A method and system for predicting optical power based on cloud map feature extraction
Through the optical power prediction method and system based on cloud map feature extraction, the problem of insufficient ultra-short-term optical power prediction accuracy caused by the evolution and movement of the cloud is solved, and accurate prediction and adaptive adjustment of optical power are achieved, thereby improving power System stability and renewable energy utilization.
Patent Information
- Application Number
- CN202410420538.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-09
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-04-09
AI Technical Summary
The existing photovoltaic power prediction systems in photovoltaic power plants have shortcomings in terms of rapid response and high-precision prediction, especially when the proportion of new energy in the power grid increases and the requirements for water-optical complementarity are increased, it is difficult to meet the power grid's new energy power. Forecast demand.
Using optical power prediction methods and systems based on cloud map feature extraction, the accuracy of optical power is achieved by establishing observation data types, data stability analysis, cloud map feature extraction index, change prediction and optical power prediction models, combined with real-time observation data and cloud map data. Predictive and adaptive adjustment.
It improves the accuracy of ultra-short-term optical power prediction and the stable operation of the power system, ensures supply and demand balance, and improves the operating efficiency of the power system and the utilization rate of renewable energy.
Smart Images

Figure CN118428522B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data science and technology, specifically to the field of cloud computing and big data analysis technology, and in particular to a method and system for predicting optical power based on cloud image feature extraction. Background Art
[0002] At present, the optical power prediction systems of most photovoltaic power plants only aim to avoid being assessed by the power grid, and the prediction level of the power generation capacity of the site is difficult to meet the requirements of rapid response. As the proportion of renewable energy in the power grid increases, the power grid will put forward higher requirements for renewable energy power prediction. At the same time, the real-time joint dispatch of water and photovoltaic complementarity also puts forward higher requirements for renewable energy power prediction.
[0003] In this context, an optical power prediction method and system based on cloud image feature extraction came into being, aiming to solve the technical problem of insufficient accuracy of ultra-short-term optical power prediction caused by the growth, disappearance and movement of clouds by combining real-time observation datasets and cloud image features. Summary of the invention
[0004] The present application provides an optical power prediction method and system based on cloud image feature extraction, aiming to solve the technical problem of insufficient accuracy of ultra-short-term optical power prediction caused by the growth, disappearance and movement changes of clouds by combining real-time observation data sets and cloud image features.
[0005] In view of the above problems, the present application provides an optical power prediction method and system based on cloud map feature extraction.
[0006] The first aspect disclosed in the present application provides a method for optical power prediction based on cloud map feature extraction, the method comprising: establishing an observation data type, and performing data extraction based on the observation data type to obtain a real-time observation data set, and obtaining a prediction time span by performing data stability analysis on the real-time observation data set; establishing a cloud map feature extraction index, and performing feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features; using the prediction time span as a prediction constraint, performing change prediction on the real-time cloud map features based on the real-time observation data set to obtain a cloud map feature prediction result; pre-constructing an optical power prediction model, and synchronizing the cloud map feature prediction result to the optical power prediction model to perform optical power prediction to obtain associated optical power parameters; using the prediction time span as a data call constraint, calling to obtain an electric power sequence, and adaptively adjusting the output power of base load power generation according to the associated optical power parameters and the electric power sequence.
[0007] Another aspect disclosed in the present application provides an optical power prediction system based on cloud map feature extraction, the system comprising: a stability analysis module, the stability analysis module is used to establish an observation data type, and perform data extraction based on the observation data type to obtain a real-time observation data set, and obtain a prediction time span by performing data stability analysis on the real-time observation data set; a feature extraction module, the feature extraction module is used to establish a cloud map feature extraction index, and perform feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features; a change prediction module, the change prediction module is used to use the prediction time span as a prediction constraint, perform change prediction on the real-time cloud map features based on the real-time observation data set, and obtain cloud map feature prediction results; an optical power prediction module, the optical power prediction module is used to pre-build an optical power prediction model, and synchronize the cloud map feature prediction results to the optical power prediction model to perform optical power prediction, and obtain associated optical power parameters; an adaptive adjustment module, the adaptive adjustment module is used to use the prediction time span as a data call constraint, call to obtain an electric power sequence, and perform adaptive adjustment of the output power of base load power generation according to the associated optical power parameters and the electric power sequence.
[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0009] The above-mentioned optical power prediction method based on cloud map feature extraction establishes the types of observation data and accurately extracts real-time observation data based on these types. After that, by conducting in-depth analysis on these real-time data, the stability of the data is evaluated, and the appropriate time span for optical power prediction is determined accordingly. In order to further improve the accuracy of the prediction, a set of cloud map feature extraction indicators is established. Based on these indicators, key features are accurately extracted from the cloud map data obtained in real time. After determining the prediction time span, this is used as a constraint condition, and the real-time cloud map features are predicted to change in combination with the real-time observation data set to obtain the change trend of the cloud map features in the future. Then, an optical power prediction model is pre-built. After the cloud map feature prediction results are input into the model, the optical power parameters associated with the cloud map feature changes are obtained through calculation and analysis of the model. Finally, based on the prediction time span, the corresponding power consumption sequence data is called. According to the obtained optical power parameters and power consumption sequence, the output power of the base load power generation is adaptively adjusted to ensure the stable operation and supply and demand balance of the power system, and to improve the operation efficiency of the power system and the utilization rate of renewable energy.
[0010] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0012] Figure 1 A schematic diagram of a process of an optical power prediction method based on cloud image feature extraction in one embodiment;
[0013] Figure 2 The present invention is an architecture diagram of an optical power prediction system based on cloud image feature extraction in one embodiment.
[0014] Explanation of the reference numerals: stability analysis module 1, feature extraction module 2, change prediction module 3, optical power prediction module 4, adaptive adjustment module 5. DETAILED DESCRIPTION
[0015] The embodiments of the present application provide an optical power prediction method and system based on cloud image feature extraction to solve the technical problem of insufficient accuracy of ultra-short-term optical power prediction caused by the growth, disappearance and movement changes of clouds.
[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0017] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.
[0018] Embodiment 1
[0019] like Figure 1 As shown, the present application provides an optical power prediction method based on cloud image feature extraction, the method comprising:
[0020] Establishing an observation data type, and performing data extraction based on the observation data type to obtain a real-time observation data set, and performing data stability analysis on the real-time observation data set to obtain a prediction time span;
[0021] With the rapid development of renewable energy, photovoltaic power generation, as an important part of it, is gradually occupying an important position in the global energy structure. However, photovoltaic power generation is affected by many factors, among which the ground irradiance is significantly affected by the growth, disappearance and movement of clouds, which makes the photovoltaic output power have great uncertainty.
[0022] In the embodiment of the present application, in order to improve the accuracy of optical power prediction, it is necessary to establish observation data types and perform data extraction based on these types. The observation data type refers to various types of data that need to be collected for optical power prediction, including meteorological data, solar radiation data, cloud map data, etc. The prediction time span refers to the effective time range for cloud map prediction determined by the data stability analysis based on the real-time observation data set, ensuring a high prediction accuracy. The system terminal clarifies the purpose and target of the required analysis according to actual needs, and identifies the key factors affecting the observation target according to the determined purpose. In optical power prediction, it includes solar radiation intensity, atmospheric conditions, cloud thickness and type, etc. Afterwards, for each identified key factor, define the specific data type that needs to be collected, that is, the observation data index. For example, for the cloud layer, it is necessary to collect index data such as cloud amount, cloud height, and cloud temperature. Then, unify the units of the data to ensure that data from different sources can be compatible and comparable with each other. Then, summarize all the defined observation data indicators to form a clear list of data types. This list is the established observation data type. The system terminal extracts data based on this data type list, obtains the real-time values of these data from various sources, and forms a real-time observation data set. This data set is the key to subsequent analysis. After obtaining the real-time observation data set, the system terminal uses the stability evaluation function to perform data stability analysis on the real-time observation data set, evaluate the quality and change trend of the data, and determine whether its changes within a certain period of time are relatively stable and regular. And based on the stability analysis results combined with the preset stability threshold, the prediction time span is determined. The prediction time span is the result of data analysis, which represents whether the change trend of the cloud map in the future period can be predicted more accurately. This time span is determined based on the stability of the data, which is crucial for subsequent prediction work because it limits the effective range of the prediction and ensures the accuracy of the final optical power prediction.
[0023] Furthermore, the present application provides obtaining a prediction time span by performing data stability analysis on the real-time observation data set, and the method further includes:
[0024] The observation data type includes K observation data indicators, and data extraction is performed based on the K observation data indicators to obtain the real-time observation data set, wherein the real-time observation data set includes K observation data time series;
[0025] Synchronizing the K observation data time series to a pre-built stability evaluation function, and calculating an optimized stability index;
[0026] Preferably, the system terminal successfully establishes the observation data type according to the same method as above. This observation data type covers K different observation data indicators, which are intended to comprehensively reflect the various factors affecting optical power. Based on these K observation data indicators, the system terminal selects data that matches the required observation data type from the data source, and checks the integrity, accuracy and consistency of the data, thereby obtaining a real-time observation data set. This data set contains K observation data time series, each of which represents the change of the corresponding observation data indicator over time. In order to evaluate the stability of these observation data time series, the system terminal synchronizes them to a pre-built stability evaluation function. This function can calculate an optimized stability index based on factors such as the observation value at a certain time point in the time series data and the weight of the time point. The higher this index is, the better the stability of the observation data is, and the more conducive it is to accurate optical power prediction. In summary, this process is to first determine a variety of observation data indicators, and then extract the real-time data of these indicators to form a time series data set. Then, the stability index of these time series data is calculated by the stability evaluation function to evaluate the quality of the data and provide reliable data support for optical power prediction.
[0027] Pre-constructing a plurality of sample stability thresholds, wherein the plurality of sample stability thresholds correspond to a plurality of sample time spans;
[0028] The optimized stability index is used to traverse the multiple sample stability thresholds, and the sample time span corresponding to the sample stability threshold into which the optimized stability index falls is used as the predicted time span.
[0029] Preferably, the system terminal needs to determine a suitable prediction time span, that is, an effective time range for predicting future cloud map changes, during the optical power prediction process. To achieve this, the system terminal pre-builds a series of sample stability thresholds, which correspond to different sample time spans. These sample stability thresholds are set based on historical data and are used to evaluate the stability of data in different time spans. In, the optimized stability index calculated previously is used to traverse these sample stability thresholds. The optimized stability index reflects the stability of the real-time observation data set, and the higher the value, the more stable the data. The system terminal compares this index with each sample stability threshold to see which threshold range it falls within. Once the optimized stability index falls into the range corresponding to a certain sample stability threshold, the sample time span corresponding to this threshold is used as the prediction time span. This means that within this time span, the system terminal can relatively accurately predict the trend of cloud map changes based on the current observation data, thereby improving the accuracy and reliability of optical power prediction.
[0030] Furthermore, the present application provides the stability evaluation function, and the method further includes:
[0031] The stability evaluation function is as follows:
[0032]
[0033]
[0034] Among them, H m is the observed value at the mth time point in the observed data time series, H current is the observed value at the current time point, w m is the weight distribution at the mth time point, n is the number of observation values in the observation data time series, SI i is the stability index of the i-th observation data time series, w i is the weight allocation of the i-th observation data indicator, K is the number of observation data indicators, and OSI is the optimization stability index.
[0035] Optionally, the stability evaluation function is a mathematical tool for evaluating the stability of the observation data time series. It calculates a stability index by weighting the observation values at each time point in the observation data time series and combining the weight distribution of different observation data indicators. Specifically, H m Represents the observed value at the mth time point in the observed data time series. Here m is an index used to identify the specific position in the time series. current represents the observed value at the current time point. It serves as a comparison benchmark to help the system terminal calculate the difference between other points in the time series and the current point. mis the weight assigned to the mth time point. The weight is used to reflect the relative importance of the observations at different time points in the stability calculation. n represents the number of observations in the observation data time series. It is used for normalization calculation to ensure that the stability index is comparable between time series of different lengths. SI i is the stability index of the i-th observation data time series. This index reflects the stability of a single observation data time series. i is the weight distribution of the ith observation data indicator. It represents the relative contribution of different observation data indicators in the overall stability assessment. K represents the number of indicators of the observation data indicator. This is the total number of different observation data types, which is used to calculate the overall stability index. OSI is the optimized stability index, which is the output of the entire function. This index combines the stability of all observation data time series and indicators, and provides the system terminal with a quantitative assessment of the stability of the entire real-time observation data set. For each observation data time series, the function first calculates the difference between the observation value at each time point in the time series and the observation value at the current time point, multiplies it by the weight of the corresponding time point, and then accumulates these weighted differences. In addition, the system terminal determines the weight of the corresponding time point by the exponential decay method. First, the system terminal determines the decay coefficient. The choice of the decay coefficient is crucial for weight distribution. It determines the speed at which the weight decays with time distance. Smaller values will cause the weight to decay more slowly, while larger values will cause the weight to decay quickly. The system terminal sets specific values according to the actual application scenario and data volume, and adjusts them according to actual conditions. After that, the difference between each time point and the current time point is calculated to obtain the time distance corresponding to each time point. The closer to the current time, the higher the weight of the observation value, and vice versa. Then, the weight of each time point is calculated using the exponential decay weight distribution function and the decay coefficient. Then, the time distance of each time point is substituted into the weight distribution function to obtain the corresponding weight value. After obtaining the cumulative value, it is normalized by the number of observation values to obtain the stability index of the time series. Further, the function weights and sums the stability indexes of different observation data time series, where the weight distribution of each observation data indicator is determined according to its importance in the prediction. Finally, these weighted summed stability indexes are synthesized into an optimized stability index to reflect the overall stability of the entire real-time observation data set. Through the stability evaluation function, the system terminal can quantitatively evaluate the stability of the observation data time series, thereby providing a basis for determining the prediction time span. The higher the optimized stability index, the more stable the observation data is, and the more conducive it is to accurate cloud map prediction.
[0036] Establishing a cloud map feature extraction index, and performing feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features;
[0037] In one embodiment, in order to more accurately analyze cloud changes and predict weather patterns, it is necessary to establish cloud map feature extraction indicators, and perform feature extraction on cloud map data acquired in real time based on these indicators. The cloud map feature extraction indicators refer to a series of quantitative standards for extracting useful information from cloud maps, including cloud height features, cloud precipitation features, etc. The system terminal establishes cloud map feature extraction indicators in order to extract key information from complex cloud map data. These indicators cover multiple aspects such as cloud height, temperature, and shape, and together they constitute a set of standards for analyzing cloud maps. Based on these indicators, the system terminal constructs a feature extraction channel, and uses the constructed feature extraction channel to process the cloud map data acquired in real time, and automatically identifies specific features of the cloud map. In this way, the system terminal can quickly obtain the feature information of the real-time cloud map, providing strong data support for subsequent cloud map predictions.
[0038] Further, the present application provides a method for establishing a cloud map feature extraction index, and extracting features from interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features. The method also includes:
[0039] The cloud image feature extraction index includes a directly obtained feature set and an indirectly extracted feature set, wherein the directly obtained feature set includes cloud height features, cloud precipitation features and cloud top temperature features, and the indirect extraction feature set includes cloud amount features, cloud shape features and optical thickness features;
[0040] Constructing a direct feature extraction channel according to the directly obtained feature set, and constructing an indirect feature extraction channel according to the indirectly extracted feature set;
[0041] Preferably, the cloud image feature extraction index established by the system terminal includes two major types of features, namely, directly obtained feature set and indirectly extracted feature set. The directly obtained feature set includes intuitive feature information such as height features, cloud precipitation features, and cloud top temperature features, which can be directly read from meteorological equipment or satellite data. The indirect extraction feature set covers cloud amount features, cloud shape features, and optical thickness features, which need to be extracted from the cloud image through image analysis and processing. Among them, the cloud height feature refers to the height of the cloud from the ground. This feature is crucial for judging the type of cloud and analyzing the vertical structure of the atmosphere. Clouds of different heights are usually associated with different weather phenomena; cloud precipitation features describe whether clouds produce precipitation, including rain, snow, hail, etc. By analyzing the brightness, texture and structure in the cloud map, it can be inferred whether the cloud contains precipitation particles, and then the probability and intensity of precipitation can be predicted; cloud top temperature features refer to the temperature at the top of the cloud layer. This feature reflects the thermodynamic state of the cloud layer and is of great significance for judging the stability and development trend of the cloud; cloud amount features refer to the coverage of clouds in the sky. It reflects the number and distribution of clouds in the atmosphere. The amount of clouds directly affects meteorological elements such as sunshine; cloud shape features refer to the shape and structure of clouds. Different cloud shapes are associated with different weather phenomena, such as cirrus, cumulus, stratus, etc. By analyzing cloud shape features, the dynamic state of the atmosphere and the evolution trend of the weather system can be inferred; optical thickness features refer to the ability of clouds to absorb and scatter light, which reflects the light transmittance of clouds and their impact on radiation. In order to effectively extract these features, the system terminal constructs feature direct extraction channels and feature indirect extraction channels respectively. First, for each feature in the directly obtained feature set, feature mapping is performed to determine its specific form and position in the data, ensuring that the text data segments related to these features can be accurately found. According to the characteristics and extraction requirements of each feature, the corresponding algorithm is constructed for feature extraction. For example, for the cloud height feature, the cloud height data recognition algorithm is used to identify the height data in the cloud layer data segment and calculate its height; for the cloud precipitation feature, the cloud precipitation data recognition algorithm is used to identify the precipitation particle data in the cloud layer data segment. Based on the above steps, a feature extraction framework is designed. This framework contains multiple identifiers and can automatically extract all features in the directly obtained feature set from the input data. After that, the designed feature extraction framework is encapsulated into a channel, namely the feature direct extraction channel. Then, the system terminal uses a similar method to construct a feature indirect extraction channel. Among them, the direct extraction channel specifically processes those features that can be directly obtained, while the indirect extraction channel targets those features that require additional processing and analysis. Such a dual-channel setting not only improves the efficiency of cloud image feature extraction, but also ensures the accuracy of the extraction results. In summary, by setting up these two categories of features and two extraction channels, valuable feature information can be extracted from the cloud map more quickly and accurately, providing strong data support for the cloud map.
[0042] Integrate the direct feature extraction channel and the indirect feature extraction channel in parallel to generate a cloud map feature extraction network;
[0043] The cloud map feature extraction network is used to extract features from the real-time cloud map data to obtain the real-time cloud map features.
[0044] Optionally, in order to extract comprehensive feature information from real-time cloud image data, the system terminal integrates the feature direct extraction channel corresponding to the directly obtained feature set and the feature indirect extraction channel corresponding to the indirectly extracted feature set in parallel, and builds a complete cloud image feature extraction network through steps such as channel interface standardization and parallel processing design. This network combines direct and indirect feature extraction methods, and can more comprehensively reflect the multi-dimensional characteristics of the cloud image. Afterwards, the system terminal uses this integrated cloud image feature extraction network to process the cloud image data obtained in real time. The network will automatically extract direct features such as cloud height, cloud precipitation, cloud top temperature, and indirect features such as cloud amount, cloud shape, and optical thickness from the data according to the preset algorithms and processes. These feature information will be integrated together to form a comprehensive description of the real-time cloud image. In summary, by integrating direct and indirect feature extraction channels, a powerful cloud image feature extraction network is successfully constructed, which can quickly and accurately extract key features from real-time cloud image data, providing strong data support for the system terminal to understand the cloud state.
[0045] Furthermore, the present application provides a method for generating the cloud map feature extraction network, and the method also includes:
[0046] According to the feature composition of the directly obtained feature set, a text recognizer is configured to obtain a cloud height data recognizer, a cloud precipitation data recognizer and a cloud top temperature recognizer;
[0047] Integrate the cloud height data identifier, cloud precipitation data identifier and cloud top temperature identifier in parallel to generate the feature direct extraction channel;
[0048] Optionally, the system terminal conducts an in-depth analysis of the feature composition of the feature set directly obtained from the data to understand the specific meaning of each feature and the data representation form. Afterwards, a regular expression of the corresponding feature is written based on the analysis results, and a cloud height data identifier, a cloud precipitation data identifier, and a cloud top temperature identifier are built based on the written regular expressions. These identifiers are used to accurately identify and extract the corresponding feature information from the cloud image data. Then, the system terminal integrates these three identifiers in parallel using the same method as mentioned above to generate a direct feature extraction channel. This channel can run multiple identifiers at the same time and process the input cloud image data in parallel. When the cloud image data enters the channel, it will pass through the cloud height data identifier, cloud precipitation data identifier, and cloud top temperature identifier respectively, so as to quickly extract these three key features.
[0049] According to the feature composition of the indirect extraction feature set, a preset algorithm is assigned to obtain a cloud amount feature identifier, a cloud shape feature identifier and an optical thickness identifier, wherein the preset algorithm includes an edge detection algorithm, a shape matching algorithm and a remote sensing inversion algorithm;
[0050] interactively obtaining sample indirect extraction data, and respectively training the cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier based on the sample indirect extraction data;
[0051] Optionally, the system terminal assigns corresponding preset algorithms to the three features of cloud amount, cloud shape and optical thickness based on the feature composition of the indirect extraction feature set. Specifically, the cloud amount feature identifier uses an edge detection algorithm to identify the edge of the cloud layer and then estimate the cloud amount; the cloud shape feature identifier uses a shape matching algorithm to identify the shape characteristics of the cloud layer by matching with a preset cloud shape template; and the optical thickness identifier applies a remote sensing inversion algorithm to invert the optical thickness of the cloud layer by processing remote sensing data. In order to train these identifiers, the system terminal obtains sample indirect extraction data in an interactive manner. These data contain the true values of cloud amount, cloud shape and optical thickness, which are used to verify and optimize the performance of the identifier. Afterwards, the system terminal uses these sample data to train the cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier respectively. By continuously adjusting the algorithm parameters and optimizing the model structure, the accuracy and stability of the identifier are improved.
[0052] The trained cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier are integrated in parallel to generate the feature indirect extraction channel;
[0053] The direct feature extraction channel and the indirect feature extraction channel are integrated in parallel to generate the cloud map feature extraction network.
[0054] Optionally, after training, the system terminal successfully obtains a cloud amount feature identifier, a cloud shape feature identifier, and an optical thickness identifier. Afterwards, the three identifiers are integrated in parallel using the same method as described above to form a feature indirect extraction channel. This channel can efficiently and indirectly extract key features such as cloud amount, cloud shape, and optical thickness from cloud image data. At the same time, a feature direct extraction channel has been constructed before, which can directly extract direct features such as cloud height, cloud precipitation, and cloud top temperature from cloud image data. Then, the system terminal uses the same method to integrate this feature direct extraction channel and the feature indirect extraction channel just generated in parallel, thereby constructing a complete cloud image feature extraction network. This network combines the two methods of direct extraction and indirect extraction, and can comprehensively and accurately extract multiple features of cloud images. Through this network, the system terminal can understand and analyze cloud image data more deeply, laying a solid foundation for subsequent cloud image analysis and prediction.
[0055] Taking the prediction time span as a prediction constraint, predicting the changes of the real-time cloud map features based on the real-time observation data set, and obtaining a cloud map feature prediction result;
[0056] In one embodiment, the system terminal uses the obtained prediction time span as a constraint condition for the prediction, and synchronizes this prediction time span, the real-time observation data set, and the real-time cloud map features to the standard cloud map prediction network. This standard cloud map prediction network has been pre-trained and optimized, and can predict changes based on the input data. By synchronizing these data to the network, the system terminal can use the prediction ability of the network to predict the changes in the features previously extracted from the real-time cloud map within the future prediction time span. Through this process, the cloud map feature prediction results can be obtained, which can reflect the state and trend of the cloud map in the future period of time, and provide support for optical power prediction.
[0057] Furthermore, the present application provides a method for predicting changes of the real-time cloud map features based on the real-time observation data set with the prediction time span as a prediction constraint to obtain a cloud map feature prediction result, and the method further includes:
[0058] Interactively obtaining sample training data, wherein the sample training data includes multiple sample time spans of association mapping, multiple sample observation data sets, multiple sample cloud map original features, and multiple sample cloud map change features;
[0059] A standard cloud map prediction network is constructed based on a back propagation neural network, wherein the input data of the standard cloud map prediction network are the time span, the observation data set and the original features of the cloud map, and the output result is the cloud map prediction feature;
[0060] The sample training data is used to perform training of the cloud image prediction network until the output accuracy of the cloud image prediction network is higher than a preset value, thereby obtaining an optimized cloud image prediction network;
[0061] Preferably, in order to train the standard cloud map prediction network, the system terminal obtains sample training data by interacting with the historical database. These data include multiple sample time spans, multiple sample observation data sets, multiple sample cloud map original features and multiple sample cloud map change features. These sample data are all associated and mapped, and can accurately reflect the changes of cloud map features over time. Afterwards, the system terminal defines the structure of the standard cloud map prediction network, which includes determining the number of neurons in the input layer, hidden layer and output layer. The input layer is used to receive the time span, observation data set and cloud map original features as input data. The hidden layer is responsible for processing the input data, performing nonlinear transformations, and extracting useful features. The output layer outputs the cloud map prediction features as the result. The weights and bias items of the network are initialized again. These parameters are randomly generated and adjusted in the subsequent training process. Subsequently, the sample training data is input into the network for forward propagation. During the forward propagation process, the input data is calculated through each layer, and the output of the network is finally obtained. This output is compared with the sample cloud map change features to calculate the error. Then, back propagation is performed to calculate the error of the output layer, and the error of the hidden layer is calculated layer by layer using the chain rule. These errors are used to update the weights and bias items of the network, so that the output of the network is closer to the real cloud map change characteristics. Repeat the process of forward propagation and back propagation until the error is less than the preset value. During the training process, the system terminal uses the gradient descent algorithm to accelerate the training process and improve the prediction accuracy. Finally, when the network training is completed, the system terminal obtains the standard cloud map prediction network. This network can receive new time spans, observation data sets, and original features of cloud maps as inputs, and predict the corresponding cloud map features. In summary, rich sample training data is obtained through interaction, and a standard cloud map prediction network is constructed using the back propagation neural network. This network can receive time spans, observation data sets, and original features of cloud maps as inputs, and predict the future features of cloud maps. This provides a powerful tool for subsequent cloud map predictions.
[0062] The prediction time span, the real-time observation data set and the real-time cloud map features are used as input data and input into the optimized cloud map prediction network to perform cloud map change prediction to obtain the cloud map feature prediction result.
[0063] Preferably, in the process of cloud map change prediction, the system terminal uses the prediction time span, real-time observation data set and real-time cloud map features as key input data. These data contain important information required for prediction, such as the predicted time range, real-time observation results and current cloud map features. After that, these input data are synchronized to the optimized cloud map prediction network. After previous training and optimization, this network already has strong prediction capabilities and can accurately predict the future changes of the cloud map based on the input data. After the input data is calculated and processed by the network, the cloud map feature prediction results are output through the output layer. This result reflects the changing trend and status of the cloud map features within the future prediction time span.
[0064] Pre-constructing an optical power prediction model, and synchronizing the cloud map feature prediction result to the optical power prediction model to perform optical power prediction, and obtain associated optical power parameters;
[0065] In one embodiment, during the process of optical power prediction, the system terminal first pre-builds an optical power prediction model by calling a standard model set. The various models in this model set are trained based on a large amount of historical data and algorithms, and can predict changes in optical power based on input characteristic parameters. Afterwards, the cloud map feature prediction results previously obtained through the cloud map prediction network are synchronized to this optical power prediction model. When the cloud map feature prediction results are input into the optical power prediction model, the model will perform corresponding calculations and analyses based on these characteristic parameters, and finally obtain the associated optical power parameters. These parameters reflect the possible changes in optical power in a specific time period in the future.
[0066] Furthermore, the present application provides a pre-built optical power prediction model, and synchronizes the cloud map feature prediction result to the optical power prediction model to perform optical power prediction, and obtains associated optical power parameters. The method also includes:
[0067] Presetting an observation data deviation threshold, wherein the observation data deviation threshold includes K observation data deviation thresholds mapped to the K types of observation data indicators;
[0068] Generate K cloud graph feature call constraints according to the K observation data time series and the K observation data deviation thresholds;
[0069] Based on the K cloud image features, a constraint call is used to obtain a sample optical power record data set, wherein the sample optical power record data set includes multiple groups of historical instantaneous observation data, multiple groups of historical instantaneous cloud image features, and multiple groups of historical optical power records;
[0070] Optionally, in order to accurately predict the optical power, the system terminal first sets the observation data deviation threshold, which is determined based on K different observation data indicators, and each indicator has a corresponding deviation threshold. The role of these thresholds is to determine whether the real-time observation data is within the normal range to ensure the accuracy and reliability of the data. After that, the system terminal compares the K observation data time series in the real-time observation data with the corresponding deviation threshold to determine whether each observation data is within the normal range. For the observation data indicators that do not exceed the deviation threshold, the normal prediction process is maintained. For the observation data indicators that exceed the deviation threshold, specific cloud map feature call constraints are formulated according to the degree and direction of the deviation. These constraints include adjusting the feature extraction algorithm, increasing the feature extraction frequency, focusing on specific cloud map features, etc. These constraints are actually a kind of guidance and restriction on the cloud map feature extraction process, ensuring that the cloud map features can be accurately called and extracted under specific observation data deviation conditions. Finally, based on these K cloud map feature call constraints, the system terminal calls out a sample optical power record data set that meets this constraint from the historical database. This data set contains multiple sets of historical instantaneous observation data, multiple sets of historical instantaneous cloud map features, and multiple sets of historical optical power records. These data provide system terminals with rich historical information that can be used for further analysis and prediction, helping them to better understand the relationship between cloud map features and optical power, thereby optimizing the prediction and management of optical power.
[0071] Pre-constructing multiple prediction channels and optical power mean channels, wherein the multiple prediction channels are generated by randomly calling and building models based on a standard model set;
[0072] The sample optical power record data set is used to train the multiple prediction channels respectively, the multiple prediction channels that have been trained are arranged in parallel, and the output ends of the multiple prediction channels are connected to the input end of the optical power mean channel to complete the construction of the optical power prediction model;
[0073] Optionally, in order to build an optical power prediction model, the system terminal pre-sets multiple prediction channels and an optical power mean channel. These prediction channels are not fixed. The prediction channels are randomly called and built based on the standard model set. Such a design increases the diversity and flexibility of the model. The optical power mean channel first sets the data receiving format of its own input end according to the format of the output data of the prediction channel. The mean is calculated according to the pre-embedded mean calculation algorithm at the execution end and output by the output end. After the prediction channel and the optical power mean channel are built, the system terminal uses the previously prepared sample optical power record data set to train multiple prediction channels. This data set contains rich historical data, including observation data, cloud map features and optical power records, which provides strong data support for model training. When all prediction channels are trained, the system terminal sets them in parallel so that each channel can predict optical power independently. Then, we connect the output ends of these prediction channels to the input end of the optical power mean channel. The function of the optical power mean channel is to average the output results of multiple prediction channels to obtain a more stable and reliable optical power prediction value. In this way, the system terminal completes the construction of the optical power prediction model. This model not only has multiple prediction channels to provide diverse prediction results, but also integrates and optimizes the prediction results through the optical power mean channel, improving the accuracy and stability of the prediction. This design enables the model to better adapt and make accurate predictions when facing complex and changing environmental conditions.
[0074] The cloud map feature prediction results are synchronized to multiple prediction channels of the optical power prediction model to perform optical power prediction, and multiple one-sided optical power prediction results are obtained. The multiple one-sided optical power prediction results are synchronized to the optical power mean channel to perform mean calculation to obtain the associated optical power parameters.
[0075] Optionally, the system terminal synchronizes the cloud map feature prediction results to multiple prediction channels of the optical power prediction model, which is a key step in the optical power prediction process. Doing so can make full use of the cloud layer change information contained in the cloud map feature prediction results to predict the changes in optical power. Each prediction channel will process and analyze the cloud map feature prediction results based on its specific model structure and parameter settings to obtain its own optical power prediction results, that is, the one-sided optical power prediction results. Since different prediction channels use different prediction algorithms, there are certain differences in the prediction results they obtain. In order to obtain a more accurate and stable optical power prediction value, the system terminal further integrates these one-sided optical power prediction results and transmits them to the optical power mean channel. This channel adds the one-sided optical power prediction results of all received prediction channels and divides them by the number of one-sided optical power prediction results to eliminate the deviation of individual channels in the prediction results, thereby obtaining a more representative optical power prediction value, that is, the associated optical power parameter. In summary, this process is to apply the cloud map feature prediction results to the optical power prediction model, obtain the partial prediction results through multiple prediction channels, and use the optical power mean channel to integrate and optimize these results, and finally obtain accurate and reliable related optical power parameters. This design not only improves the accuracy of optical power prediction, but also enhances the robustness and adaptability of the model.
[0076] The predicted time span is used as a data call constraint to obtain an electric power sequence, and the output power of the base load power generation is adaptively adjusted according to the associated optical power parameter and the electric power sequence.
[0077] In one embodiment, the system terminal uses the predicted time span as a data call constraint to call and obtain the power sequence within the corresponding time range. This sequence reflects the user's power demand in a specific time period. Afterwards, the output power of the base load power generation is adaptively adjusted in combination with the associated optical power parameters previously obtained through the optical power prediction model. The purpose of adaptive adjustment is to dynamically adjust the output power of the power generation equipment according to the real-time optical power prediction results and power demand to better match the power supply and demand. By doing so, renewable energy can be used more effectively, dependence on traditional energy can be reduced, and the operating efficiency of the power grid can be optimized. In summary, this process is to dynamically adjust the output power of the power generation equipment according to the predicted time span and optical power parameters to meet the user's power demand and realize the intelligent management and optimized operation of the power grid.
[0078] In summary, the embodiments of the present application have at least the following technical effects:
[0079] The embodiment of the present application first establishes the observation data type, and determines the prediction time span by extracting the real-time observation data set and performing data stability analysis. At the same time, the real-time cloud map data is feature extracted by the cloud map feature extraction index to obtain the real-time cloud map features. Based on the real-time observation data set, the real-time cloud map features are predicted to change, and the cloud map feature prediction results are obtained. In this method, a real-time observation data set is generated by data extraction. By constructing a stability evaluation function and setting multiple sample stability thresholds, the optimized stability index is calculated to determine the prediction time span. In addition, through the text recognizer configuration and the preset algorithm allocation, a feature direct extraction channel and a feature indirect extraction channel are established. Based on the prediction time span, a standard cloud map prediction network is constructed by back propagation neural network, and the cloud map prediction network is trained and optimized to realize the change prediction of cloud map features. The optical power prediction model is pre-constructed, and the cloud map feature prediction results are synchronized to the optical power prediction model to obtain the associated optical power parameters. By presetting the observation data deviation threshold, calling the sample optical power record data set for training, constructing multiple prediction channels and optical power mean channels, the prediction of optical power and the acquisition of associated parameters are realized. This method achieves adaptive regulation of base load power output by comprehensively utilizing observation data, cloud map features, and optical power prediction models, improving prediction accuracy and system stability. These technical effects jointly solve the technical problem of insufficient ultra-short-term optical power prediction accuracy caused by the growth, disappearance, and movement of clouds, ensuring the safety and stability of the power grid.
[0080] Embodiment 2
[0081] Based on the same inventive concept as the optical power prediction method based on cloud image feature extraction in the aforementioned embodiment, Figure 2 As shown, the present application provides an optical power prediction system based on cloud image feature extraction, the system comprising:
[0082] Stability analysis module 1: The stability analysis module 1 is used to establish an observation data type, and perform data extraction based on the observation data type to obtain a real-time observation data set, and obtain a prediction time span by performing data stability analysis on the real-time observation data set;
[0083] Feature extraction module 2: the feature extraction module 2 is used to establish a cloud map feature extraction index, and perform feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features;
[0084] Change prediction module 3: the change prediction module 3 is used to predict the changes of the real-time cloud map features based on the real-time observation data set with the prediction time span as the prediction constraint, and obtain the cloud map feature prediction result;
[0085] Optical power prediction module 4: the optical power prediction module 4 is used to pre-build an optical power prediction model, and synchronize the cloud map feature prediction result to the optical power prediction model to perform optical power prediction, and obtain associated optical power parameters;
[0086] Adaptive adjustment module 5: The adaptive adjustment module 5 is used to use the predicted time span as a data call constraint, call to obtain the power consumption sequence, and adaptively adjust the output power of the base load power generation according to the associated optical power parameters and the power consumption sequence.
[0087] Furthermore, the stability analysis module 1 is used to perform the following method:
[0088] The observation data type includes K observation data indicators, and data extraction is performed based on the K observation data indicators to obtain the real-time observation data set, wherein the real-time observation data set includes K observation data time series;
[0089] Synchronizing the K observation data time series to a pre-built stability evaluation function, and calculating an optimized stability index;
[0090] Pre-constructing a plurality of sample stability thresholds, wherein the plurality of sample stability thresholds correspond to a plurality of sample time spans;
[0091] The optimized stability index is used to traverse the multiple sample stability thresholds, and the sample time span corresponding to the sample stability threshold into which the optimized stability index falls is used as the predicted time span.
[0092] Furthermore, the stability analysis module 1 is used to perform the following method:
[0093] The stability evaluation function is as follows:
[0094]
[0095]
[0096] Among them, H m is the observed value at the mth time point in the observed data time series, H current is the observed value at the current time point, w m is the weight distribution at the mth time point, n is the number of observation values in the observation data time series, SI i is the stability index of the i-th observation data time series, w i is the weight allocation of the i-th observation data indicator, K is the number of observation data indicators, and OSI is the optimization stability index.
[0097] Furthermore, the stability analysis module 1 is used to perform the following method:
[0098] Presetting an observation data deviation threshold, wherein the observation data deviation threshold includes K observation data deviation thresholds mapped to the K types of observation data indicators;
[0099] Generate K cloud graph feature call constraints according to the K observation data time series and the K observation data deviation thresholds;
[0100] Based on the K cloud image features, a constraint call is used to obtain a sample optical power record data set, wherein the sample optical power record data set includes multiple groups of historical instantaneous observation data, multiple groups of historical instantaneous cloud image features, and multiple groups of historical optical power records;
[0101] Pre-constructing multiple prediction channels and optical power mean channels, wherein the multiple prediction channels are generated by randomly calling and building models based on a standard model set;
[0102] The sample optical power record data set is used to train the multiple prediction channels respectively, the multiple prediction channels that have been trained are arranged in parallel, and the output ends of the multiple prediction channels are connected to the input end of the optical power mean channel to complete the construction of the optical power prediction model;
[0103] The cloud map feature prediction results are synchronized to multiple prediction channels of the optical power prediction model to perform optical power prediction, and multiple one-sided optical power prediction results are obtained. The multiple one-sided optical power prediction results are synchronized to the optical power mean channel to perform mean calculation to obtain the associated optical power parameters.
[0104] Furthermore, the feature extraction module 2 is used to perform the following method:
[0105] The cloud image feature extraction index includes a directly obtained feature set and an indirectly extracted feature set, wherein the directly obtained feature set includes cloud height features, cloud precipitation features and cloud top temperature features, and the indirect extraction feature set includes cloud amount features, cloud shape features and optical thickness features;
[0106] Constructing a direct feature extraction channel according to the directly obtained feature set, and constructing an indirect feature extraction channel according to the indirectly extracted feature set;
[0107] Integrate the direct feature extraction channel and the indirect feature extraction channel in parallel to generate a cloud map feature extraction network;
[0108] The cloud map feature extraction network is used to extract features from the real-time cloud map data to obtain the real-time cloud map features.
[0109] Furthermore, the feature extraction module 2 is used to perform the following method:
[0110] According to the feature composition of the directly obtained feature set, a text recognizer is configured to obtain a cloud height data recognizer, a cloud precipitation data recognizer and a cloud top temperature recognizer;
[0111] Integrate the cloud height data identifier, cloud precipitation data identifier and cloud top temperature identifier in parallel to generate the feature direct extraction channel;
[0112] According to the feature composition of the indirect extraction feature set, a preset algorithm is assigned to obtain a cloud amount feature identifier, a cloud shape feature identifier and an optical thickness identifier, wherein the preset algorithm includes an edge detection algorithm, a shape matching algorithm and a remote sensing inversion algorithm;
[0113] interactively obtaining sample indirect extraction data, and respectively training the cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier based on the sample indirect extraction data;
[0114] The trained cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier are integrated in parallel to generate the feature indirect extraction channel;
[0115] The direct feature extraction channel and the indirect feature extraction channel are integrated in parallel to generate the cloud map feature extraction network.
[0116] Furthermore, the change prediction module 3 is used to perform the following method:
[0117] Interactively obtaining sample training data, wherein the sample training data includes multiple sample time spans of association mapping, multiple sample observation data sets, multiple sample cloud map original features, and multiple sample cloud map change features;
[0118] A standard cloud map prediction network is constructed based on a back propagation neural network, wherein the input data of the standard cloud map prediction network are the time span, the observation data set and the original features of the cloud map, and the output result is the cloud map prediction feature;
[0119] The sample training data is used to perform training of the cloud image prediction network until the output accuracy of the cloud image prediction network is higher than a preset value, thereby obtaining an optimized cloud image prediction network;
[0120] The prediction time span, the real-time observation data set and the real-time cloud map features are used as input data and input into the optimized cloud map prediction network to perform cloud map change prediction to obtain the cloud map feature prediction result.
[0121] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0122] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
[0123] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.
Claims
1. A method for predicting optical power based on cloud image feature extraction, characterized in that: The method comprises: Establishing an observation data type, and performing data extraction based on the observation data type to obtain a real-time observation data set, and performing data stability analysis on the real-time observation data set to obtain a prediction time span; Establishing a cloud map feature extraction index, and performing feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features; Taking the prediction time span as a prediction constraint, predicting the changes of the real-time cloud map features based on the real-time observation data set, and obtaining a cloud map feature prediction result; Pre-constructing an optical power prediction model, and synchronizing the cloud map feature prediction result to the optical power prediction model to perform optical power prediction, and obtain associated optical power parameters; Taking the predicted time span as a data call constraint, calling to obtain an electric power sequence, and adaptively adjusting the output power of base load power generation according to the associated optical power parameter and the electric power sequence; Wherein, by performing data stability analysis on the real-time observation data set to obtain the prediction time span, the method further includes: The observation data type includes K observation data indicators, and data extraction is performed based on the K observation data indicators to obtain the real-time observation data set, wherein the real-time observation data set includes K observation data time series; Synchronizing the K observation data time series to a pre-built stability evaluation function, and calculating an optimized stability index; Pre-constructing a plurality of sample stability thresholds, wherein the plurality of sample stability thresholds correspond to a plurality of sample time spans; The optimized stability index is used to traverse the multiple sample stability thresholds, and the sample time span corresponding to the sample stability threshold into which the optimized stability index falls is used as the prediction time span; Wherein, the stability evaluation function is as follows: Among them, H m is the observed value at the mth time point in the observed data time series, H current is the observed value at the current time point, w m is the weight distribution at the mth time point, n is the number of observation values in the observation data time series, SI i is the stability index of the i-th observation data time series, w i is the weight distribution of the i-th observation data indicator, K is the number of observation data indicators, and OSI is the optimization stability index; Wherein, an optical power prediction model is pre-constructed, and the cloud map feature prediction result is synchronized to the optical power prediction model to perform optical power prediction to obtain associated optical power parameters. The method also includes: Presetting an observation data deviation threshold, wherein the observation data deviation threshold includes K observation data deviation thresholds mapped to the K types of observation data indicators; Generate K cloud graph feature call constraints according to the K observation data time series and the K observation data deviation thresholds; Based on the K cloud image features, a constraint call is used to obtain a sample optical power record data set, wherein the sample optical power record data set includes multiple groups of historical instantaneous observation data, multiple groups of historical instantaneous cloud image features, and multiple groups of historical optical power records; Pre-constructing multiple prediction channels and optical power mean channels, wherein the multiple prediction channels are generated by randomly calling and building models based on a standard model set; The sample optical power record data set is used to train the multiple prediction channels respectively, the multiple prediction channels that have been trained are arranged in parallel, and the output ends of the multiple prediction channels are connected to the input end of the optical power mean channel to complete the construction of the optical power prediction model; The cloud map feature prediction results are synchronized to multiple prediction channels of the optical power prediction model to perform optical power prediction, and multiple one-sided optical power prediction results are obtained. The multiple one-sided optical power prediction results are synchronized to the optical power mean channel to perform mean calculation to obtain the associated optical power parameters.
2. The method according to claim 1, characterized in that Establishing a cloud map feature extraction index, and performing feature extraction on the interactively obtained real-time cloud map data based on the cloud map feature extraction index to obtain real-time cloud map features, the method further includes: The cloud image feature extraction index includes a directly obtained feature set and an indirectly extracted feature set, wherein the directly obtained feature set includes cloud height features, cloud precipitation features and cloud top temperature features, and the indirect extraction feature set includes cloud amount features, cloud shape features and optical thickness features; Constructing a direct feature extraction channel according to the directly obtained feature set, and constructing an indirect feature extraction channel according to the indirectly extracted feature set; Integrate the direct feature extraction channel and the indirect feature extraction channel in parallel to generate a cloud map feature extraction network; The cloud map feature extraction network is used to extract features from the real-time cloud map data to obtain the real-time cloud map features.
3. The method according to claim 2, characterized in that The method further comprises: According to the feature composition of the directly obtained feature set, a text recognizer is configured to obtain a cloud height data recognizer, a cloud precipitation data recognizer and a cloud top temperature recognizer; Integrate the cloud height data identifier, cloud precipitation data identifier and cloud top temperature identifier in parallel to generate the feature direct extraction channel; According to the feature composition of the indirect extraction feature set, a preset algorithm is assigned to obtain a cloud amount feature identifier, a cloud shape feature identifier and an optical thickness identifier, wherein the feature composition of the indirect extraction feature set is three features of cloud amount, cloud shape and optical thickness, and the preset algorithm includes an edge detection algorithm, a shape matching algorithm and a remote sensing inversion algorithm; interactively obtaining sample indirect extraction data, and respectively training the cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier based on the sample indirect extraction data; Integrate the trained cloud amount feature identifier, cloud shape feature identifier and optical thickness identifier in parallel to generate the feature indirect extraction channel; The direct feature extraction channel and the indirect feature extraction channel are integrated in parallel to generate the cloud map feature extraction network.
4. The method according to claim 1, characterized in that Taking the prediction time span as a prediction constraint, performing change prediction on the real-time cloud map feature based on the real-time observation data set to obtain a cloud map feature prediction result, the method further includes: Interactively obtaining sample training data, wherein the sample training data includes multiple sample time spans of association mapping, multiple sample observation data sets, multiple sample cloud map original features, and multiple sample cloud map change features; A standard cloud map prediction network is constructed based on a back propagation neural network, wherein the input data of the standard cloud map prediction network are the time span, the observation data set and the original features of the cloud map, and the output result is the cloud map prediction feature; The sample training data is used to perform training of the cloud image prediction network until the output accuracy of the cloud image prediction network is higher than a preset value, thereby obtaining an optimized cloud image prediction network; The prediction time span, the real-time observation data set and the real-time cloud map features are used as input data and input into the optimized cloud map prediction network to perform cloud map change prediction to obtain the cloud map feature prediction result.
5. An optical power prediction system based on cloud image feature extraction, characterized in that: For executing the method according to any one of claims 1 to 4, the system comprises: Stability analysis module: establishes observation data types, and extracts data based on the observation data types to obtain real-time observation data sets, and obtains the prediction time span by performing data stability analysis on the real-time observation data sets; Feature extraction module: establish cloud map feature extraction indicators, and perform feature extraction on interactively acquired real-time cloud map data based on the cloud map feature extraction indicators to obtain real-time cloud map features; Change prediction module: taking the prediction time span as a prediction constraint, performing change prediction on the real-time cloud map features based on the real-time observation data set, and obtaining a cloud map feature prediction result; Optical power prediction module: pre-builds an optical power prediction model, and synchronizes the cloud map feature prediction result to the optical power prediction model to perform optical power prediction, and obtains associated optical power parameters; Adaptive adjustment module: using the predicted time span as a data call constraint, calling to obtain the power sequence, and adaptively adjusting the output power of the base load power generation according to the associated optical power parameters and the power sequence.
