling. 3) The excitation layer performs nonlinear mapping on the output of the convolutional layer. In convolutional neural networks, the excitation is generally the ReLU function, which has the characteristics of fast convergence and simple gradient calculation. 4) The fully connected layer maps the output feature sequence of the pooling layer into a column vector of a certain length to facilitate subsequent regression operations; LSTM controls the transmission state through the gate structure, which is divided into forget gate, selective memory gate and output gate; The forget gate determines the degree to which the previous unit state is forgotten. It uses the sigmoid function to determine the discarded useless information by reading and: F t =σ(W f ·[h t-1 ,x t ]+b f ) Where: F t is the state matrix of the forget gate at time t; h t-1 is the state matrix of the state unit at the previous moment; x t is the input value at time t; σ is the activation function; W f is the weight matrix of the forget gate; b f is the constant parameter matrix of the forget gate; In the input determination stage, the sigmoid function is used to obtain the new information that needs to be determined. and input c t The ratio is: i t =σ(W i ·[h t-1 ,x t ]+b i ) Where: W i and b i ——weight matrix and bias vector of sigmoid layer; tanh——hyperbolic tangent function; W c and b c ——The weight matrix and bias vector of the tanh layer; Output stage: c processed by tanh function t And the data classified by sigmoid function o t , passed to the hidden layer data value h at the next moment t , generate output results; about t =σ(W0 [h t-1 ,x t ]+b0) h t =o t ·tanh(c t )。 4. The photovoltaic station output prediction method based on GRA-CNN-LSTM according to claim 1 is characterized in that: Predictions based on the optimized model include: GRA is used to analyze the factors affecting electricity load and find out the main meteorological factors that have a significant impact on photovoltaic output. On this basis, the historical data of the screened meteorological factors are used as independent variables, and the output power of photovoltaic stations is used as the dependent variable. The data is input into the CNN-LSTM model constructed in the previous article for training to obtain a load accuracy prediction model. Finally, the meteorological data to be predicted is input into the trained prediction model to obtain the prediction results.