Photovoltaic power station cluster generation power forecasting method and system

Abstract

The invention discloses a photovoltaic power station cluster generation power forecasting method and system, and the method comprises the steps: S101, obtaining a first power forecasting result through a first power forecasting model obtained through a first training mode for a first type of photovoltaic power station in a cluster; step S102, for a second type of photovoltaic power station in the cluster, performing fine tuning on the first power forecasting model to obtain a second power forecasting model, and using the second power forecasting model to obtain a second power forecasting result; and S103, summarizing the first power forecasting result and the second power forecasting result in the cluster to obtain a power generation power forecasting result of the photovoltaic power station cluster. According to the method, the problem that a newly added photovoltaic power station cannot be fully trained due to scarcity of training samples is solved. Moreover, the forecasting model of the photovoltaic power station in the photovoltaic power station cluster is updated progressively by using the newly accumulated data, so that the generalization ability and forecasting accuracy of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power forecasting, and for example, to a method and system for forecasting power generation of a photovoltaic power station cluster. Background Art

[0002] With the rapid development of the global economy, the demand for energy is also increasing. However, traditional energy sources such as oil and coal are non-renewable resources, and their exploitation and utilization are already facing serious situations. At the same time, the supply of these energy sources is often affected by factors such as natural disasters, and there is a great deal of uncertainty. Photovoltaic energy, as a clean and renewable energy source, comes from sunlight and is an inexhaustible energy source. The development of photovoltaic energy can effectively reduce dependence on traditional energy and ensure national energy security. Benefiting from national policy encouragement and technological development drive, my country's total photovoltaic installed capacity and increment are both among the highest in the world.

[0003] Photovoltaic power generation is affected by factors such as sunshine and climate, which leads to the volatility and periodicity of photovoltaic power generation, which interferes with the normal operation of the power grid. Therefore, with the increase of photovoltaic power generation, the prediction of photovoltaic power generation has become more and more important. With the development of deep learning technology, the use of deep learning models to predict photovoltaic power generation has gradually become the mainstream. For deep learning models, a large amount of historical data is required for model training. However, for newly added photovoltaic power stations, the historical data accumulation is insufficient and cannot support the training of the model. In addition, with the operation of photovoltaic power stations, the photovoltaic power stations themselves are aging and the panels are covered with dust. Therefore, updating the model according to the newly accumulated data of the photovoltaic power station to adapt to various changes is also a problem that needs to be solved. Summary of the invention

[0004] To solve the above problems, the purpose of the present invention is to provide a method and system for predicting the power generation of a photovoltaic power station cluster, so as to achieve accurate prediction of the power generation of various types of power stations in the cluster, especially for newly installed photovoltaic power stations, by making use of the rich historical data of other types of power stations in the cluster to train the prediction model, and by using a small amount of its own data to fine-tune the model to obtain the prediction model, and as the data accumulates, the model is progressively updated, thereby improving the accuracy of the power generation forecast of the photovoltaic power station cluster.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for predicting power generation of a photovoltaic power station cluster, comprising the following steps: Step S101: for a photovoltaic power station of a first type in a cluster, a first power forecasting model obtained by a first training method is used to obtain a first power forecasting result.

[0006] Step S102: for the second type of photovoltaic power station in the cluster, fine-tune the first power forecast model to obtain a second power forecast model, and use the second power forecast model to obtain a second power forecast result.

[0007] Step S103: Summarize the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

[0008] Furthermore, the first type of photovoltaic power station has sufficient first historical data for training the first power forecast model; the amount of second historical data accumulated by the second type of photovoltaic power station cannot support the training of the first power forecast model according to the first training method; the first training method is a supervised training method or a semi-supervised training method, and the first power forecast model includes a feature extraction layer and an output layer.

[0009] Further, the first power forecast model is fine-tuned to obtain the second power forecast model, specifically: Step S1021: divide the second historical data into a second training set and a second validation set; step S1022: freeze the parameters of the feature extraction layer of the first power forecasting model; step S1023: use the second training set to train the first power forecasting model, and adjust the parameters of the output layer of the first power forecasting model; step S1024: evaluate the prediction performance of the fine-tuned first power forecasting model in the second validation set; step S1025: if the prediction error is greater than the set threshold, obtain the second power forecasting model; otherwise, adjust the learning rate and re-execute steps S1021-S1025.

[0010] Furthermore, for the second type of photovoltaic power station in the cluster, after using the second power forecasting model to forecast the power generation for a certain period, third historical data are obtained through data accumulation; the third historical data are divided into a third training set and a third verification set; the third training set is used to perform incremental sample training on the second power forecasting model to obtain a third power forecasting model; if the prediction performance of the third power forecasting model on the third verification set is better than that of the second power forecasting model, the second type of photovoltaic power station uses the third power forecasting model to replace the second power forecasting model for power forecasting.

[0011] Further, the use of the third training set to perform incremental sample training on the second power forecast model is specifically as follows: Freeze the output layer, perform incremental sample training on the second power prediction model using the third training set, and adjust parameters of the feature extraction layer; or, The third training set is divided into two parts; the output layer is frozen, and the second power prediction model is trained with incremental samples using the first part of the third training set, and the parameters of the feature extraction layer are adjusted; the feature extraction layer is frozen, and the second power prediction model is trained with incremental samples using the second part of the third training set, and the parameters of the output layer are adjusted; Alternatively, the feature extraction layer and the output layer are not frozen, and the third training set is used to perform incremental sample training on the second power prediction model, and the parameters of the feature extraction layer and the output layer are adjusted.

[0012] In another aspect, the present invention provides a photovoltaic power station cluster power generation forecasting system, comprising the following functional modules: The first power forecasting module is used to perform power forecasting on a first type of photovoltaic power station in the cluster, and obtain a first power forecasting result using a first power forecasting model obtained by a first training method.

[0013] The second power forecasting module is used to forecast the power of the second type of photovoltaic power station in the cluster, fine-tune the first power forecasting model, obtain the second power forecasting model, and use the second power forecasting model to obtain the second power forecasting result.

[0014] Forecast result summary module: summarizes the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

[0015] Furthermore, the second power forecasting module includes a model fine-tuning unit.

[0016] Model fine-tuning unit: divide the second historical data into a second training set and a second validation set, freeze the parameters of the feature extraction layer of the first power forecasting model, use the second training set to train the first power forecasting model, adjust the parameters of the output layer of the first power forecasting model, evaluate the prediction performance of the fine-tuned first power forecasting model in the second validation set, if the prediction error is greater than the set threshold, obtain the second power forecasting model; otherwise, adjust the learning rate and use the second historical data to re-fine-tune the first power forecasting model.

[0017] Furthermore, the first power forecasting module includes a first optimization unit.

[0018] The first optimization unit is used to optimize the first power forecast model; after using the first power forecast model to forecast the power generation for a certain period of time, fourth historical data is obtained through data accumulation; the fourth historical data is divided into a fourth training set and a fourth validation set; the fourth training set is used to perform incremental sample training on the first power forecast model to obtain a fourth power forecast model; if the prediction performance of the fourth power forecast model on the fourth validation set is better than that of the first power forecast model, the first type of photovoltaic power station uses the fourth power forecast model to replace the first power forecast model for power forecasting.

[0019] Furthermore, the second power forecasting module includes a second optimization unit.

[0020] The second optimization unit is used to optimize the second power forecast model; for the second type of photovoltaic power station in the cluster, after using the second power forecast model to forecast the power generation for a certain period of time, third historical data are obtained through data accumulation; the third historical data are divided into a third training set and a third verification set; the third training set is used to perform incremental sample training on the second power forecast model to obtain a third power forecast model; if the prediction performance of the third power forecast model on the third verification set is better than that of the second power forecast model, the second type of photovoltaic power station uses the third power forecast model to replace the second power forecast model for power forecasting.

[0021] Furthermore, the second power forecasting module includes an incremental sample training mode selection unit, The incremental sample training mode selection unit is used to select an incremental training mode; the incremental training mode includes: Method 1: freeze the output layer, use the third training set to perform incremental sample training on the second power prediction model, and adjust the parameters of the feature extraction layer; Method 2, dividing the third training set into two parts; freezing the output layer, using the first part of the third training set to perform incremental sample training on the second power prediction model, and adjusting the parameters of the feature extraction layer; freezing the feature extraction layer, using the second part of the third training set to perform incremental sample training on the second power prediction model, and adjusting the parameters of the output layer; Method 3: The feature extraction layer and the output layer are not frozen, and the third training set is used to perform incremental sample training on the second power prediction model, and the parameters of the feature extraction layer and the output layer are adjusted.

[0022] The present invention has achieved the following beneficial effects: for newly added photovoltaic power stations in a photovoltaic power station cluster, the prediction model suitable for the newly added photovoltaic power station can be quickly trained by using the prediction model of other types of power stations in the existing cluster and a small amount of sample data of the present invention, thus solving the problem that the newly added photovoltaic power station cannot be fully trained due to the scarcity of training samples. At the same time, with the operation of the photovoltaic power station cluster, the prediction model of the photovoltaic power station in the photovoltaic power station cluster is progressively updated by using the newly accumulated data, thus improving the generalization ability and prediction accuracy of the model, which is of great significance for the consumption of photovoltaic power and the safe operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and should not be regarded as limiting the scope.

[0024] Figure 1 It is a flow chart of a photovoltaic power station cluster power generation power forecasting method of the present invention; Figure 2 is a flow chart of fine-tuning the first power forecast model of the present invention; Figure 3 is a flow chart of the incremental sample training of the second power forecast model of the present invention; Figure 4 It is a schematic diagram of functional modules of a photovoltaic power station cluster power generation forecasting system of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0026] See also Figure 1 As shown in the flowchart, in the first aspect, the present invention provides a method for predicting power generation of a photovoltaic power station cluster, comprising the following steps: Step S101: for a photovoltaic power station of a first type in a cluster, a first power forecasting model obtained by a first training method is used to obtain a first power forecasting result.

[0027] In the present invention, the power stations in the photovoltaic power station cluster are divided into two types. The first type of photovoltaic power station has sufficient first historical data for training the first power forecast model. The amount of second historical data accumulated by the second type of photovoltaic power station cannot support the training of the first power forecast model according to the first training method. The first training method is a supervised training method or a semi-supervised training method, and the first power forecast model includes a feature extraction layer and an output layer. Based on the first historical data, the first power forecast model is trained using the first training method. The first power forecast model can adopt convolutional neural networks (CNN), recurrent neural networks (RNN), long short-term memory (LSTM), gated recurrent neural networks (Gated Recurrent Neural Network), or a combination or variant of the above-mentioned deep models, which is not specifically limited by the present invention.

[0028] Step S102: for the second type of photovoltaic power station in the cluster, fine-tune the first power forecast model to obtain a second power forecast model, and use the second power forecast model to obtain a second power forecast result.

[0029] Preferably, the first power forecast model is fine-tuned to obtain a second power forecast model, such as Figure 2 As shown, the method includes the following steps. Step S1021: divide the second historical data into a second training set and a second validation set; Step S1022: freeze the parameters of the feature extraction layer of the first power forecast model; Step S1023: train the first power forecast model using the second training set, and adjust the parameters of the output layer of the first power forecast model; Step S1024: evaluate the prediction performance of the fine-tuned first power forecast model in the second validation set; Step S1025: if the prediction error is greater than the set threshold, obtain the second power forecast model; otherwise, adjust the learning rate and re-execute steps S1021-S1025. The second type of photovoltaic power station itself has less accumulated data, which is not enough to fully train the deep learning model. The first power forecast model of other similar power stations in the cluster is selected, the parameters of the feature extraction layer remain unchanged, and the parameters of the output layer are adjusted using the second historical data accumulated by the second type of photovoltaic power station itself to obtain a second power forecast model suitable for the second type of photovoltaic power station. For the selection of other similar power stations, power stations with the same photovoltaic panel model, rated power, and inverter model are preferred.

[0030] Step S103: Summarize the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

[0031] The forecast results of various types of photovoltaic power stations in the cluster are collected and summarized, and the accumulated results are used as the power forecast results of the photovoltaic power station cluster.

[0032] As a preferred embodiment of the present invention, incremental sample training can be performed on the second type of photovoltaic power station, such as Figure 3 As shown. For the second type of photovoltaic power station in the cluster, after using the second power forecast model to forecast the power generation for a certain period of time, the third historical data is obtained through data accumulation. The third historical data is divided into a third training set and a third validation set. The third training set is used to perform incremental sample training on the second power forecast model to obtain a third power forecast model. If the prediction performance of the third power forecast model on the third validation set is better than that of the second power forecast model, the second type of photovoltaic power station uses the third power forecast model to replace the second power forecast model for power forecasting. The forecast model is incrementally trained through accumulated samples, which improves the forecast accuracy of the forecast model under long-term operation.

[0033] As a preferred embodiment of the present invention, the third training set is used to perform incremental sample training on the second power forecast model, specifically: Freeze the output layer, use the third training set to perform incremental sample training on the second power forecast model, and adjust the parameters of the feature extraction layer; or, divide the third training set into two parts; freeze the output layer, use the first part of the third training set to perform incremental sample training on the second power forecast model, and adjust the parameters of the feature extraction layer; freeze the feature extraction layer, use the second part of the third training set to perform incremental sample training on the second power forecast model, and adjust the parameters of the output layer; or, do not freeze the feature extraction layer and the output layer, use the third training set to perform incremental sample training on the second power forecast model, and adjust the parameters of the feature extraction layer and the output layer. Select an appropriate incremental training method based on the model complexity of the second power forecast model used and the amount of incremental sample data.

[0034] In addition, the present invention also provides a photovoltaic power station cluster power generation forecasting system, including a first power forecasting module, a second power forecasting module and a forecast result summary module, such as Figure 4 shown.

[0035] The first power forecasting module is used to perform power forecasting on a first type of photovoltaic power station in the cluster, and obtain a first power forecasting result using a first power forecasting model obtained by a first training method.

[0036] The second power forecasting module is used to forecast the power of the second type of photovoltaic power station in the cluster, fine-tune the first power forecasting model, obtain the second power forecasting model, and use the second power forecasting model to obtain the second power forecasting result.

[0037] Forecast result summary module: summarizes the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

[0038] As a preferred embodiment of the present invention, the second power forecasting module includes a model fine-tuning unit. The model fine-tuning unit: divides the second historical data into a second training set and a second validation set, freezes the parameters of the feature extraction layer of the first power forecasting model, trains the first power forecasting model using the second training set, adjusts the parameters of the output layer of the first power forecasting model, evaluates the prediction performance of the fine-tuned first power forecasting model in the second validation set, and obtains the second power forecasting model if the prediction error is greater than a set threshold; otherwise, adjusts the learning rate and re-fine-tunes the first power forecasting model using the second historical data.

[0039] As a preferred embodiment of the present invention, the first power forecasting module includes a first optimization unit. The first optimization unit is used to optimize the first power forecasting model; after using the first power forecasting model to forecast the power generation for a certain period of time, fourth historical data is obtained through data accumulation; the fourth historical data is divided into a fourth training set and a fourth validation set; the fourth training set is used to perform incremental sample training on the first power forecasting model to obtain a fourth power forecasting model; if the prediction performance of the fourth power forecasting model on the fourth validation set is better than that of the first power forecasting model, the first type of photovoltaic power station uses the fourth power forecasting model to replace the first power forecasting model for power forecasting.

[0040] As a preferred embodiment of the present invention, the second power forecasting module includes a second optimization unit. The second optimization unit is used to optimize the second power forecasting model; for the second type of photovoltaic power station in the cluster, after using the second power forecasting model to forecast the power generation for a certain period of time, third historical data is obtained through data accumulation; the third historical data is divided into a third training set and a third validation set; the third training set is used to perform incremental sample training on the second power forecasting model to obtain a third power forecasting model; if the prediction performance of the third power forecasting model on the third validation set is better than that of the second power forecasting model, the second type of photovoltaic power station uses the third power forecasting model to replace the second power forecasting model for power forecasting.

[0041] As a preferred embodiment of the present invention, the second power forecasting module includes an incremental sample training mode selection unit. The incremental sample training mode selection unit is used to select an incremental training mode; the incremental training mode includes: Method 1, freezing the output layer, using the third training set to perform incremental sample training on the second power prediction model, and adjusting the parameters of the feature extraction layer.

[0042] Method 2, divide the third training set into two parts; freeze the output layer, use the first part of the third training set to perform incremental sample training on the second power forecasting model, and adjust the parameters of the feature extraction layer; freeze the feature extraction layer, use the second part of the third training set to perform incremental sample training on the second power forecasting model, and adjust the parameters of the output layer.

[0043] Method 3: The feature extraction layer and the output layer are not frozen, and the third training set is used to perform incremental sample training on the second power prediction model, and the parameters of the feature extraction layer and the output layer are adjusted.

[0044] For newly added photovoltaic power stations in the photovoltaic power station cluster, it is difficult to model them due to insufficient historical data. The present invention uses the forecast models of other similar power stations in the existing cluster and a small amount of sample data of itself to quickly train a forecast model suitable for the newly added photovoltaic power station, solving the problem that the newly added photovoltaic power station cannot be fully trained due to the scarcity of training samples. In addition, as the photovoltaic power station cluster is running, the forecast model of the photovoltaic power station in the photovoltaic power station cluster is gradually updated using the newly accumulated data, which improves the long-term forecast accuracy of the cluster power station.

[0045] It should be pointed out that, according to the needs of implementation, the various components / steps described in the embodiments of the present application can be split into more components / steps, or two or more components / steps or partial operations of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of the present application.

[0046] The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as software or computer code that can be stored in a recording medium (such as a CD ROM, RAM, floppy disk, hard disk or magneto-optical disk), or implemented as a computer code originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded through a network and to be stored in a local recording medium, so that the device described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor or programmable or dedicated hardware (such as an ASIC or FPGA). It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, a processor or hardware, the device described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the device shown here, the execution of the code converts the general-purpose computer into a dedicated computer for executing the device shown here.

[0047] Those of ordinary skill in the art will appreciate that the units and device steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different devices to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present application.

[0048] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.

[0049] The above implementation methods are only used to illustrate the embodiments of the present application, and are not limitations on the embodiments of the present application. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the spirit and scope of the embodiments of the present application. Therefore, all equivalent technical solutions also belong to the scope of the embodiments of the present application. The scope of patent protection of the embodiments of the present application should be limited by the claims.

Claims

1. A method for predicting power generation of a photovoltaic power station cluster, characterized in that: The following steps are involved: Step S101: for a photovoltaic power station of a first type in a cluster, a first power forecasting model obtained by a first training method is used to obtain a first power forecasting result; Step S102: for a second type of photovoltaic power station in the cluster, fine-tune the first power forecast model to obtain a second power forecast model, and use the second power forecast model to obtain a second power forecast result; Step S103: Summarize the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

2. A photovoltaic power station cluster power generation prediction method according to claim 1, characterized in that: The photovoltaic power stations in the cluster are divided into a first type of photovoltaic power station and a second type of photovoltaic power station according to the amount of historical data accumulation; the first type of photovoltaic power station uses the first historical data accumulated by itself to train a first power forecast model; the second type of photovoltaic power station uses the second historical data accumulated by itself to fine-tune the first power forecast model; the first training method is a supervised training method or a semi-supervised training method, and the first power forecast model includes a feature extraction layer and an output layer.

3. A photovoltaic power station cluster power generation prediction method according to claim 2, characterized in that: The first power forecast model is fine-tuned to obtain the second power forecast model, specifically: Step S1021: Divide the second historical data into a second training set and a second validation set; Step S1022: Freeze the parameters of the feature extraction layer of the first power prediction model; Step S1023: training the first power prediction model using the second training set, and adjusting the parameters of the output layer of the first power prediction model; Step S1024: evaluating the prediction performance of the fine-tuned first power prediction model on the second validation set; Step S1025: If the prediction error is greater than the set threshold, obtain a second power forecast model; otherwise, adjust the learning rate and re-execute steps S1021-S1025.

4. A photovoltaic power station cluster power generation prediction method according to claim 3, characterized in that: For the second type of photovoltaic power station in the cluster, after using the second power forecast model to forecast the power generation for a certain period, third historical data are obtained through data accumulation; the third historical data are divided into a third training set and a third validation set; the third training set is used to perform incremental sample training on the second power forecast model to obtain a third power forecast model; if the prediction performance of the third power forecast model on the third validation set is better than that of the second power forecast model, the second type of photovoltaic power station uses the third power forecast model to replace the second power forecast model for power forecasting.

5. A photovoltaic power station cluster power generation forecasting method according to claim 4, characterized in that: The use of the third training set to perform incremental sample training on the second power forecast model is specifically as follows: Freeze the output layer, perform incremental sample training on the second power prediction model using the third training set, and adjust parameters of the feature extraction layer; or, The third training set is divided into two parts; the output layer is frozen, and the second power prediction model is trained with incremental samples using the first part of the third training set, and the parameters of the feature extraction layer are adjusted; the feature extraction layer is frozen, and the second power prediction model is trained with incremental samples using the second part of the third training set, and the parameters of the output layer are adjusted; Alternatively, the feature extraction layer and the output layer are not frozen, and the third training set is used to perform incremental sample training on the second power prediction model, and the parameters of the feature extraction layer and the output layer are adjusted.

6. A photovoltaic power station cluster power generation forecasting system, used to execute the photovoltaic power station cluster power generation forecasting method according to any one of claims 1 to 5, characterized in that: Includes the following functional modules: A first power forecasting module: used to forecast power for a first type of photovoltaic power station in the cluster, and obtain a first power forecasting result using a first power forecasting model obtained by a first training method; The second power forecasting module is used to forecast the power of the second type of photovoltaic power station in the cluster, fine-tune the first power forecasting model to obtain the second power forecasting model, and use the second power forecasting model to obtain the second power forecasting result; Forecast result summary module: summarizes the first power forecast result and the second power forecast result in the cluster to obtain the power generation forecast result of the photovoltaic power station cluster.

7. A photovoltaic power station cluster power generation forecasting system according to claim 6, characterized in that: The second power forecasting module includes a model fine-tuning unit, Model fine-tuning unit: divide the second historical data into a second training set and a second validation set, freeze the parameters of the feature extraction layer of the first power forecasting model, use the second training set to train the first power forecasting model, adjust the parameters of the output layer of the first power forecasting model, evaluate the prediction performance of the fine-tuned first power forecasting model in the second validation set, if the prediction error is greater than the set threshold, obtain the second power forecasting model; otherwise, adjust the learning rate and use the second historical data to re-fine-tune the first power forecasting model.

8. A photovoltaic power station cluster power generation forecasting system according to claim 7, characterized in that: The first power forecasting module includes a first optimization unit, The first optimization unit is used to optimize the first power forecast model; after using the first power forecast model to forecast the power generation for a certain period of time, fourth historical data is obtained through data accumulation; the fourth historical data is divided into a fourth training set and a fourth validation set; the fourth training set is used to perform incremental sample training on the first power forecast model to obtain a fourth power forecast model; if the prediction performance of the fourth power forecast model on the fourth validation set is better than that of the first power forecast model, the first type of photovoltaic power station uses the fourth power forecast model to replace the first power forecast model for power forecasting.

9. A photovoltaic power station cluster power generation forecasting system according to claim 8, characterized in that: The second power forecasting module includes a second optimization unit, The second optimization unit is used to optimize the second power forecast model; for the second type of photovoltaic power station in the cluster, after using the second power forecast model to forecast the power generation for a certain period of time, third historical data are obtained through data accumulation; the third historical data are divided into a third training set and a third verification set; the third training set is used to perform incremental sample training on the second power forecast model to obtain a third power forecast model; if the prediction performance of the third power forecast model on the third verification set is better than that of the second power forecast model, the second type of photovoltaic power station uses the third power forecast model to replace the second power forecast model for power forecasting.

10. A photovoltaic power station cluster power generation forecasting system according to claim 9, characterized in that: The second power forecasting module includes an incremental sample training mode selection unit, The incremental sample training mode selection unit is used to select an incremental training mode; the incremental training mode includes: Method 1: freeze the output layer, use the third training set to perform incremental sample training on the second power prediction model, and adjust the parameters of the feature extraction layer; Method 2, dividing the third training set into two parts; freezing the output layer, using the first part of the third training set to perform incremental sample training on the second power prediction model, and adjusting the parameters of the feature extraction layer; freezing the feature extraction layer, using the second part of the third training set to perform incremental sample training on the second power prediction model, and adjusting the parameters of the output layer; Method 3: The feature extraction layer and the output layer are not frozen, and the third training set is used to perform incremental sample training on the second power prediction model, and the parameters of the feature extraction layer and the output layer are adjusted.

Images