Wind and solar farm power prediction method based on multi-dimensional information perception and aggregation

Abstract

The present application relates to the field of renewable energy power generation technology, and provides a method for predicting the power of a wind and solar farm group under the convergence of multi-dimensional information perception. The method includes: collecting a multi-dimensional information perception data set; constructing a regional joint probability distribution model to analyze the joint distribution of wind energy and light energy in multiple regions; inputting each regional data set into the model, calculating the joint probability distribution of each pair of regions, and outputting a joint probability distribution set; dividing regions according to the joint probability distribution set, and outputting complementary and non-complementary regional sets; establishing complementary and non-complementary power prediction models respectively; loading the two models into the cloud platform to perform power prediction on the wind and solar farm group. The present application solves the technical problem of large power prediction errors caused by uneven distribution of wind energy and light energy between regions in a wind and solar farm group, and realizes the precise division of complementary and non-complementary regions through the convergence of multi-dimensional information perception data and the construction of a joint probability distribution model, thereby improving the accuracy and stability of power prediction for a wind and solar farm group.

Description

Technical Field

[0001] The present application relates to the field of energy technology, specifically to the field of new energy power generation technology, and in particular to a method for predicting wind and solar power plant group power under multi-dimensional information perception and convergence. Background Art

[0002] With the continuous growth of global energy demand and increasingly stringent environmental protection requirements, new energy power generation technology has become an important direction for future energy development. Wind energy and solar energy, as two major renewable energy sources, have been widely used due to their rich resources and environmental friendliness. However, since wind and solar resources are greatly affected by natural environmental factors and have obvious intermittent and volatile characteristics, there is great uncertainty and volatility in the power output of wind and solar power generation. This not only poses challenges to the stable operation of the power grid, but also affects the power generation efficiency and economy of the wind and solar farm group. The existing wind and solar power generation power prediction methods mainly rely on a single data source or limited information perception means, and cannot comprehensively and accurately obtain wind and solar energy data in the wind and solar farm group, resulting in insufficient power prediction accuracy, especially in the face of uneven distribution or rapid changes in wind and solar resources in multiple regions. The prediction error is more significant. In addition, the traditional power prediction model is difficult to fully consider the complementarity of wind and solar resources between different regions, and fails to effectively utilize the joint probability distribution characteristics between regions, further limiting the accuracy and applicability of the prediction model. Summary of the invention

[0003] This application aims to solve the technical problem of large power prediction errors caused by uneven distribution of wind energy and light energy between regions in a wind and solar farm group by providing a wind and solar farm group power prediction method under multi-dimensional information perception convergence.

[0004] In view of the above problems, the present application provides a wind and solar farm group power prediction method under multi-dimensional information perception and convergence.

[0005] The present application provides a method for predicting power of a wind-solar farm group under multi-dimensional information perception convergence, the method comprising: collecting a multi-dimensional information perception data set of each area in the wind-solar farm group through a cloud platform; constructing a regional joint probability distribution model, the regional joint probability distribution model is used to analyze the joint distribution of wind energy and light energy in multiple areas in the same time period; inputting the multi-dimensional information perception data set of each area into the regional joint probability distribution model, calculating the joint probability distribution of each pair of areas, and outputting a joint probability distribution set; performing regional division according to the joint probability distribution set, and outputting a complementary area set and a non-complementary area set; based on the complementary area set and the non-complementary area set, respectively establishing a complementary power prediction model and a non-complementary power prediction model; loading the complementary power prediction model and the non-complementary power prediction model into the cloud platform to perform power prediction on the wind-solar farm group.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above-mentioned wind and solar farm group power prediction method under the multi-dimensional information perception convergence, this method collects multi-dimensional perception data of each area of ​​the wind and solar farm group through the cloud platform, and constructs a joint probability distribution model to analyze the joint distribution of wind energy and light energy in each area in the same time period. After the data is input into the model, the joint probability distribution of each pair of regions is calculated, and the joint probability distribution set is output. Based on this, the regions are divided to obtain a set of complementary regions and non-complementary regions. Power prediction models are established for these two types of regions respectively, and finally the models are loaded into the cloud platform for power prediction of wind and solar farm groups. Through this method, the accuracy of power prediction can be effectively improved, the complementarity of wind and solar resources between regions can be fully utilized, the prediction error can be reduced, and the stability of the power grid can be improved.

[0008] 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

[0009] 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.

[0010] Figure 1 It is a schematic diagram of a flow chart of a method for predicting wind and solar farm group power under multi-dimensional information perception and convergence in one embodiment;

[0011] Figure 2 It is a schematic diagram of the power prediction process of a wind and solar farm group power prediction method under multi-dimensional information perception convergence in an embodiment. DETAILED DESCRIPTION

[0012] The embodiment of the present application solves the technical problem of large power prediction error caused by uneven distribution of wind energy and light energy between regions in a wind and solar farm group by providing a wind and solar farm group power prediction method under multi-dimensional information perception convergence.

[0013] 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.

[0014] 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.

[0015] Examples, such as Figure 1 As shown, the present application provides a method for predicting wind and solar farm group power under multi-dimensional information perception and convergence, the method comprising:

[0016] The multi-dimensional information perception data set of each area in the wind and solar farm group is collected through the cloud platform.

[0017] In an embodiment of the present application, the system terminal collects various types of information perception data from different areas in the wind and solar farm group through the cloud platform, and generates a multi-dimensional information perception data set. This multi-dimensional information perception data set includes multi-dimensional data related to wind energy and light energy, such as wind speed, light intensity, temperature, and humidity. In this process, the cloud platform aggregates real-time data sets from various regions for subsequent analysis and modeling. In this way, comprehensive and detailed wind and solar resource information can be obtained, laying the foundation for more accurate power prediction.

[0018] A regional joint probability distribution model is constructed, and the regional joint probability distribution model is used to analyze the joint distribution of wind energy and light energy in multiple regions within the same time period; the multidimensional information perception data set of each region is input into the regional joint probability distribution model, the joint probability distribution of each pair of regions is calculated, and the joint probability distribution set is output.

[0019] In one embodiment, in order to understand the joint distribution of wind energy and light energy in each area of ​​the wind and solar farm group in the same time period, the system terminal first determines the overall framework of the model. The framework includes an input layer, a core computing layer, and an output layer. The input layer is used to receive multi-dimensional information perception data of each area, the core computing layer is used to process data and calculate joint probability distribution, and the output layer is used to output a joint probability distribution set. Subsequently, the pre-constructed regional joint probability distribution calculation formula is encapsulated, and the encapsulated regional joint probability distribution calculation formula is integrated into the core computing layer of the regional joint probability distribution model. After the integration of the formula is completed, other parts of the regional joint probability distribution model are configured, such as parameter settings, calculation order, etc. Ensure that the regional joint probability distribution model can efficiently and accurately process the input data and output a joint probability distribution set. When the regional joint probability distribution model is constructed, the system terminal combines the multi-dimensional information perception data sets of each area in pairs to form multiple groups of regional pairs and corresponding data pairs, and uses these data pairs as input and imports them into the constructed regional joint probability distribution model. The regional joint probability distribution model receives the data pair of each region (such as region i and region j), and passes the data of this pair of regions to the core computing layer, and calculates the joint probability distribution between the two regions through the integrated regional joint probability distribution calculation formula. For each pair of regions in the wind and solar farm group, repeat the above calculation steps to generate the joint probability distribution of each pair of regions. All calculated joint probability distributions will form a joint probability distribution set. This set contains the joint distribution information of wind energy and light energy between different regions, reflects the relationship between them, helps to distinguish complementary and non-complementary regions, and provides data support for the power prediction model.

[0020] Further, the present application provides a method for calculating the joint probability distribution of each pair of regions, including:

[0021] Calculate the joint probability distribution of each pair of regions, expressed as: .

[0022] Preferably, in order to calculate the joint probability distribution of each pair of regions, the system terminal pre-constructs a regional joint probability distribution calculation formula, which is as follows: ; among them, among them, and They represent the wind energy resources in area i and area j respectively, such as wind speed, wind direction and other data. and They represent the light energy resources of area i and area j respectively, such as light intensity, solar radiation and other data. is the joint probability distribution of wind and solar energy in region i. It indicates the possibility of simultaneous occurrence of wind and solar resources in region i, and is calculated based on the wind and solar energy sensing data in the region. It represents the conditional probability distribution of the wind and solar resources in region j under the condition that the wind and solar resources in region i are known. Represents the joint probability distribution of wind energy and solar energy in region i and region j. It reflects how the wind and solar resources in the two regions are related to each other at the same time. This result can be used to analyze the resource complementarity or interdependence between different regions, thereby helping to optimize the power forecast of the entire wind and solar farm group.

[0023] Region division is performed according to the joint probability distribution set, and a complementary region set and a non-complementary region set are output.

[0024] In one embodiment, after obtaining the joint probability distribution set, the system terminal uses a pre-built complementarity calculation formula and combines the joint probability distribution set to calculate the complementarity index of each pair of regions. Subsequently, all regions are binary clustered according to the size of the complementarity index, that is, according to the preset complementarity index, the region with a higher complementarity index is selected as the complementary region set. These regions have complementary resources in different time periods, can support and schedule each other, and have a good coordination effect. For regions with a lower complementarity index, they are divided into non-complementary region sets. The resource distribution of these regions does not have an obvious complementary effect, and may be in a high resource or low resource state at the same time in the same time period, and resources cannot be effectively scheduled. Finally, the results of the complementary region set and the non-complementary region set will be output. The regions in the complementary region set can cooperate with each other in power scheduling, while the non-complementary region set needs to be processed separately in power prediction.

[0025] Through this process, it is possible to effectively identify which areas in the wind and solar farm group have complementary resources, thus providing a basis for power scheduling and forecasting.

[0026] Furthermore, the present application provides a method of performing region division according to the joint probability distribution set and outputting a complementary region set and a non-complementary region set, including:

[0027] According to the joint probability distribution set, the complementarity index of each pair of regions in the same time period is calculated, and a complementarity index set is output, wherein the expression for calculating each complementarity index is: ; Based on the complementarity index set, a complementarity matrix is ​​constructed; the complementarity matrix is ​​subjected to binary clustering, and a complementary region set and a non-complementary region set are output, wherein the complementary region set is a region having a complementarity index greater than or equal to a preset complementarity index, and the non-complementary region set is a region having a complementarity index less than the preset complementarity index.

[0028] Optionally, before calculating the complementarity index, the system terminal checks the joint probability distribution of each pair of region i and region j in the wind and solar farm group according to the joint probability distribution set, and evaluates the resource correlation between the two regions, that is, the impact of the wind and solar resources of region i on region j. Subsequently, the complementarity of region i and region j is further quantified according to the inter-regional correlation provided by the joint probability distribution. The quantification is performed according to the pre-constructed complementarity calculation formula, which is as follows: ;in, is the wind energy value of area i at time t, including wind speed or other wind energy related data. is the light energy value of region j at time t, including data related to light energy such as light intensity. T is the length of the time period, indicating the time range covered by the calculation of the complementarity index. is the complementarity index of region i and region j. This index is used to measure the complementarity of the two regions in terms of wind and solar energy resources, and its size is proportional to the complementarity. By calculating the complementarity index of each pair of regions through this complementarity calculation formula, a set of complementarity indexes can be obtained. Afterwards, the correction coefficient of each pair of regions is determined based on the joint probability distribution set, that is, the joint probability distribution of each pair of regions is added to 1. The correction coefficient of each pair of regions is then used to correct the complementarity index set so that the resource relationship between the two regions can be better reflected. The corrected complementarity indexes are then summarized to form a complementarity matrix, and each element in the matrix Represents the complementarity index of region i and region j. This matrix reflects the resource complementarity between all pairs of regions and can more accurately describe the resource distribution between regions. Then, a preset complementarity index is set according to business needs and data characteristics to distinguish between regions with strong complementarity and non-complementary regions. The elements in the complementarity matrix are then compared with the preset complementarity index, and binary clustering is performed to divide the elements in the complementarity matrix into a complementary region set and a non-complementary region set. Among them, the complementarity index in the complementary region set is greater than or equal to the preset complementarity index. The resources of these regions in different time periods can support and schedule each other, and the complementarity is strong. The complementarity index in the non-complementary region set is less than the preset complementarity index. The complementarity of resources between these regions is poor, and the resource change trends are similar. They may be in a high resource or low resource state at the same time and cannot be effectively complemented.

[0029] Further, the present application provides updating the complementarity index set, including:

[0030] The wind and solar field group is divided into time periods to obtain a plurality of characteristic time periods, wherein the plurality of characteristic time periods include daytime, nighttime, seasonal time periods and extreme weather time periods; a plurality of complementary indexes corresponding to the plurality of characteristic time periods for each pair of regions are obtained; and the plurality of complementary indexes are integrated to determine an updated complementary index for each pair of regions, so as to update the complementary index set.

[0031] Optionally, the system terminal divides the operating time of the wind and solar farm group to obtain multiple characteristic time periods. These characteristic time periods include daytime, nighttime, seasonal time periods and extreme weather time periods. Among them, the daytime time period is the time period with strong solar radiation during the day. The nighttime time period is the time period without solar radiation. The seasonal time period is the different seasons divided by quarter or month (such as spring, summer, autumn, and winter). The extreme weather time period is the time period under abnormal weather conditions such as strong winds, heavy rains, and strong sunshine. Each time period has unique meteorological conditions, which will affect the distribution of wind and solar resources, and then affect the resource complementarity between regions. Subsequently, the complementarity analysis of each pair of regions is performed for each characteristic time period. Specifically, in each characteristic time period, the complementarity index of region i and region j is calculated separately using the same method as described above. . Based on the wind and light energy data in the daytime period, the daytime regional complementarity index is calculated. Based on the wind energy data in the nighttime period, the nighttime regional complementarity index is calculated. According to the changes in wind and light energy data in different seasons, the seasonal regional complementarity index is calculated. Based on the wind and light energy data under extreme weather conditions, the extreme weather regional complementarity index is calculated. After obtaining the complementarity index of each pair of regions in each characteristic time period, the index is fused, that is, the weighted sum of these regional complementarity indexes is performed through the weight of each characteristic time period to determine the updated complementarity index of each pair of regions. The weight of each characteristic time period reflects the impact of different time periods on the final complementarity index, which is set according to the importance and frequency of occurrence of the characteristic time period. The updated complementarity index of each pair of regions can more comprehensively reflect the resource complementarity between regions in different time periods. These updated complementarity indexes are summarized into the complementarity index set to replace the original complementarity index, so as to provide a basis for subsequent regional division and power prediction.

[0032] Based on the complementary region set and the non-complementary region set, a complementary power prediction model and a non-complementary power prediction model are established respectively.

[0033] In one embodiment, the system terminal confirms which areas belong to complementary areas and which areas belong to non-complementary areas based on the complementary area set and the non-complementary area set, and obtains the historical multi-dimensional information perception data of these areas. Subsequently, the training set is divided according to the historical multi-dimensional information perception data of the complementary area, and the complementary power prediction model is trained based on the divided training set and the long short-term memory network. The goal of this complementary power prediction model is to smooth the fluctuation of power output and improve the accuracy and stability of power prediction on the basis of making full use of the resource complementarity between complementary areas. Similarly, the system terminal divides the training set of the historical multi-dimensional information perception data of the non-complementary area, and trains the non-complementary power prediction model based on the divided training set and the long short-term memory network. The goal of this non-complementary power prediction model is to independently predict the power output of each non-complementary area, focusing on capturing the resource trends and potential fluctuations of each area. By training the complementary power prediction model and the non-complementary power prediction model, it is possible to better cope with the complex resource distribution and power output changes in the wind and solar field group, and improve the accuracy of the overall power prediction.

[0034] Further, the present application provides a method for establishing a complementary power prediction model, including:

[0035] Based on the complementary mapping relationship of the complementary region set, a first training data set and identification samples for identifying power sizes, and a second training data set and identification samples for identifying power sizes are determined, wherein the first training data set and the second training data set are in a complementary mapping relationship; an initialized power prediction model is constructed using a long short-term memory network; the initialized power prediction model is trained according to the first training data set and identification samples for identifying power sizes, and the second training data set and identification samples for identifying power sizes, and a converged complementary power prediction model is output.

[0036] Optionally, the system terminal establishes a mapping relationship based on the resource complementarity of each area in the complementary area set. The complementary mapping relationship reflects how the resources between different areas complement each other in a time period. For example, when the wind energy in a certain area is low, the light energy in another area is high, and the two complement each other. Then, the wind energy and light energy data of these areas in different time periods are sorted out from the historical multi-dimensional information perception data of the complementary area. These data will be used to establish a complementary mapping relationship to reflect the complementarity of different areas when resources fluctuate. Subsequently, based on the complementary mapping relationship, the historical data of resources (such as wind energy or light energy) in a regional set is determined as the first training data set. The characteristic of this data set is that it corresponds to the power data of regional resources in a certain period of time, reflecting the resource fluctuation in the area. Then, the size of the power output is annotated for each sample in the first training data set. These identified samples will be used for supervised learning to help the model learn the power change law between regions. Corresponding to the first training data set, another set of resource data in the complementary area is selected as the second training data set. This data set reflects the resource fluctuation of another area in the same time period and is complementary to the first data set. Similarly, the power size of each sample in the second training data set is annotated to ensure that the model can learn the complementary power relationship between regions. After that, a long short-term memory network (LSTM) is used as the basis for initializing the power prediction model. LSTM is suitable for processing time series data and can capture resource changes and time series characteristics between complementary regions. The input layer of the model receives the resource data in the first and second training data sets, which are processed by the LSTM layer, and the output layer is used to predict the power output of each complementary region in the future time period. LSTM can use memory cells to capture the trend of historical resource fluctuations and thus predict future power changes. Then, the first training data set is input into the initialized power prediction model, which contains the historical data of region i and the corresponding identification power samples. The initialized power prediction model uses its memory cells to process the resource change characteristics of region i through the input first training data set. After each time step of data passes through the LSTM layer, a hidden state is generated to capture the resource change pattern in the time series. The output of the model is then compared with the actual identification power sample, and the error between the predicted value and the true value is calculated by the mean square error MSE. Then, an error back propagation algorithm is used, such as the back propagation algorithm BPTT, to calculate the gradient layer by layer, and adjust the parameters of the model, such as weights and biases, according to these gradients to reduce the error. Repeat the process of forward propagation, error calculation, and back propagation until the loss function of the model on the first training data set converges and the error is minimized. After the first stage of training is completed, the system terminal begins to use the second training data set for further model training. This data set contains the historical resource data of region j and its power identification samples. Similar to the first training data set, the initialized power prediction model captures the resource change pattern of region j by processing the time series data of the second training data set.Since the first and second training data sets are complementary, the initialization power prediction model learns how to combine the resource complementarity of the two regions through this process to smoothly predict the power output. The error of the model on the second training data set is then calculated, and the parameters of the model are further adjusted through the back propagation algorithm. At this stage, the model gradually learns to handle the resource characteristics between the complementary regions and enhances the sensitivity to power fluctuations during prediction. By processing the data of two different complementary regions, the system terminal enables the initialization power prediction model to integrate the resource information of the two regions and adjust the parameters so that the model can not only predict the power output of a single region, but also capture the power balance brought by complementarity. The training process usually alternates for multiple cycles, that is, repeatedly training between the first and second training data sets until the model error on both data sets reaches a stable state. Through alternating training, the model can not only learn the independent resource characteristics of each region, but also capture the complementarity between the two regions, thereby more accurately predicting future power output. During the training process, the performance of the model is regularly evaluated through the validation set to monitor whether the model is overfitting or underfitting. If the error of the validation set is high, adjust the model's hyperparameters, such as the number of hidden layer units and the learning rate, to improve the generalization ability of the model. Through the above training process, the initialized power prediction model can efficiently learn the resource complementarity and power output variation rules between complementary areas, and finally output a stable and accurate complementary power prediction model.

[0037] Further, the present application provides a method for establishing the non-complementary power prediction model, including:

[0038] Based on the non-complementary region set, a third training data set and an identification sample identifying the power size are determined; the initialized power prediction model is trained and fitted according to the third training data set and the identification sample identifying the power size, and a converged non-complementary power prediction model is output.

[0039] Optionally, the system terminal uses the same method as described above to obtain the first training data set, divides the third training data set according to the historical multi-dimensional information perception data of the non-complementary area, and annotates the corresponding power output for each time step data sample in the third training data set as a sample to identify the power size. Subsequently, a training method similar to the above is used to train and fit the initialized power prediction model. Through multiple forward propagations and back propagations, the initialized power prediction model continuously adjusts parameters until the loss function gradually converges and the prediction error of the model on the third training data set is minimized, and the converged initialized power prediction model is output as a non-complementary power prediction model. The model can independently predict the power output of each non-complementary area, identify resource fluctuations in these areas, and predict potential power changes, thereby providing reliable data support for power prediction.

[0040] The complementary power prediction model and the non-complementary power prediction model are loaded into the cloud platform to perform power prediction on the wind and solar farm group.

[0041] In one embodiment, the system terminal deploys the trained complementary power prediction model and non-complementary power prediction model to the cloud platform respectively. The cloud platform is responsible for managing the computing resources and operating environment of these models to ensure that the models can accept new input data and make predictions at any time. For areas belonging to the complementary area set, the cloud platform inputs the real-time data of these areas into the complementary power prediction model. The model uses the resource complementarity between regions to predict the overall power output of each area in the future time period and smooth out power fluctuations. For non-complementary areas, the cloud platform inputs the real-time data of each non-complementary area into the non-complementary power prediction model. The model will independently predict the power output of each area, focusing on the power changes caused by resource fluctuations. After the complementary power prediction model and the non-complementary power prediction model complete the prediction respectively, the system terminal integrates the predicted power output of each area in the cloud platform to generate the power prediction results of the entire wind and solar farm group. These prediction results can be used for tasks such as power scheduling and load balancing of wind and solar farm groups to help improve the operational stability of the power grid.

[0042] Further, if Figure 2 As shown, the present application provides a method of loading the complementary power prediction model and the non-complementary power prediction model into the cloud platform to perform power prediction on the wind and solar farm group, including:

[0043] Obtain the area to be predicted; identify the category to which the area to be predicted belongs; if the area to be predicted is an area of ​​the complementary area set, activate the complementary power prediction model to perform power prediction on the area to be predicted, and obtain a power prediction result; if the area to be predicted is an area of ​​the non-complementary area set, activate the non-complementary power prediction model to perform power prediction on the area to be predicted, and obtain a power prediction result.

[0044] Preferably, the system terminal first obtains the target area for power prediction from the wind and solar farm group, that is, the area to be predicted. These areas to be predicted can be a certain area in the entire wind and solar farm group, or multiple areas that need to be predicted at the same time, and then the real-time multi-dimensional information perception data of these areas to be predicted is transmitted to the cloud platform for processing. After receiving the data of the area to be predicted, the cloud platform checks which category the area belongs to. If the area to be predicted is in the previously divided complementary area set, it means that the area has strong resource complementarity with other areas. If the area to be predicted belongs to the non-complementary area set, it means that the resource changes of the area and other areas are independent, and the complementarity is weak or non-existent. If it is identified that the area to be predicted belongs to the complementary area set, the cloud platform will activate the complementary power prediction model, and perform multiple predictions based on the complementary power prediction model, and output the power prediction results of the area in the future time period. If it is identified that the area to be predicted belongs to the non-complementary area set, the cloud platform will activate the non-complementary power prediction model, and the non-complementary power prediction model will perform power prediction based on the real-time multi-dimensional information perception data of the area, and output the power prediction results of the area in the future time period.

[0045] Furthermore, the present application provides activating the complementary power prediction model to perform power prediction on the area to be predicted, including:

[0046] The complementary power prediction model performs direct power prediction to obtain a power prediction result of the area to be predicted; obtains a complementary area of ​​the area to be predicted; the complementary power prediction model performs complementary power prediction based on the complementary area to obtain a second power prediction result of the area to be predicted; verifies and updates the power prediction result based on the second power prediction result, and outputs the power prediction result of the area to be predicted.

[0047] Optionally, the complementary power prediction model first performs direct power prediction on the real-time multi-dimensional information perception data of the area to be predicted, and generates a preliminary power prediction result for the area to be predicted. After the direct prediction result is generated, the system terminal obtains the complementary area corresponding to the area to be predicted according to the set complementary area set. These complementary areas are areas with strong resource complementarity with the area to be predicted, and can produce a balancing and regulating effect on the power output of the area to be predicted through their power changes. Then, the real-time multi-dimensional information perception data of the prediction area and its complementary area are simultaneously input into the complementary power prediction model, and complementary power prediction is performed to calculate the second power prediction result of the area to be predicted. This result reflects the possible adjustment of the power output of the area to be predicted under the action of the complementary area resources. Subsequently, the preliminary power prediction result is compared with the second power prediction result. If the difference between the two is greater than or equal to the verification update threshold, it means that the resources of the complementary area have a strong regulating effect on the area to be predicted, and the system terminal outputs the second power prediction result as the final power prediction result of the area to be predicted. On the contrary, it means that the preliminary power prediction result is more reliable and the impact of the complementary area is small. At this time, the system terminal will calculate the mean of the preliminary power prediction result and the second power prediction result, and output the calculation result as the final power prediction result of the area to be predicted, ensuring the accuracy and stability of the power prediction.

[0048] In summary, the embodiments of the present application have at least the following technical effects:

[0049] The embodiment of the present application collects multi-dimensional information perception data sets of each area of ​​the wind and solar farm group through a cloud platform, and constructs a regional joint probability distribution model to analyze the joint distribution of wind energy and light energy in multiple areas in the same time period. Subsequently, the multi-dimensional data sets of each area are input into the model, the joint probability distribution of each pair of areas is calculated, a joint probability distribution set is generated, and the areas are divided according to the set, and a complementary area set and a non-complementary area set are output. Based on the complementary area set and the non-complementary area set, a complementary power prediction model and a non-complementary power prediction model are established respectively, and these models are loaded into the cloud platform to perform power prediction of the wind and solar farm group. For the area to be predicted, it is identified whether it belongs to the complementary area set or the non-complementary area set, and the corresponding power prediction model is activated respectively to obtain the power prediction result. The core calculation step of the model also includes obtaining multiple complementary indexes according to the characteristic time periods of the complementary areas and the non-complementary areas, constructing a complementary matrix based on these indexes, and outputting the complementary area and the non-complementary area set through binary clustering. These technical effects jointly solve the technical problem of large power prediction errors caused by uneven distribution of wind and light energy among regions in wind and solar farms, and achieve the effect of accurately dividing complementary and non-complementary areas through the aggregation of multi-dimensional information perception data and the construction of a joint probability distribution model, thereby improving the accuracy and stability of power prediction for wind and solar farms.

[0050] 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. The processes depicted in the accompanying 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.

[0051] 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.

[0052] 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 wind and solar farm power prediction method based on multi-dimensional information perception and convergence, characterized in that: The method comprises: Collect multi-dimensional information perception data sets of each area in the wind and solar farm group through the cloud platform; Constructing a regional joint probability distribution model, wherein the regional joint probability distribution model is used to analyze the joint distribution of wind energy and solar energy in multiple regions within the same time period; Inputting the multidimensional information perception data sets of each region into the regional joint probability distribution model, calculating the joint probability distribution of each pair of regions, and outputting a joint probability distribution set; Performing region division according to the joint probability distribution set, and outputting a complementary region set and a non-complementary region set; Based on the complementary region set and the non-complementary region set, respectively establish a complementary power prediction model and a non-complementary power prediction model; Loading the complementary power prediction model and the non-complementary power prediction model into the cloud platform to perform power prediction on the wind and solar farm group; Among them, the joint probability distribution of each pair of regions is calculated, and the expression is: ； in, and represents the wind energy resources in regions i and j, and represents the light energy resources of region i and region j, It represents the conditional probability distribution of wind and solar resources in region j under the condition that the wind and solar resources in region i are known; The method further comprises: performing region division according to the joint probability distribution set, outputting a complementary region set and a non-complementary region set. According to the joint probability distribution set, the complementarity index of each pair of regions in the same time period is calculated, and a complementarity index set is output, wherein the expression for calculating each complementarity index is: ； in, is the wind energy value of region i at time t, is the light energy value of region j at time t, T is the length of the time period, is the complementarity index between region i and region j, The size of is proportional to the complementarity; Constructing a complementarity matrix according to the complementarity index set; The complementarity matrix is ​​subjected to binary clustering to output a complementary region set and a non-complementary region set, wherein the complementary region set is a region having a complementarity index greater than or equal to a preset complementarity index, and the non-complementary region set is a region having a complementarity index less than the preset complementarity index.

2. The method according to claim 1, characterized in that The complementary power prediction model and the non-complementary power prediction model are loaded into the cloud platform to perform power prediction on the wind and solar farm group, the method comprising: Get the area to be predicted; Identify the category to which the area to be predicted belongs, and if the area to be predicted is an area of ​​the complementary area set, activate the complementary power prediction model to perform power prediction on the area to be predicted, and obtain a power prediction result; If the area to be predicted is an area of ​​the non-complementary area set, the non-complementary power prediction model is activated to perform power prediction on the area to be predicted to obtain a power prediction result.

3. The method according to claim 1, characterized in that The method further comprises: Dividing the wind and solar farm group into time periods to obtain multiple characteristic time periods, wherein the multiple characteristic time periods include daytime, nighttime, seasonal time periods, and extreme weather time periods; Obtaining a plurality of complementarity indexes corresponding one-to-one to the plurality of characteristic time periods for each pair of regions; The plurality of complementarity indices are fused to determine an updated complementarity index for each pair of regions, so as to update the complementarity index set.

4. The method according to claim 2, characterized in that The method of establishing a complementary power prediction model includes: Based on the complementary mapping relationship of the complementary region set, determining a first training data set and an identification sample of an identification power size, and a second training data set and an identification sample of an identification power size, wherein the first training data set and the second training data set are in a complementary mapping relationship; Using long short-term memory network, an initialization power prediction model is constructed; The initialized power prediction model is trained according to the first training data set and the identification samples identifying the power size, and the second training data set and the identification samples identifying the power size, and a converged complementary power prediction model is output.

5. The method according to claim 4, characterized in that Activating the complementary power prediction model to perform power prediction on the to-be-predicted area, the method further comprising: The complementary power prediction model performs direct power prediction to obtain a power prediction result of the area to be predicted; Acquire a complementary region to the region to be predicted; The complementary power prediction model performs complementary power prediction according to the complementary area to obtain a second power prediction result of the area to be predicted; The power prediction result is verified and updated according to the second power prediction result, and the power prediction result of the area to be predicted is output.

6. The method according to claim 4, characterized in that The non-complementary power prediction model is established, and the method includes: Based on the non-complementary region set, determining a third training data set and an identification sample of an identification power size; The initialized power prediction model is trained and fitted according to the third training data set and the identification samples identifying the power magnitude, and a converged non-complementary power prediction model is output.

Images