Power equipment power prediction method and system based on machine learning

Abstract

The invention discloses an electrical equipment power prediction method and system based on machine learning, relates to the technical field of electrical energy, and solves the technical problem that the generation power prediction precision is seriously influenced because the mutual influence between meteorological data is not considered and the accuracy of meteorological prediction data is difficult to guarantee in the prior art. Historical record data of distributed photovoltaic equipment is acquired, environment record data is divided into a plurality of data sets according to a solar activity track, and a mapping relation between platform prediction data and actual detection data in the data sets is constructed through a machine learning algorithm; according to the method, the historical record data is divided through the sun activity trajectory to mine the mapping relationship in different time periods, so that the data set quality can be improved; the mapping relation is constructed through the machine learning algorithm, the mutual influence between the environmental factors can be accurately mined, the construction efficiency and precision of the mapping relation are improved, and meanwhile, the construction cost can be reduced because the mapping relation does not need to be constructed frequently.

Description

Technical Field

[0001] This application belongs to the field of power energy technology, specifically a power prediction method and system for power equipment based on machine learning. Background Technology

[0002] Power prediction for distributed photovoltaic (PV) systems refers to the process of forecasting the power generation capacity of these systems over a future period using specific methods. This forecast is of great significance for grid dispatching, power system management, and the optimized operation of PV power generation equipment.

[0003] The power generation of distributed photovoltaic (PV) systems is primarily influenced by meteorological factors, such as solar irradiance, temperature, and humidity. Current machine learning-based distributed PV power prediction systems mainly rely on historical data to uncover the mapping relationship between meteorological factors and their corresponding power generation at specific times. This mapping relationship is then combined with meteorological forecast data to obtain the desired predicted power generation. However, meteorological data primarily comes from meteorological forecasting platforms. These platforms, when forecasting large-scale meteorological data, cannot accurately account for the interactions between different meteorological data points, making it difficult to guarantee the accuracy of meteorological forecasts. This significantly impacts the accuracy of power generation predictions.

[0004] This application provides a machine learning-based method and system for predicting the power of electrical equipment to solve the aforementioned technical problems. Summary of the Invention

[0005] This application aims to at least solve one of the technical problems existing in the prior art; to this end, this application proposes a power equipment power prediction method and system based on machine learning to solve the technical problem that the prior art does not consider the mutual influence between meteorological data, making it difficult to guarantee the accuracy of meteorological prediction data, which will seriously affect the accuracy of power generation prediction.

[0006] To achieve the above objectives, a first aspect of this application provides a machine learning-based method for predicting the power output of electrical equipment, comprising: Acquire historical data of distributed photovoltaic equipment; the historical data includes equipment status data, environmental record data and corresponding power generation, and the environmental record data includes platform prediction data and actual detection data; The environmental record data is divided into several datasets according to the solar activity trajectory. The mapping relationship between the platform's predicted data and the actual detection data in these datasets is constructed using machine learning algorithms. Based on environmental prediction data and mapping relationships, actual environmental data is determined, and the power generation of distributed photovoltaic equipment is predicted based on equipment status data and actual environmental data.

[0007] Preferably, the environmental record data is divided into several datasets according to the solar activity trajectory, including: The effective time of several environmental records in the historical data is obtained; among them, the platform prediction data and the actual detection data are time-registered and have the same effective time. The day is divided into several time periods based on the solar activity trajectory; environmental records are matched for each time period based on the time of solar activity to obtain datasets for several time periods.

[0008] Preferably, the day is divided into several time periods based on the solar activity trajectory, including: The solar activity trajectory is marked based on the solar terms, resulting in several feature markers; these feature markers are used to assist in matching environmental record data. The target point is set when the sun is at noon. Based on the target point, the time period is divided towards the sunrise or sunset side according to a set interval. The set interval may be one hour or fifteen minutes. Associate feature tags with several time periods.

[0009] Preferably, a mapping relationship between platform prediction data and actual detection data is constructed using machine learning algorithms in several datasets, including: A mapping model is constructed based on a machine learning algorithm; wherein the machine learning algorithm includes at least one hidden layer. The platform prediction data in the dataset is used as the model input data, and the difference between the actual detection data and the platform prediction data is used as the model output data; a mapping model is trained based on the model input data and the model output data. The influence weights of each environmental element on other environmental elements are extracted from the trained mapping model, and the mapping relationship between environmental elements and other environmental elements is constructed based on the influence weights; among them, environmental elements include solar irradiance, temperature, wind speed and humidity.

[0010] Preferably, the mapping relationship between environmental elements and other environmental elements is constructed based on influence weights, including: The environmental elements in the actual detection data are used as target elements in sequence; The weight matrix corresponding to the target element is extracted from the trained mapping model, and the mapping relationship between the weight matrix and the platform prediction data is established. The mapping relationship is constructed based on a function model or an artificial intelligence model.

[0011] Preferably, the actual environmental data is determined based on environmental prediction data and mapping relationships, including: Environmental forecast data and the timing of their effects are obtained through a meteorological platform; Determine the feature markers and time periods corresponding to the time of action, and extract the mapping relationship based on the feature markers and time periods; calculate the actual environmental data based on the environmental prediction data and the mapping relationship.

[0012] Preferably, predicting the power generation of distributed photovoltaic equipment based on equipment status data and actual environmental data includes: Obtain a power prediction model; wherein, the power prediction model is constructed based on an artificial intelligence model; After preprocessing, the equipment status data and actual environmental data are input into the power prediction model to obtain the predicted power generation of the distributed photovoltaic equipment.

[0013] Preferably, after predicting the power generation of the photovoltaic power generation equipment, the predicted power is fitted with the power generation during the period to be predicted to obtain the predicted power curve.

[0014] Preferably, a power prediction model is constructed, including: Obtain the training dataset; the training dataset includes equipment status data, actual detection data and corresponding power generation, and the equipment status data includes equipment age, photoelectric conversion efficiency and power factor; The constructed artificial intelligence model is trained based on the training dataset to obtain the power prediction model; the artificial intelligence model includes a BP neural network model or an RBF neural network model.

[0015] The second aspect of this application provides a power prediction system for power equipment based on machine learning, including a data processing layer and a data acquisition layer that interacts with the data thereon. Data acquisition layer: used to acquire historical data of distributed photovoltaic equipment; the historical data includes equipment status data, environmental record data and corresponding power generation, and the environmental record data includes platform prediction data and actual detection data; Data processing layer: used to divide environmental record data into several datasets according to the solar activity trajectory, and to construct the mapping relationship between the platform's predicted data and the actual detection data in several datasets through machine learning algorithms; In addition, it is used to determine actual environmental data based on environmental prediction data and mapping relationships, and to predict the power generation of distributed photovoltaic equipment based on equipment status data and actual environmental data.

[0016] Compared with the prior art, the beneficial effects of this application are: 1. This application obtains historical data of distributed photovoltaic equipment, divides the environmental record data into several datasets according to the solar activity trajectory, and constructs a mapping relationship between the platform prediction data and the actual detection data in several datasets through machine learning algorithms. This application divides the historical data by solar activity trajectory to mine the mapping relationship in different time periods, which can improve the quality of the dataset. Constructing the mapping relationship through machine learning algorithms can accurately mine the mutual influence between environmental factors, improve the effectiveness and accuracy of the mapping relationship construction, and reduce the construction cost since the mapping relationship does not need to be constructed frequently.

[0017] 2. This application determines the actual environmental data based on environmental prediction data and mapping relationships, and predicts the power generation of distributed photovoltaic equipment based on equipment status data and actual environmental data. After determining the actual environmental data corresponding to the prediction time, this application combines the equipment status data at the prediction time to predict the power generation of distributed photovoltaic equipment. It predicts the power generation from both the equipment itself and the environmental impact, which can improve the prediction accuracy. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the method flow for the distributed photovoltaic power prediction method in Embodiment 1 of this application; Figure 2 This is a schematic diagram illustrating the process of constructing the mapping relationship in Embodiment 1 of this application; Figure 3 This is a schematic diagram illustrating the system principle of the distributed photovoltaic prediction system of this application. Detailed Implementation

[0020] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] Distributed photovoltaic (PV) systems are systems that convert solar energy into electrical energy, characterized by distributed generation and high flexibility. The power generation capacity of distributed PV systems is of great significance for grid dispatch and power system management; therefore, accurate prediction of their power generation capacity is essential.

[0022] Since the power generation capacity of distributed generation equipment is affected by various factors, such as solar irradiance, temperature, wind speed, and humidity, prediction can generally be achieved by mining the relationship between these data and power generation capacity. Current prediction methods mainly rely on machine learning algorithms. However, due to the wide distribution of distributed generation equipment, models cannot be standardized, and data is insufficient in some areas, affecting the prediction results.

[0023] Most importantly, environmental forecast data is acquired through a meteorological platform during the forecasting process, and power generation is predicted based on this data. However, the meteorological platform relies on a wide range of sensors to acquire environmental forecast data, resulting in insufficient spatiotemporal resolution. Therefore, the environmental forecast data is not accurate for the location of distributed photovoltaic (PV) equipment. Furthermore, due to the interaction of environmental factors, there can be a significant discrepancy between the actual environmental data at the location of the distributed PV equipment and the obtained environmental forecast data.

[0024] The distributed photovoltaic power prediction method and system provided in this example are used to solve the above problems.

[0025] Please see Figures 1-2 The first aspect of this application provides a machine learning-based method for predicting the power output of electrical equipment, comprising: Acquire historical data of distributed photovoltaic (PV) devices; divide environmental records into several datasets according to solar activity trajectories; construct mapping relationships between platform-predicted data and actual detection data in these datasets using machine learning algorithms; determine actual environmental data based on environmental prediction data and mapping relationships; and predict the power generation of distributed PV devices based on device status data and actual environmental data.

[0026] Historical data includes equipment status data, environmental record data, and corresponding power generation. Environmental record data includes platform prediction data and actual monitoring data. Equipment status data is mainly used to assess the overall equipment status of distributed photovoltaic (PV) equipment, thereby determining its impact on power generation. Platform prediction data and actual monitoring data, aside from differing acquisition channels, contain the same types of content, both including various environmental factors affecting power generation. Platform prediction data is obtained from a meteorological platform, while actual monitoring data is obtained from sensors placed around the distributed PV equipment. It should be noted that the sensors acquiring the actual monitoring data can be multiple types of sensors or integrated sensors capable of acquiring multiple data types.

[0027] This embodiment does not directly use platform prediction data and power generation from historical data to build a prediction model as in traditional methods. This is because the prediction model built based on these two types of data does not consider the impact of the aging of the distributed photovoltaic equipment itself, and the platform prediction data is not consistent with the environmental data of the distributed photovoltaic equipment, and does not take into account the mutual influence between environmental data, which will lead to inaccurate models.

[0028] Similarly, there is no prediction model built based on the actual measured detection data and power generation. Although the actual detection data is the environmental data when the distributed photovoltaic equipment is working, if the prediction model is built based on the actual detection data, then the environmental data at future moments needs to be predicted. If the future environmental data is predicted based on the actual detection data, the impact of environmental changes cannot be predicted, which will also lead to inaccurate data and thus affect the prediction accuracy of power generation.

[0029] This embodiment mines the relationship between platform-predicted data and actual detection data from environmental records based on solar activity trajectories, establishing a mapping relationship between the two. When predicting future power generation, platform-predicted data for the corresponding time can be obtained from the meteorological platform, and the corresponding actual detection data can be determined by combining the mapping relationship. The subsequent prediction process is then completed based on this actual detection data. Since the actual detection data for this prediction is based on platform-predicted data, the impact of sudden environmental changes is considered. Furthermore, by using the mapping relationship as a foundation, the interaction between data from different seasons and times is also taken into account, thus improving prediction accuracy.

[0030] In this embodiment, the environmental record data is divided according to the solar activity trajectory, which is essentially based on the sun's position in a year and its position in a day.

[0031] First, you need to divide the day into several time periods based on the sun's activity trajectory. You can refer to the following steps: The solar activity trajectory is marked based on the solar terms to obtain several feature markers; the sun's noon position is taken as the target point; based on the target point, the sun is divided into several time periods according to the set intervals towards the sunrise or sunset side; and the feature markers are associated with several time periods.

[0032] The aforementioned feature markers are used to assist in matching environmental recording data; the set intervals include one hour or fifteen minutes. The sun's daily trajectory can include sunrise, noon, and sunset, but its data for each day of the year is not entirely the same. For example, sunrise and sunset in summer and winter differ significantly. Therefore, this embodiment introduces solar terms to set feature markers for each day of the year. Based on the solar terms, the year is divided into 24 time periods, each corresponding to a feature marker (the feature markers for each day within each time period are the same). In some other preferred embodiments, time periods can also be divided using methods such as seasons to set feature markers.

[0033] Furthermore, even within the same time period, sunrise and sunset times are not exactly the same on different days. In this embodiment, noon is used as the reference point, and the time is divided to both sides at set intervals, i.e., towards the sunrise or sunset side, to obtain several time periods. All the time periods of each day are spliced ​​together to form the time range corresponding to the sun from sunrise to sunset. Since there is no sun at night and the distributed photovoltaic equipment does not work, the nighttime period is not analyzed.

[0034] For example, the next solar term after the winter solstice is the minor cold. A feature marker is set for the days between the winter solstice and the minor cold (including the day of the winter solstice, but excluding the day of the minor cold). The earliest sunrise point of the day corresponding to the feature marker is marked as the left base point, and the latest sunset point is marked as the right base point. Assuming the base point on the left is 6:00 AM (relative to the location of the distributed photovoltaic equipment), the points on the right are 6:00 PM, noon corresponds to 12:00 PM, and the time interval is one hour; then we can obtain the time intervals [6, 7], (7, 8], (8, 9], (9, 10], (10, 11], (11, 12], (12, 13], (13, 14], (14, 15], (15, 16], (16, 17], (17, 18).

[0035] After determining the feature markers and time periods, the effective time of several environmental record data in the historical data is obtained. This effective time is both the effective time of the platform prediction data and the effective time of the actual detection data. It can be understood that the platform prediction data and the actual detection data in each environmental record data are registered.

[0036] It should be noted that the "action time" here refers to the actual time when the data takes effect. For example, if the platform predicts that the data will be applied to the distributed photovoltaic equipment at 8:00 AM this morning, then 8:00 AM is the action time of the platform's predicted data. The actual time of data collection can be used as the action time.

[0037] Historical data is divided into datasets corresponding to different time periods based on the time of application. Since the sunrise and sunset times corresponding to the same feature are not consistent each day, the amount of data in each dataset is also different. However, as long as the historical data is sufficient, it does not affect the accuracy of the mapping relationship obtained based on the dataset.

[0038] Next, we will use machine learning algorithms to construct a mapping relationship between the platform's predicted data and the actual detection data in several datasets. Please refer to the following steps for details: A mapping model is constructed based on a machine learning algorithm; wherein the machine learning algorithm includes at least one hidden layer. The platform prediction data in the dataset is used as the model input data, and the difference between the actual detection data and the platform prediction data is used as the model output data; a mapping model is trained based on the model input data and the model output data. The influence weights of each environmental element on other environmental elements are extracted from the trained mapping model, and the mapping relationship between environmental elements and other environmental elements is constructed based on the influence weights; among them, environmental elements include solar irradiance, temperature, wind speed and humidity.

[0039] Machine learning algorithms such as decision tree models, support vector machine models, and neural network models can all extract weight matrices. Taking neural network models as an example, a mapping model is constructed based on the neural network model. This mapping model includes an input layer, an output layer, and at least one hidden layer.

[0040] The dataset corresponding to the time period is extracted. The platform's predicted data and the actual detection data in the dataset are preprocessed to generate model input data and model output data, respectively. It should be noted that the model output data is the difference between the actual detection data and the platform's predicted data for each environmental factor.

[0041] The mapping model is trained using model input data and model output data. The weight relationship between the changes of each environmental factor in the model output data and the weights of each environmental factor in the platform prediction data can be extracted from the trained mapping model, and the weight matrix corresponding to the changes of each environmental factor can be obtained.

[0042] For example, suppose there are environmental factors A1, A2, and A3 in the platform's predicted data, and environmental factor changes B1, B2, and B3 in the actual detected data. In the trained mapping model, B1 corresponds to a set of weights with A1, A2, and A3. The weight corresponding to A1 in this set of weights is the sum of the weights of each neuron of environmental factor A1 in the hidden layer. The other weights are calculated in the same way.

[0043] The change in environmental factors is equivalent to the dependent variable, and each environmental factor in the model input data is equivalent to an independent variable. Each independent variable corresponds to an influence weight, and all influence weights form the weight matrix of the dependent variable. The mapping relationship between environmental elements based on influence weights can be established by referring to the following steps: The environmental elements in the actual detection data are used as target elements in sequence; The weight matrix corresponding to the target element is extracted from the trained mapping model, and the mapping relationship between the weight matrix and the platform prediction data is established. The mapping relationship is constructed based on a function model or an artificial intelligence model.

[0044] For example, taking a functional model, suppose the environmental factors in the environmental prediction data are A1, A2, and A3, and the corresponding weights of A3 with A1 and A2 are α1 and α2, respectively. Considering the mutual influence between environmental factors, A3' = A3 + α1 × A1 + α2 × A2. This functional expression can be used as a mapping relationship.

[0045] After determining the mapping relationship, environmental prediction data and the corresponding action time are obtained from the platform. Based on the feature tags and time periods corresponding to the action time, the mapping relationship is matched. Through this mapping relationship and the environmental prediction data, the actual environmental data of the environment where the distributed photovoltaic equipment is located at the action time can be calculated. Combining the equipment status data of the distributed photovoltaic equipment and the calculated actual environmental data, the power generation can be predicted.

[0046] It is worth noting that the above mapping relationship generally remains unchanged with sufficient historical data. However, the use of distributed photovoltaic equipment and changes in climate conditions may alter the mapping relationship. Therefore, the mapping relationship should be updated using newly collected historical data. Since the update is periodic and does not require real-time processing, it can be performed at night to improve resource utilization.

[0047] Example 2: Based on Example 1, this example provides a method for predicting the power generation of distributed photovoltaic equipment based on equipment status data and actual environmental data. The following steps can be referred to: Obtain the power prediction model; after preprocessing the equipment status data and actual environmental data, input them into the power prediction model to obtain the predicted power generation of the distributed photovoltaic equipment.

[0048] Constructing a power prediction model includes: Obtain the training dataset; the training dataset includes equipment status data, actual detection data and corresponding power generation, and the equipment status data includes equipment age, photoelectric conversion efficiency and power factor; The constructed artificial intelligence model is trained based on the training dataset to obtain the power prediction model; the artificial intelligence model includes a BP neural network model or an RBF neural network model.

[0049] It should be noted that the preprocessing here includes not only data processing suitable for model training, but also determining the device state data corresponding to the actual environmental data at the time of impact. After determining the time of impact corresponding to the actual environmental data, time-series modeling can be used to perform time-series analysis on the device state data to predict the device state data corresponding to the time of impact.

[0050] A power prediction model can determine the power generation at various future times. Fitting the power generation over a future time period yields a predicted power curve. Integrating this predicted power curve provides the power generation for the corresponding integration time period.

[0051] Please see Figure 3 The second aspect of this application provides a power prediction system for power equipment based on machine learning, including a data processing layer and a data acquisition layer that interacts with the data thereon. Data acquisition layer: used to acquire historical data of distributed photovoltaic equipment; the historical data includes equipment status data, environmental record data and corresponding photovoltaic power generation, and the environmental record data includes platform prediction data and actual detection data; Data processing layer: used to divide environmental record data into several datasets according to the solar activity trajectory, and to construct the mapping relationship between the platform's predicted data and the actual detection data in several datasets through machine learning algorithms; In addition, it is used to determine actual environmental data based on environmental prediction data and mapping relationships, and to predict the power generation of distributed photovoltaic equipment based on equipment status data and actual environmental data.

[0052] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A power prediction method for electrical equipment based on machine learning, characterized in that, include: Acquire historical data of distributed photovoltaic equipment; the historical data includes equipment status data, environmental record data and corresponding power generation, and the environmental record data includes platform prediction data and actual detection data; The environmental record data is divided into several datasets according to the solar activity trajectory, and the mapping relationship between the platform prediction data and the actual detection data in the several datasets is constructed by machine learning algorithm; Based on the environmental prediction data and the mapping relationship, the actual environmental data is determined, and the power generation of the distributed photovoltaic equipment is predicted based on the equipment status data and the actual environmental data.

2. The power prediction method for power equipment based on machine learning according to claim 1, characterized in that, The environmental record data is divided into several datasets according to the solar activity trajectory, including: The effective time of several environmental record data in the historical data is obtained; wherein, the platform prediction data and the actual detection data are time-registered and their effective times are the same. The day is divided into several time periods based on the solar activity trajectory; environmental record data is matched for each time period based on the time of influence to obtain a dataset for each time period.

3. The power prediction method for power equipment based on machine learning according to claim 2, characterized in that, The day is divided into several time periods based on the solar activity trajectory, including: The solar activity trajectory is marked based on the solar terms to obtain several feature markers; among them, the feature markers are used to assist in matching environmental record data; The target point is set when the sun is at noon. Based on the target point, the time period is divided towards the sunrise or sunset side according to a set interval. The set interval includes one hour or fifteen minutes. The feature markers are associated with several time periods.

4. The power prediction method for power equipment based on machine learning according to claim 1, characterized in that, The mapping relationship between the platform's predicted data and the actual detection data in several datasets is constructed using machine learning algorithms, including: A mapping model is constructed based on a machine learning algorithm; wherein the machine learning algorithm includes at least one hidden layer. The platform prediction data in the dataset is used as the model input data, and the difference between the actual detection data and the platform prediction data is used as the model output data; the mapping model is trained based on the model input data and the model output data. The influence weights of each environmental element on other environmental elements are extracted from the trained mapping model, and the mapping relationship between the environmental element and other environmental elements is constructed based on the influence weights; wherein, the environmental elements include solar irradiance, temperature, wind speed and humidity.

5. The power prediction method for power equipment based on machine learning according to claim 4, characterized in that, Based on the influence weights, a mapping relationship is constructed between the environmental elements and other environmental elements, including: The environmental elements in the actual detection data are sequentially used as target elements; The weight matrix corresponding to the target element is extracted from the trained mapping model, and the mapping relationship between the weight matrix and the platform prediction data is established for the target element; wherein the mapping relationship is constructed based on a function model or an artificial intelligence model.

6. The power prediction method for power equipment based on machine learning according to claim 3, characterized in that, Based on the environmental prediction data and the mapping relationship, the actual environmental data is determined, including: Environmental forecast data and the timing of their effects are obtained through a meteorological platform; The feature marker and time period corresponding to the time of action are determined, and the mapping relationship is extracted based on the feature marker and the time period; the actual environmental data is calculated based on the environmental prediction data and the mapping relationship.

7. The power prediction method for power equipment based on machine learning according to claim 1, characterized in that, Predicting the power generation of the distributed photovoltaic equipment based on equipment status data and actual environmental data includes: Obtain a power prediction model; wherein, the power prediction model is constructed based on an artificial intelligence model; After preprocessing, the equipment status data and actual environmental data are input into the power prediction model to obtain the predicted power generation of the distributed photovoltaic equipment.

8. The power prediction method for power equipment based on machine learning according to claim 7, characterized in that, After predicting the power generation of the photovoltaic power generation equipment, the predicted power curve is obtained by fitting the predicted power generation during the period to be predicted.

9. The power prediction method for power equipment based on machine learning according to claim 7, characterized in that, Constructing the power prediction model includes: Obtain the training dataset; the training dataset includes equipment status data, actual detection data and corresponding power generation, and the equipment status data includes equipment age, photoelectric conversion efficiency and power factor; The constructed artificial intelligence model is trained based on the training dataset to obtain a power prediction model; wherein the artificial intelligence model includes a BP neural network model or an RBF neural network model.

10. A machine learning-based power equipment power prediction system, used to execute the machine learning-based power equipment power prediction method according to any one of claims 1 to 9, characterized in that, It includes a data processing layer and a data acquisition layer that interacts with the data. Data acquisition layer: used to acquire historical data of distributed photovoltaic equipment; the historical data includes equipment status data, environmental record data and corresponding power generation, and the environmental record data includes platform prediction data and actual detection data; Data processing layer: used to divide the environmental record data into several datasets according to the solar activity trajectory, and to construct the mapping relationship between the platform prediction data and the actual detection data in several datasets through machine learning algorithms; In addition, it is used to determine actual environmental data based on environmental prediction data and the mapping relationship, and to predict the power generation of the distributed photovoltaic equipment based on equipment status data and actual environmental data.

Images