Photovoltaic power generation prediction method and system based on fusion information

Abstract

The present invention discloses a photovoltaic power generation prediction method and system based on fusion information, which relates to the field of digital energy. The method comprises the following steps: S1: using a time convolutional neural network (TCN) to obtain corresponding future time series features based on historical time series photovoltaic power generation data; S2: using a convolutional neural network (ResNet-50) to extract environmental features based on environmental data at the same historical time and calculate self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data. The convolutional neural network (ResNet-50) is used to extract corresponding future environmental features F. k , assign the self-attention weight to the environment feature F k , get the weighted environmental feature F k W ; S3: Splicing and fusing time series features and environmental features with self-attention weight information F k W , obtain the fusion feature F (t,k) ; S4: Using LSTM network to combine features F (t,k) Obtain photovoltaic power generation prediction results.

Description

Technical Field

[0001] The present invention discloses a method and a system, which relate to the field of digital energy, and specifically to a photovoltaic power generation prediction method and system based on fusion information. Background Art

[0002] Digital energy is the premise and guarantee for the current digitalization of society, and it is also the basic guarantee for the survival and life of mankind. Human development is inseparable from energy, and the development and utilization of new energy is an essential guarantee for the current and future development of mankind. Solar energy is an inexhaustible clean energy. The rational use of solar energy is related to the development direction of the entire country and even the entire human race. At present, photovoltaic power generation has entered thousands of households. The three major applications of photovoltaic power generation are: (1) providing living electricity for residents in areas without electricity, including mobile power supplies and backup power supplies; (2) solar charging equipment, such as solar chargers and solar street lights; (3) grid-connected power generation, which is the application business scenario that various countries are currently competing to develop.

[0003] However, photovoltaic power generation is characterized by significant intermittency, randomness, and volatility. Large-scale photovoltaic power generation integration can impact the safe and stable operation of the power grid. Therefore, photovoltaic power generation forecasting is necessary. Currently, photovoltaic power generation forecasting relies on single time series or single environmental data, resulting in low accuracy. Summary of the Invention

[0004] This invention addresses the problems of the existing technology and provides a photovoltaic power generation prediction method and system based on fusion information to solve the problem of low photovoltaic power generation prediction accuracy and provide technical support for the rational production and utilization of energy.

[0005] The specific scheme proposed by the present invention is:

[0006] The present invention provides a photovoltaic power generation prediction method based on fusion information, the method steps are:

[0007] S1: Use the temporal convolutional neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data;

[0008] S2: Use the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculate self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data.

[0009] Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features

[0010] S3: Concatenating and fusing time series features with environmental features containing self-attention weight information Get the fusion feature F (t,k) ;

[0011] S4: Using LSTM network to combine the features F (t,k) Obtain photovoltaic power generation prediction results.

[0012] Furthermore, the step of obtaining the corresponding future time series features in S1 in the photovoltaic power generation prediction method based on fusion information includes:

[0013] The time series photovoltaic power generation data at intervals within a set time period is input into the temporal convolutional neural network TCN, and the time series features of the set corresponding future time period are obtained through the temporal convolutional neural network TCN.

[0014] Furthermore, the photovoltaic power generation prediction method based on fusion information described in S2 includes:

[0015] S21: Use the convolutional neural network ResNet-50 to extract historical environmental features using formula 1. Formula 1 is as follows: F i =Resnet50(X j )Formula 1

[0016] j∈(1,7), X j Indicates the numerical information of average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, longitude and latitude, F i Represents the environmental features extracted from the environmental data;

[0017] S22: According to environmental characteristics F i The self-attention weight is assigned through the self-attention mechanism using Formula 2, which is as follows:

[0018] W i n =SelfAttention(F1,…,F n ) Formula 2

[0019] W i n is the environmental feature F i The self-attention weight between

[0020] S23: Calculate the corresponding future environmental characteristics F using formula 1 k , the self-attention weight W i n Assigned to the corresponding future environmental characteristics F k , as shown in Formula 3:

[0021]

[0022] Get weighted environmental features

[0023] Furthermore, the photovoltaic power generation prediction method based on fusion information described in S3 splices and fuses the time series features and the environmental features with self-attention weight information As shown in Formula 4:

[0024]

[0025] Increase the feature dimension of the prediction.

[0026] Furthermore, in the photovoltaic power generation prediction method based on fusion information, the LSTM model is used in S4 to predict photovoltaic power generation using the LSTM network structure. The LSTM network structure is the first two layers of LSTM, the dropout layer and the fully connected layer. (t,k) After the LSTM network outputs the feature G with time series (t,k) , G (t,k) Input the fully connected network and get the prediction result.

[0027] The present invention also provides a photovoltaic power generation prediction system based on fusion information, including a time series feature extraction module, an environmental feature extraction module, a fusion module and a prediction module.

[0028] The time series feature extraction module uses the time convolution neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data;

[0029] The environmental feature extraction module uses the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculates self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data.

[0030] Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features

[0031] The fusion module splices and fuses time series features with environmental features with self-attention weight information Get the fusion feature F (t,k) ;

[0032] The prediction module uses the LSTM network to calculate the fusion feature F (t,k) Obtain photovoltaic power generation prediction results.

[0033] Furthermore, the time series feature extraction module in the photovoltaic power generation prediction system based on fusion information obtains the corresponding future time series features, including:

[0034] The time series photovoltaic power generation data at intervals within a set time period is input into the temporal convolutional neural network TCN, and the time series features of the set corresponding future time period are obtained through the temporal convolutional neural network TCN.

[0035] Furthermore, the environmental feature extraction module in the photovoltaic power generation prediction system based on fusion information includes:

[0036] The convolutional neural network ResNet-50 is used to extract historical environmental features using Formula 1, which is as follows: i =Resnet50(X j )Formula 1

[0037] j∈(1,7), X j Indicates the numerical information of average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, longitude and latitude, F i Represents the environmental features extracted from the environmental data;

[0038] According to environmental characteristics F i The self-attention weight is assigned through the self-attention mechanism using Formula 2, which is as follows:

[0039] W i n =SelfAttention(F1,…,F n ) Formula 2

[0040] W i n is the environmental feature F i The self-attention weight between

[0041] Use formula 1 to calculate the corresponding future environmental characteristics F k , the self-attention weight W i n Assigned to the corresponding future environmental characteristics F k , as shown in Formula 3:

[0042]

[0043] Get weighted environmental features

[0044] Furthermore, in the photovoltaic power generation prediction system based on fusion information, the fusion module splices and fuses the time series features and the environmental features with self-attention weight information. As shown in Formula 4:

[0045]

[0046] Increase the feature dimension of the prediction.

[0047] Furthermore, in the photovoltaic power generation prediction system based on fusion information, the prediction module uses the LSTM model to use the LSTM network structure to predict photovoltaic power generation. The LSTM network structure is the first two layers of LSTM, the dropout layer and the fully connected layer. (t,k) After the LSTM network outputs the feature G with time series (t,k) , G (t,k) Input the fully connected network and get the prediction result.

[0048] The benefits of the present invention are:

[0049] The present invention provides a photovoltaic power generation prediction method based on fused information, which provides technical support for the rational production and utilization of energy, solves the problem that the power generation prediction in the photovoltaic power generation field is affected by multiple factors and is difficult to predict accurately. It integrates time domain information with environmental information to predict photovoltaic power generation, obtains relevant information in a timely and accurate manner among the changing influencing factors, and thus obtains more accurate photovoltaic power generation prediction data, thereby improving the accuracy of photovoltaic power generation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 It is a schematic diagram of the application process of the method of the present invention.

[0052] Figure 2 It is a schematic diagram of extracting timing characteristics by the method of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0054] The present invention provides a photovoltaic power generation prediction method based on fusion information, the method steps are:

[0055] S1: Use the temporal convolutional neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data;

[0056] S2: Use the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculate self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data.

[0057] Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features

[0058] S3: Concatenating and fusing time series features with environmental features containing self-attention weight information Get the fusion feature F (t,k) ;

[0059] S4: Using LSTM network to combine the features F (t,k) Obtain photovoltaic power generation prediction results.

[0060] Photovoltaic power generation predictions rely on both time series information and spatial environmental characteristics. Photovoltaic power generation is closely related to the surrounding climate and geographic environment. Solar irradiance is a key indicator directly impacting photovoltaic power generation. This irradiance is significantly affected by the surrounding environment, including rainfall, air particle size, temperature, wind speed, and wind direction, all of which directly impact photovoltaic power generation efficiency. The method of the present invention utilizes time-domain information—time series features—and environmental information—external environmental characteristics—to extract and fuse features from different domains, achieving more accurate predictions.

[0061] In specific applications, in some embodiments of the method of the present invention, based on the technical solution of the method of the present invention, the step S1 of obtaining corresponding future time series features includes:

[0062] Input the time series photovoltaic power generation data at intervals within the set time period into the time convolution neural network TCN, and obtain the time series features of the set corresponding future time period through the time convolution neural network TCN. For example, input the time series photovoltaic power generation data every 15 minutes in the same season of the previous two years into the time convolution neural network TCN, and obtain the time series features of the set corresponding future time period through the time convolution neural network TCN, refer to Figure 2 In the process, the long-term dependency information is extracted through the TCN network, and the future 96 feature information is obtained, which is recorded as F t .

[0063] The temporal convolutional neural network (TCN) can predict and judge data at future points in time based on the historical order of a known sequence. Furthermore, TCN utilizes causal convolution, and its architecture can map sequences of arbitrary length to output sequences of the same length. It can construct residual structures and dilated convolutions to construct very long, effective historical information, enabling predictions using historical information from further back in time. Photovoltaic power generation forecasts exhibit cyclical variations based on daily and seasonal cycles, relying on relatively distant historical data. Therefore, the present invention utilizes TCN networks to extract time series features.

[0064] Furthermore, the photovoltaic power generation prediction method based on fusion information described in S2 includes:

[0065] S21: Use the convolutional neural network ResNet-50 to extract historical environmental features using formula 1. Formula 1 is as follows: F i =Resnet50(X j )Formula 1

[0066] j∈(1,7), X j Indicates the numerical information of average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, longitude and latitude, F i Represents the environmental features extracted from environmental data.

[0067] S22: According to environmental characteristics F i Use formula 2 to assign self-attention weights through the self-attention mechanism SelfAttention. Formula 2 is as follows:

[0068] W i n =SelfAttention(F1,…,F n ) Formula 2

[0069] W i n is the environmental feature F i The self-attention weight between

[0070] The self-attention mechanism SelfAttention actually assigns weights to each input item based on the interaction between the input items, that is, internal voting.

[0071] S23: Calculate the corresponding future environmental characteristics F using formula 1 k , the self-attention weight W i n Assigned to the corresponding future environmental characteristics F k , as shown in Formula 3:

[0072]

[0073] Get weighted environmental features Among them, F n It means the environmental characteristics information every 15 minutes in the same season in the previous two years, that is, F n Represents historical environmental characteristics. The historical basis is the data of the same season two years before the current time, and F k It is the future feature information of the current time. For example, if the current time is t, then n∈(tn,t-n+1,…t-1), and k∈(t+1,…). This can be a chronological relationship, or a relationship between the past and the future. The purpose of using historical information is to calculate the impact of various environmental characteristics on future environmental characteristics, such as the impact of rainfall in the past few days on future temperature, and to add appropriate weight information to various future environmental characteristics.

[0074] Furthermore, the splicing fusion of time series features and environmental features with self-attention weight information in S3 As shown in Formula 4:

[0075]

[0076] Increase the feature dimension of the prediction.

[0077] Furthermore, the LSTM model in S4 is used to predict photovoltaic power generation using an LSTM network structure. The LSTM network structure consists of the first two layers of LSTM, a dropout layer, and a fully connected layer.

[0078] Using Formula 5:

[0079] G (t,k) = LSTM(F (t,k) ) Formula 5

[0080] F (t,k) After the LSTM network outputs the feature G with time series (t,k) , G (t,k) Enter the fully connected network, as shown in Formula 6:

[0081] R (t,k) =fc(G) Formula 6

[0082] The prediction results are obtained to achieve the purpose of accurately predicting photovoltaic power generation by integrating features.

[0083] The present invention also provides a photovoltaic power generation prediction system based on fusion information, including a time series feature extraction module, an environmental feature extraction module, a fusion module and a prediction module.

[0084] The time series feature extraction module uses the time convolution neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data;

[0085] The environmental feature extraction module uses the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculates self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data.

[0086] Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features

[0087] The fusion module splices and fuses time series features with environmental features with self-attention weight information Get the fusion feature F (t,k) ;

[0088] The prediction module uses the LSTM network to calculate the fusion feature F (t,k) Obtain photovoltaic power generation prediction results.

[0089] The information interaction, execution process, etc. between the modules in the above system are based on the same concept as the embodiment of the method of the present invention. For specific contents, please refer to the description in the embodiment of the method of the present invention and will not be repeated here.

[0090] Similarly, the system of the present invention provides technical support for the rational production and utilization of energy, solves the problem that the prediction of power generation in the field of photovoltaic power generation is affected by various factors and is difficult to predict accurately, integrates time domain information with environmental information, predicts photovoltaic power generation, and obtains relevant information in a timely and accurate manner among the changing influencing factors, thereby obtaining more accurate photovoltaic power generation prediction data and improving the accuracy of photovoltaic power generation prediction.

[0091] It should be noted that not all steps and modules in the above-mentioned processes and system structures are required, and certain steps or modules can be omitted according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in the above-mentioned embodiments can be a physical structure or a logical structure, that is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or may be implemented by certain components in multiple independent devices.

[0092] The above embodiments are merely preferred embodiments for the purpose of fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are within the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims.

Claims

1. Photovoltaic power generation prediction method based on fusion information, which is characterized by The method steps are: S1: Use the temporal convolutional neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data; S2: Use the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculate self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data. Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features S3: Concatenating and fusing time series features with environmental features containing self-attention weight information Get the fusion feature F (t,k) ; S4: Using LSTM network to combine the features F (t,k) Obtain photovoltaic power generation prediction results.

2. The photovoltaic power generation prediction method based on fusion information according to claim 1 is characterized by The corresponding future time series features are obtained in S1, including: The time series photovoltaic power generation data at intervals within a set time period is input into the temporal convolutional neural network TCN, and the time series features of the set corresponding future time period are obtained through the temporal convolutional neural network TCN.

3. The photovoltaic power generation prediction method based on fusion information according to claim 1 is characterized by Said S2 includes: S21: Use the convolutional neural network ResNet-50 to extract historical environmental features using formula 1. Formula 1 is as follows: F i =Resnet50(X j )Formula 1 j∈(1,7), X j Indicates the numerical information of average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, longitude and latitude, F i Represents the environmental features extracted from the environmental data; S22: According to environmental characteristics F i The self-attention weight is assigned through the self-attention mechanism using Formula 2, which is as follows: W i n = SelfAttention(F1,…,F n ) Formula 2 W i n is the environmental feature F i The self-attention weight between S23: Calculate the corresponding future environmental characteristics F using formula 1 k , the self-attention weight W i n Assigned to the corresponding future environmental characteristics F k , as shown in Formula 3: Get weighted environmental features 4. The photovoltaic power generation prediction method based on fusion information according to claim 3 is characterized by The splicing and fusion of time series features and environmental features with self-attention weight information in S3 As shown in Formula 4: Increase the feature dimension of the prediction.

5. The photovoltaic power generation prediction method based on fusion information according to claim 3 or 4, characterized in that In S4, the LSTM model is used to predict photovoltaic power generation using the LSTM network structure. The LSTM network structure consists of the first two layers of LSTM, the dropout layer, and the fully connected layer. (t,k) After the LSTM network outputs the feature G with time series (t,k) , G (t,k) Input the fully connected network and get the prediction result.

6. Photovoltaic power generation prediction system based on fusion information, characterized by Including time series feature extraction module, environmental feature extraction module, fusion module and prediction module, The time series feature extraction module uses the time convolution neural network (TCN) to obtain the corresponding future time series features based on the historical time series photovoltaic power generation data; The environmental feature extraction module uses the convolutional neural network ResNet-50 to extract environmental features based on environmental data at the same historical time and calculates self-attention weights based on the environmental features. The environmental data includes average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, and longitude and latitude data. Use convolutional neural network ResNet-50 to extract the corresponding future environmental features F k , assign the self-attention weight to the environment feature F k , get the weighted environmental features The fusion module splices and fuses time series features with environmental features with self-attention weight information Get the fusion feature F (t,k) ; The prediction module uses the LSTM network to calculate the fusion feature F (t,k) Obtain photovoltaic power generation prediction results.

7. The photovoltaic power generation prediction system based on fusion information according to claim 6 is characterized by The time series feature extraction module obtains the corresponding future time series features, including: The time series photovoltaic power generation data at intervals within a set time period is input into the temporal convolutional neural network TCN, and the time series features of the set corresponding future time period are obtained through the temporal convolutional neural network TCN.

8. The photovoltaic power generation prediction system based on fusion information according to claim 6 is characterized by The environmental feature extraction module includes: The convolutional neural network ResNet-50 is used to extract historical environmental features using Formula 1, which is as follows: i =Resnet50(X j )Formula 1 j∈(1,7), X j Indicates the numerical information of average temperature, temperature difference, rainfall, wind speed, wind direction, irradiance, longitude and latitude, F i Represents the environmental features extracted from the environmental data; According to environmental characteristics F i The self-attention weight is assigned through the self-attention mechanism using Formula 2, which is as follows: W i n = SelfAttention(F1,…,F n ) Formula 2 W i n is the environmental feature F i The self-attention weight between Use formula 1 to calculate the corresponding future environmental characteristics F k , the self-attention weight W i n Assigned to the corresponding future environmental characteristics F k , as shown in Formula 3: Get weighted environmental features 9. The photovoltaic power generation prediction system based on fusion information according to claim 8 is characterized by The fusion module splices and fuses time series features and environmental features with self-attention weight information As shown in Formula 4: Increase the feature dimension of the prediction.

10. The photovoltaic power generation prediction system based on fusion information according to claim 8 or 9, characterized in that The prediction module uses the LSTM model to use the LSTM network structure to predict photovoltaic power generation. The LSTM network structure is composed of the first two layers of LSTM, the dropout layer and the fully connected layer. (t,k) After the LSTM network outputs the feature G with time series (t,k) , G (t,k) Input the fully connected network and get the prediction result.

Images