Method, system, device and medium for predicting short-term trends in direct solar radiation changes

Abstract

The application belongs to the technical field of new energy and energy saving, and discloses a short-term solar direct radiation change trend prediction method, system, device and medium, which comprises the following steps: acquiring data of each solar radiation feature of a current period; extracting a feature dimension topology vector and a time dimension topology vector of the solar radiation feature by a topology data analysis technology according to the data of each solar radiation feature of the current period; inputting the solar radiation feature data of the current period and the feature dimension topology vector and the time dimension topology vector of the solar radiation feature into a preset solar direct radiation trend prediction model based on LSTM to obtain a solar direct radiation change trend of a next period. The application overcomes the problem that the traditional DNI short-term prediction lacks deep mining of the correlation between features, mines the correlation between features based on a topology data analysis technology, and utilizes the time sequence prediction advantage of an LSTM deep neural network to improve the accuracy of short-term solar direct radiation change trend prediction.

Description

Technical Field

[0001] This invention belongs to the field of new energy and energy-saving technology, and relates to a method, system, equipment and medium for predicting short-term direct solar radiation variation trends. Background Technology

[0002] With increasingly tight fossil fuel supplies and environmental degradation, the vigorous development of renewable and clean energy, including solar energy, has become an international consensus. As a widely used renewable energy source, solar energy is experiencing rapid growth and holds a leading position in global energy consumption. Unlike photovoltaic power generation, solar thermal power plants concentrate the energy of direct solar radiation (DNI) onto a heat pump absorber through a condensation system to generate electricity. Changes in DNI significantly impact the operational reliability and power generation efficiency of solar thermal power plants. Therefore, accurate prediction of DNI not only contributes to the safe and stable operation of solar thermal power plants but also provides crucial data support for the resource allocation and optimized control of multi-energy complementary systems.

[0003] Based on the forecast time interval, DNI trend forecasts can be divided into long-term, medium-term, and short-term forecasts. Long-term forecasts generally predict solar radiation for the next few days based on numerical weather models, medium-term forecasts obtain solar radiation values ​​for the next few hours based on satellite imaging, and short-term forecasts predict changes in solar radiation for the next few minutes by analyzing cloud information from all-day infrared cloud images. However, predicting DNI from sky images presents many challenges. Cloud optical thickness is considered to be most relevant to DNI decay and is difficult to obtain directly. Furthermore, considering that cloud movement is often accompanied by its own formation and dissipation, this can lead to inaccurate cloud cover predictions. Moreover, capturing significant short-term changes in DNI within seconds to minutes is expensive and has limited availability. More importantly, for most regions, the proportion of cloudy days throughout the year is relatively small, resulting in less data to analyze. Therefore, it is difficult to predict DNI from the perspective of sky images.

[0004] Compared to image-based DNI prediction, time-series-based DNI prediction has been a focus of industry attention due to its large data volume and diverse methods. It is generally believed that the influence of all meteorological factors affecting DNI is integrated into the changes in DNI time series, meaning that learning the regularity of DNI time series can lead to more accurate DNI predictions. However, for short-term DNI prediction, the randomness of DNI changes caused by complex and variable climate conditions is difficult for general machine learning algorithms to capture. Although deep learning algorithms, represented by convolutional neural networks, have achieved some research results, they are still far from meeting practical application standards, resulting in inaccurate current short-term DNI predictions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing short-term DNI prediction results in the above-mentioned prior art, and to provide a method, system, device and medium for predicting the short-term direct solar radiation variation trend.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] In a first aspect, the present invention provides a method for predicting short-term trends in direct solar radiation, comprising:

[0008] Obtain data on the characteristics of solar radiation during the current time period;

[0009] Based on the solar radiation characteristics data for the current time period, topological data analysis techniques are used to extract the characteristic dimension topological vector and the time dimension topological vector of the solar radiation characteristics.

[0010] Input the solar radiation characteristic data of the current period, along with the feature dimension topology vector and time dimension topology vector of the solar radiation characteristics, into a preset LSTM-based solar direct radiation trend prediction model to obtain the solar direct radiation change trend for the next period.

[0011] Optionally, the solar radiation characteristics include total solar radiation intensity, direct solar radiation intensity, solar scattered intensity, average atmospheric pressure, relative humidity, wind direction, and wind speed.

[0012] Optionally, the feature dimension topology vector for extracting solar radiation features includes:

[0013] The data for each solar radiation feature are divided into segments with time window 'a' and time step 'b', and the segmented data for each solar radiation feature are organized according to the following formula to obtain sample data for several feature dimensions:

[0014] S = {S1, ..., S2} t S a}

[0015] Where S represents the feature dimension of the sample data, S t S represents the data of various solar radiation characteristics at time t. t ={s t，1 s t，2 , ..., s t，n , ..., s t，c}, S t，n Let be the data of the nth solar radiation feature at time t, and c be the total number of solar radiation features;

[0016] A feature persistence map is constructed based on sample data of several feature dimensions, and the feature persistence map is processed by the Persistence Landscape method to obtain the feature dimension topology vector of solar radiation features.

[0017] Optionally, the time-dimensional topological vector for extracting solar radiation features includes:

[0018] The data for each solar radiation feature are divided into time windows (a) and time steps (b), and the divided data are organized according to the following formula to obtain several time-dimensional sample data:

[0019] S′={S′1,...,S′ n , ..., S′ c}

[0020] Where S′ represents the time dimension sample data, S′ n Let S′ be the time subsequence of the nth solar radiation feature. n ={S 1，n , ..., s t，n , ..., s a，n}, S t，n Let be the data of the nth solar radiation feature at time t, and c be the total number of solar radiation features;

[0021] A time-duration map is constructed based on sample data from several time dimensions, and the time-duration map is processed using the Persistence Landscape method to obtain the time-dimensional topological vector of solar radiation characteristics.

[0022] Optionally, the LSTM-based solar direct radiation trend prediction model includes: a first DNN network, a second DNN network, a first feature joint network, several first fusion networks, and several second fusion networks.

[0023] The input to the first DNN network is the topological vector of the feature dimension of solar radiation characteristics;

[0024] The input to the second DNN network is a topological vector of the time dimension of solar radiation features;

[0025] The first fusion network includes c1 first LSTM networks, a second feature joint network, and a third DNN network. The inputs of the c1 first LSTM networks are data of c1 solar radiation features, respectively. The outputs of the first LSTM networks are sequentially connected to the second feature joint network and the third DNN network.

[0026] The second fusion network includes several sub-fusion networks, a third feature joint network, and a fourth DNN network; the sub-fusion network includes c2 second LSTM networks, a fourth feature joint network, and a fifth DNN network, the inputs of the c2 second LSTM networks are data of c2 solar radiation features respectively; the outputs of the c2 second LSTM networks are sequentially connected to the fourth feature joint network, the fifth DNN network, the third feature joint network, and the fourth DNN network;

[0027] The outputs of the first, second, third, and fourth DNN networks are all connected to the input of the first feature joint network.

[0028] The output of the first feature joint network is set to the sixth DNN network.

[0029] Optionally, the solar radiation characteristics include total solar radiation intensity, direct solar radiation intensity, solar scattered intensity, average atmospheric pressure, relative humidity, wind direction, and wind speed; one first fusion network and one second fusion network are each set, and three first LSTM networks are set. The inputs of the three first LSTM networks are the data of total solar radiation intensity, direct solar radiation intensity, and solar scattered intensity, respectively; two sub-fusion networks are set. The first sub-fusion network sets two second LSTM networks. The inputs of the two second LSTM networks of the first sub-fusion network are the data of average atmospheric pressure and relative humidity, respectively. The inputs of the two second LSTM networks of the second sub-fusion network are the data of wind direction and wind speed, respectively.

[0030] Optionally, the activation function of the LSTM-based solar direct radiation trend prediction model is a linear rectified function; the cost function of the LSTM-based solar direct radiation trend prediction model is a binary cross-entropy function; and the LSTM-based solar direct radiation trend prediction model uses classification accuracy, recall, and F1 score as evaluation metrics during training.

[0031] In a second aspect, the present invention provides a system for predicting short-term trends in direct solar radiation, comprising:

[0032] The data acquisition module is used to acquire data on various solar radiation characteristics during the current time period;

[0033] The data preprocessing module is used to extract the feature dimension topological vector and time dimension topological vector of solar radiation characteristics based on the data of various solar radiation characteristics in the current time period through topological data analysis technology.

[0034] The trend prediction module is used to input the solar radiation characteristic data of the current period, as well as the feature dimension topology vector and time dimension topology vector of the solar radiation characteristics, into a preset LSTM-based solar direct radiation trend prediction model to obtain the solar direct radiation change trend for the next period.

[0035] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting short-term direct solar radiation variation trends.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for predicting short-term direct solar radiation trends.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] This invention presents a method for predicting short-term direct solar radiation trends. Based on data of various solar radiation characteristics in the current time period, it extracts feature dimension topological vectors and time dimension topological vectors of the solar radiation characteristics using topological data analysis technology. These feature dimension and time dimension topological vectors, along with the current time period's solar radiation characteristic data, are then input into an LSTM-based direct solar radiation trend prediction model constructed based on prior knowledge. This method predicts the trend of direct solar radiation changes in the next time period. It overcomes the problem of traditional DNI short-term predictions lacking in-depth analysis of the relationships between features. By leveraging topological data analysis technology to deeply explore these relationships and utilizing the temporal predictive advantages of LSTM deep neural networks, the accuracy of short-term direct solar radiation trend prediction is significantly improved. Attached Figure Description

[0039] Figure 1 This is a flowchart of a method for predicting short-term direct solar radiation trends according to an embodiment of the present invention.

[0040] Figure 2 This is a feature dimension topology continuous graph according to an embodiment of the present invention.

[0041] Figure 3 This is a time-dimensional topology persistence graph according to an embodiment of the present invention.

[0042] Figure 4 This is a network framework diagram of the solar direct radiation trend prediction model based on LSTM, as described in an embodiment of the present invention.

[0043] Figure 5 This is a diagram of the LSTM network framework according to an embodiment of the present invention.

[0044] Figure 6 This is a diagram of the DNN network framework according to an embodiment of the present invention.

[0045] Figure 7 This is a block diagram of the short-term solar direct radiation variation trend prediction system according to an embodiment of the present invention. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0047] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0048] See Figure 1 In one embodiment of the present invention, a method for predicting the short-term trend of solar direct radiation is provided. The method utilizes topological data analysis (TDA) to obtain the topological vector representations of the time dimension and feature dimension of multidimensional time series data, and then constructs a solar direct radiation trend prediction model based on long short-term memory neural network (LSTM) that incorporates prior knowledge to predict the trend of solar direct radiation values.

[0049] Specifically, the method for predicting short-term direct solar radiation trends includes the following steps:

[0050] S1: Obtain data on the characteristics of solar radiation during the current time period.

[0051] S2: Based on the solar radiation characteristics data for the current time period, extract the feature dimension topological vector and the time dimension topological vector of the solar radiation characteristics through topological data analysis technology.

[0052] S3: Input the solar radiation characteristic data of the current period, as well as the feature dimension topology vector and time dimension topology vector of the solar radiation characteristics, into the preset LSTM-based solar direct radiation trend prediction model to obtain the solar direct radiation change trend of the next period.

[0053] Specifically, based on the solar radiation characteristics data for the current period, topological data analysis techniques are used to extract the feature dimension topological vector and the time dimension topological vector of the solar radiation characteristics. Then, the feature dimension topological vector and the time dimension topological vector of the solar radiation characteristics, along with the solar radiation characteristic data for the current period, are input into an LSTM-based solar direct radiation trend prediction model constructed based on prior knowledge, thereby predicting the trend of solar direct radiation changes in the next period.

[0054] In summary, the short-term solar direct radiation variation trend prediction method of this invention overcomes the problem of the lack of in-depth mining of the correlation between features in the traditional DNI short-term prediction. Based on topological data analysis technology, it deeply mines the correlation between features, and then utilizes the temporal prediction advantage of LSTM deep neural network, which greatly improves the accuracy of short-term solar direct radiation variation trend prediction.

[0055] In one possible implementation, the solar radiation characteristics mainly include seven characteristics, namely total solar radiation intensity (GHI), direct solar radiation intensity (DNI), solar diffuse intensity (DIF), average atmospheric pressure (AP), relative humidity (RH), wind direction (WD), and wind speed (WS).

[0056] In one possible implementation, the extraction of the feature dimension topology vector of solar radiation features includes: dividing the data of each solar radiation feature with a time window of a and a time step of b, and organizing the divided data of each solar radiation feature according to the following formula to obtain several feature dimension sample data:

[0057] S = {S1, ..., S2} t S a}

[0058] Where S represents the feature dimension of the sample data, S t S represents the data of various solar radiation characteristics at time t. t ={s t，1 s t，2 , ..., s t，n , ..., s t，c}, s t，n Let t be the data of the nth solar radiation feature at time t, and c be the total number of solar radiation features.

[0059] A feature persistence map is constructed based on sample data of several feature dimensions, and the feature persistence map is processed by the Persistence Landscape method to obtain the feature dimension topology vector of solar radiation features.

[0060] Among them, the Persistence Landscape method: Persistence Landscape, proposed by Peter Bubenik, is a topological feature representation method that has been widely applied in many fields. The principle of Persistence Landscape is to perform certain transformations on a persistent graph. The concept of a landscape is defined by considering a set of functions created by superimposing each point representing a birth-death pair p = (d, b) in the persistent graph, as shown below:

[0061]

[0062] The persistence landscape of a persistent graph is a set of functions.

[0063]

[0064] In this embodiment, the data for each solar radiation characteristic are first normalized. Then, the original time series is divided into seven feature series data with a length of 10, using a time window of 10 and a time step of 1. The predictor variable is the trend of solar direct radiation change in the next hour. A value of solar direct radiation that remains unchanged or increases is categorized as 1, and a value of solar direct radiation that decreases is categorized as -1.

[0065] For each sample in a multidimensional time series dataset, obtain the topological vector representation of the feature dimension. Assume a sample data point is represented as S = {S1, ..., S2}. t S 10}, where S t ={s t，1 s t，2 , ..., s t，7} represents the data for seven solar radiation characteristics at time t. See also Figure 2 First, a feature persistence graph is constructed, and then the topological vector representation of the feature dimension is obtained from the feature persistence graph, with the vector length chosen to be 50.

[0066] In one possible implementation, the extraction of the time-dimensional topological vector of solar radiation features includes: dividing the data of each solar radiation feature into segments with a time window of 'a' and a time step of 'b', and organizing the segmented data of each solar radiation feature according to the following formula to obtain several time-dimensional sample data:

[0067] S′={S′1,...,S′ n , ..., S′ c}

[0068] Where S′ represents the time dimension sample data, S′ n Let S′ be the time subsequence of the nth solar radiation feature. n ={s 1，n , ..., s t，n , ..., s a，n}, s t，n Let t be the data of the nth solar radiation feature at time t, and c be the total number of solar radiation features.

[0069] A time-duration map is constructed based on sample data from several time dimensions, and the time-duration map is processed using the Persistence Landscape method to obtain the time-dimensional topological vector of solar radiation characteristics.

[0070] In this embodiment, the data of each solar radiation feature are still normalized, and then the original time series is divided into data with a time window of 10 and a time step of 1, so as to obtain a feature sequence data containing seven lengths of 10.

[0071] For each sample in a multidimensional time series dataset, obtain the topological vector representation of the time dimension. Assume a sample data point is represented as S = {S1, ..., S2}. n S7, ..., S7}, where S n ={s 1，n s t，n , ..., s 10，n} represents the time subsequence of the nth feature. See also Figure 3 Construct a time-duration graph, and then obtain the topological vector representation of the time dimension from the time-duration graph, with the vector length set to 50.

[0072] In one possible implementation, the LSTM-based solar direct radiation trend prediction model includes: a first DNN network, a second DNN network, a first feature joint network, several first fusion networks, and several second fusion networks; the input of the first DNN network is the feature dimension topology vector of the solar radiation features; the input of the second DNN network is the time dimension topology vector of the solar radiation features; the first fusion network includes c1 first LSTM networks, a second feature joint network, and a third DNN network, the inputs of the c1 first LSTM networks are data of c1 solar radiation features respectively; the outputs of the first LSTM networks are sequentially connected to the second feature joint network and the third DNN network. The first DNN network comprises a second fusion network, a third feature joint network, and a fourth DNN network. The second fusion network includes several sub-fusion networks, a third feature joint network, and a fifth DNN network. The inputs of the c2 second LSTM networks are c2 solar radiation feature data. The outputs of the c2 second LSTM networks are sequentially connected to the fourth feature joint network, the fifth DNN network, the third feature joint network, and the fourth DNN network. The outputs of the first, second, third, and fourth DNN networks are all connected to the input of the first feature joint network. A sixth DNN network is configured at the output of the first feature joint network.

[0073] Among them, the feature joint network is a concatenate network, the DNN network consists of two DNN layers connected in sequence, and the LSTM network consists of two LSTM layers and two DNN layers connected in sequence.

[0074] See Figures 4 to 6 In this embodiment, one first fusion network and one second fusion network are each set up, and three first LSTM networks are set up. The inputs of the three first LSTM networks are data on total solar radiation intensity, direct solar radiation intensity, and solar scattering intensity, respectively. Two sub-fusion networks are set up. The first sub-fusion network sets up two second LSTM networks. The inputs of the two second LSTM networks of the first sub-fusion network are data on average atmospheric pressure and relative humidity, respectively. The inputs of the two second LSTM networks of the second sub-fusion network are data on wind direction and wind speed, respectively.

[0075] Specifically, inspired by how the human brain processes information in segments, we first divide the seven feature variables into three groups based on their physical meaning: total solar radiation intensity, direct solar radiation intensity, and solar scattering intensity; average atmospheric pressure and relative humidity; and wind direction and wind speed. Then, we use LSTM and DNN networks to extract features from each group for predicting the trend of changes in direct solar radiation.

[0076] The original data is first processed through a two-layer LSTM network to extract time-series features. Then, the extracted features are fused through a DNN network to obtain features within each group. Finally, the DNN network is used to fuse these features with the time-dimensional topological vector and the feature-dimensional topological vector to obtain fused features containing the correlation between various meteorological factors. These features are then used for short-term prediction of the trend of changes in direct solar radiation.

[0077] In one possible implementation, the activation function of the LSTM-based solar direct radiation trend prediction model is a linear rectified function; the cost function of the LSTM-based solar direct radiation trend prediction model is a binary cross-entropy function; and during the training process, the LSTM-based solar direct radiation trend prediction model uses classification accuracy, recall, and F1 score as evaluation metrics.

[0078] Specifically, the Rectified Linear Function (ReLU), commonly used in artificial neural networks, is selected as the activation function of the neural network, as shown below:

[0079] f(x) = max(0, x)

[0080] Choose the binary cross-entropy function as the cost function:

[0081]

[0082] Among them, y i These are real data tags. The data labels output by the model, where N is the size of a batch of data.

[0083] The evaluation metrics are classification accuracy, recall, and F1 score, and their specific calculation methods are as follows:

[0084] Assuming the confusion matrix is ​​as shown in Table 1, then:

[0085]

[0086]

[0087]

[0088]

[0089] Table 1 Confusion Matrix

[0090]

[0091] In one possible implementation, to verify the effectiveness of the method of the present invention, testing will be conducted on actual data. The data selected is hourly data from the Jinta County meteorological station in Gansu Province for the entire year of 2014. First, the raw data is processed to obtain topological vectors for the time series of seven meteorological factors, including feature dimension topology vectors and time dimension topology vectors. The original normalized data (or_data) is compared with data on which feature dimension topology vectors are added (TDA_data_feature), data on which time dimension topology vectors are added (TDA_data_time), and data on which both time dimension and feature dimension topology vectors are added (TDA_data_feature_time). The effectiveness of the added topological features is verified by comparing the classification performance of different classifiers (Random Forest (RF), k-Nearest Neighbors (KNN)). In the experiment, the training dataset contained 5000 data points, and the test set contained 3749 data points. The experimental results are shown in Table 2.

[0092] Table 2 Experimental Results

[0093]

[0094]

[0095] Experimental results show that the classification performance of all three classification models improved to varying degrees after adding topological feature vectors. This directly demonstrates that for meteorological time series, the presence of topological features in both the feature and time dimensions is helpful for predicting short-term direct solar radiation, indirectly verifying the effectiveness of the proposed method. By comparing the best results in Table 2 with those of TDA, LSTM, and deep neural networks (TDA-LSTM), we can see that TDA-LSTM consistently outperforms traditional machine learning methods in terms of ACC, recall, and F1 scores. Furthermore, the prediction results meet the requirement of the National Key Research and Development Program's "Energy Storage and Smart Grid Technology" key project: "Prediction accuracy of over 95% for provincial wind power / photovoltaic power within 4-24 hours." The experimental model has reached the application standard.

[0096] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0097] See Figure 7In another embodiment of the present invention, a short-term direct solar radiation variation trend prediction system is provided, which can be used to implement the above-mentioned short-term direct solar radiation variation trend prediction method. Specifically, the short-term direct solar radiation variation trend prediction system includes a data acquisition module, a data preprocessing module, and a trend prediction module. The data acquisition module is used to acquire data on various solar radiation characteristics in the current time period; the data preprocessing module is used to extract the feature dimension topological vector and time dimension topological vector of the solar radiation characteristics based on the data of various solar radiation characteristics in the current time period through topological data analysis technology; the trend prediction module is used to input the solar radiation characteristic data of the current time period, as well as the feature dimension topological vector and time dimension topological vector of the solar radiation characteristics, into a preset LSTM-based direct solar radiation trend prediction model to obtain the solar direct radiation variation trend for the next time period.

[0098] All relevant content of each step involved in the aforementioned embodiments of the short-term direct solar radiation variation trend prediction method can be referenced to the functional description of the corresponding functional module of the short-term direct solar radiation variation trend prediction system in the embodiments of the present invention, and will not be repeated here.

[0099] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0100] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a short-term solar direct radiation variation trend prediction method.

[0101] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the short-term direct solar radiation variation trend prediction method in the above embodiments.

[0102] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting short-term trends in direct solar radiation, characterized in that, include: Obtain data on the characteristics of solar radiation during the current time period; Based on the solar radiation characteristics data for the current time period, topological data analysis techniques are used to extract the characteristic dimension topological vector and the time dimension topological vector of the solar radiation characteristics. Input the solar radiation characteristic data of the current period, as well as the feature dimension topology vector and time dimension topology vector of the solar radiation characteristics, into the preset LSTM-based solar direct radiation trend prediction model to obtain the solar direct radiation change trend of the next period. The LSTM-based solar direct radiation trend prediction model includes: a first DNN network, a second DNN network, a first feature joint network, several first fusion networks, and several second fusion networks; The input to the first DNN network is the topological vector of the feature dimension of solar radiation characteristics; The input to the second DNN network is a topological vector of the time dimension of solar radiation features; The first fusion network includes c1 first LSTM networks, a second feature joint network, and a third DNN network. The inputs of the c1 first LSTM networks are data of c1 solar radiation features, respectively. The outputs of the first LSTM networks are sequentially connected to the second feature joint network and the third DNN network. The second fusion network includes several sub-fusion networks, a third feature joint network, and a fourth DNN network; the sub-fusion network includes c2 second LSTM networks, a fourth feature joint network, and a fifth DNN network, the inputs of the c2 second LSTM networks are data of c2 solar radiation features respectively; the outputs of the c2 second LSTM networks are sequentially connected to the fourth feature joint network, the fifth DNN network, the third feature joint network, and the fourth DNN network; The outputs of the first, second, third, and fourth DNN networks are all connected to the input of the first feature joint network. The output of the first feature joint network is set to the sixth DNN network.

2. The method for predicting short-term direct solar radiation trends according to claim 1, characterized in that, The solar radiation characteristics include total solar radiation intensity, direct solar radiation intensity, solar scattered intensity, average atmospheric pressure, relative humidity, wind direction, and wind speed.

3. The method for predicting short-term direct solar radiation trends according to claim 1, characterized in that, The feature dimension topology vector for extracting solar radiation features includes: Based on time window a The time step is b The data for each solar radiation characteristic are divided and organized according to the following formula to obtain sample data for several feature dimensions: in, S For feature dimension sample data, for t Data on solar radiation characteristics at various times. , for t The first moment n Data on solar radiation characteristics The total number of solar radiation characteristics; A feature persistence map is constructed based on sample data of several feature dimensions, and the feature persistence map is processed by the Persistence Landscape method to obtain the feature dimension topology vector of solar radiation features.

4. The method for predicting short-term direct solar radiation trends according to claim 1, characterized in that, The time-dimensional topological vector for extracting solar radiation features includes: Based on time window a The time step is b The data for each solar radiation characteristic are divided and organized according to the following formula to obtain sample data for several time dimensions: in, For time-dimension sample data, For the first n A time subsequence of solar radiation characteristics. , for t The first moment n Data on solar radiation characteristics The total number of solar radiation characteristics; A time-duration map is constructed based on sample data from several time dimensions, and the time-duration map is processed using the Persistence Landscape method to obtain the time-dimensional topological vector of solar radiation characteristics.

5. The method for predicting short-term direct solar radiation trends according to claim 1, characterized in that, The solar radiation characteristics include total solar radiation intensity, direct solar radiation intensity, solar scattered intensity, average atmospheric pressure, relative humidity, wind direction, and wind speed. One first fusion network and one second fusion network are each configured, and three first LSTM networks are configured. The inputs to the three first LSTM networks are data on total solar radiation intensity, direct solar radiation intensity, and solar scattered intensity, respectively. Two sub-fusion networks are configured. The first sub-fusion network uses two second LSTM networks. The inputs to the two second LSTM networks of the first sub-fusion network are data on average atmospheric pressure and relative humidity, respectively. The inputs to the two second LSTM networks of the second sub-fusion network are data on wind direction and wind speed, respectively.

6. The method for predicting short-term direct solar radiation trends according to claim 1, characterized in that, The activation function of the LSTM-based solar direct radiation trend prediction model is a linear rectified function; the cost function of the LSTM-based solar direct radiation trend prediction model is a binary cross-entropy function; and during the training process, the LSTM-based solar direct radiation trend prediction model uses classification accuracy, recall, and F1 score as evaluation metrics.

7. A short-term solar direct radiation variation trend prediction system, characterized in that, include: The data acquisition module is used to acquire data on various solar radiation characteristics during the current time period; The data preprocessing module is used to extract the feature dimension topological vector and time dimension topological vector of solar radiation characteristics based on the data of various solar radiation characteristics in the current time period through topological data analysis technology. The trend prediction module is used to input the solar radiation characteristic data of the current period, as well as the feature dimension topology vector and time dimension topology vector of the solar radiation characteristics, into a preset LSTM-based solar direct radiation trend prediction model to obtain the solar direct radiation change trend for the next period. The LSTM-based solar direct radiation trend prediction model includes: a first DNN network, a second DNN network, a first feature joint network, several first fusion networks, and several second fusion networks; The input to the first DNN network is the topological vector of the feature dimension of solar radiation characteristics; The input to the second DNN network is a topological vector of the time dimension of solar radiation features; The first fusion network includes c1 first LSTM networks, a second feature joint network, and a third DNN network. The inputs of the c1 first LSTM networks are data of c1 solar radiation features, respectively. The outputs of the first LSTM networks are sequentially connected to the second feature joint network and the third DNN network. The second fusion network includes several sub-fusion networks, a third feature joint network, and a fourth DNN network; the sub-fusion network includes c2 second LSTM networks, a fourth feature joint network, and a fifth DNN network, the inputs of the c2 second LSTM networks are data of c2 solar radiation features respectively; the outputs of the c2 second LSTM networks are sequentially connected to the fourth feature joint network, the fifth DNN network, the third feature joint network, and the fourth DNN network; The outputs of the first, second, third, and fourth DNN networks are all connected to the input of the first feature joint network. The output of the first feature joint network is set to the sixth DNN network.

8. 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 short-term direct solar radiation variation trend prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting short-term direct solar radiation variation trends as described in any one of claims 1 to 6.

Images