Incremental photovoltaic power generation prediction method, device, equipment, storage medium and product

Abstract

The invention discloses an incremental photovoltaic power generation prediction method, device and equipment, a storage medium and a product. The method comprises the following steps: acquiring newly added sequence data of photovoltaic power generation in real time; screening out a target sequence subset from the old sequence data based on a first distribution difference with the existing old sequence data and a second distribution difference with the newly added sequence data; generating a target training sample according to the target sequence subset and the newly added sequence data, and training a current original prediction model according to the target training sample to update and obtain an incremental prediction model; and predicting the future photovoltaic power generation power according to the incremental prediction model. According to the method and the system, dependency on task boundaries is avoided, so that time sequence prediction characteristics of photovoltaic power generation are met, the method and the system can adapt to continuously changing data environments, the sampling process of previous data is simplified, representativeness of sampling samples and generalization of a sampling method are improved, computing resources are saved, and prediction accuracy and timeliness are improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of photovoltaic power generation technology, and in particular to an incremental photovoltaic power generation prediction method, device, equipment, storage medium and product. Background Art

[0002] Photovoltaic power generation has been widely studied because of its simple way of obtaining energy. Since the amount of photovoltaic power generation has a direct impact on the stability and reliability of the power grid operation, the prediction of photovoltaic power generation has been listed as a research focus. Photovoltaic power generation prediction is a continuous, periodic and complex process. Weather factors such as temperature and humidity will affect the prediction results, and these factors may change at any time in the next few hours, which may cause the previous model to have a decrease in accuracy when predicting. In order to solve this problem, online continuous learning methods can usually be used to update the model in real time so that the model can maintain a stable prediction for subsequent data. Continuous learning is a method that combines new data streams to train models while retaining previous data information as much as possible. Its main research problem is to overcome the catastrophic forgetting of neural networks, that is, after learning new knowledge, the previously learned content is almost completely forgotten. This dilemma is one aspect of the trade-off between learning plasticity and memory stability. This is also very suitable for the scenario where the amount of data arrives in the form of a stream when predicting photovoltaic power generation.

[0003] At present, the mainstream methods of continuous learning can be divided into three categories. One is the method centered on prior knowledge, which uses penalty terms to constrain parameters. This type of method focuses on how to constrain the model from the importance of parameters and how to dynamically adjust the network structure / number of parameters of the model. The parameter-level algorithm based on regularization must maintain a parameter importance matrix equal to the size of the model for each incremental learning stage, which causes its storage cost to increase linearly over time. In addition, because different preferences for each parameter at different stages may lead to conflicts between different parameter importance matrices, the general applicability of this type of method is limited. When the task chain is long, the parameters will gradually deviate from the feasible domain. At this time, the prior-based method is generally combined with the replay-based method to obtain better results. The other is the dynamic model architecture method, which dynamically changes the model composition or generates a new model to adapt to specific task requirements. This type of method mainly focuses on how to establish a connection between the new and old models and achieve the purpose of knowledge transfer. Copying models and knowledge distillation belong to this type of method. However, only considering the constraints of the old model on the new model limits the characteristics of the model, making the model tend to maintain the extraction of features that are conducive to discriminating old data, which is not conducive to learning new data. Both of the above strategies deeply bind the neural network to a certain task, which results in very clear task boundaries during training and testing, which also means that the model is dependent on the task boundaries. These methods prefer multi-task learning rather than continuous incremental learning, and are not very suitable for photovoltaic power generation prediction, because the time series data of photovoltaic power generation arrives in a streaming manner, will continue to deform, and there is no clear task boundary division. Another method is based on data replay, which stores information in the sample space or directly in the replay buffer. When learning new knowledge, copy old examples from the replay buffer, or use a generative model to simulate the generation of old data for retraining or as a constraint on the current learning knowledge, that is, actively review previously learned knowledge when learning new knowledge. Regarding how to construct a replay buffer, existing technologies have adopted methods such as reservoir sampling, uncertainty assessment under disturbance based on data augmentation, deep generative replay, and sample selection based on a gradient-based greedy algorithm that does not rely on task boundary information and the iid assumption. These methods have problems such as complex sampling processes that increase computational overhead, insufficient sampling when facing non-uniformly distributed data, resulting in a lack of representativeness of sample data, and poor generalization when facing fuzzy task boundaries or highly correlated tasks. Therefore, how to effectively construct a replay buffer remains a challenge. Summary of the invention

[0004] Embodiments of the present invention provide an incremental photovoltaic power generation prediction method, device, equipment, storage medium and product to avoid dependence on task boundaries, save computing resources, and improve the accuracy and timeliness of predictions.

[0005] In a first aspect, an embodiment of the present invention provides an incremental photovoltaic power generation prediction method, the method comprising:

[0006] Acquire new series data of photovoltaic power generation in real time;

[0007] Screening the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data;

[0008] Generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model;

[0009] The future photovoltaic power generation power is predicted according to the incremental prediction model.

[0010] Optionally, the selecting the target sequence subset from the old sequence data based on the first distribution difference between the old sequence data and the existing old sequence data and the second distribution difference between the new sequence data and the newly added sequence data includes:

[0011] Dividing the old sequence data into subsequences with equal amounts and with a preset step length;

[0012] respectively calculating the first distribution difference and the second distribution difference of each of the subsequences;

[0013] Calculating a screening score for each of the subsequences according to the first distribution difference and the second distribution difference of each of the subsequences, wherein the screening score is used to indicate the degree of being close to the distribution of the old sequence data and deviating from the distribution of the newly added sequence data;

[0014] The target sequence subset is obtained by selecting and combining the subsequences according to the screening scores.

[0015] Optionally, the calculation formula of the screening score is:

[0016] C i =C2 i -λC1 i ;

[0017] λ = L / T;

[0018] Among them, C i represents the screening score of the ith subsequence, C1 i represents the first distribution difference of the i-th subsequence, C2 i represents the second distribution difference of the i-th subsequence, λ represents the weight coefficient, L represents the length of the newly added sequence data, and T represents the length of the old sequence data.

[0019] Optionally, selecting and combining the subsequences according to the screening scores to obtain the target sequence subset includes:

[0020] The screening scores are sorted, and the larger L / e corresponding subsequence combinations are selected to obtain the target sequence subset, where e represents the preset step length.

[0021] Optionally, the preset step size is the greatest common divisor between the length of the newly added sequence data and the length of the old sequence data.

[0022] Optionally, the first distribution difference and the second distribution difference are both maximum mean differences.

[0023] In a second aspect, an embodiment of the present invention further provides an incremental photovoltaic power generation prediction device, the device comprising:

[0024] Incremental data acquisition module, used to acquire new sequence data of photovoltaic power generation in real time;

[0025] A sequence subset screening module, configured to screen the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data;

[0026] A prediction model updating module, used to generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model;

[0027] The power generation prediction module is used to predict the future photovoltaic power generation according to the incremental prediction model.

[0028] In a third aspect, an embodiment of the present invention further provides a computer device, the computer device comprising:

[0029] one or more processors;

[0030] A memory for storing one or more programs;

[0031] When the one or more programs are executed by the one or more processors, the one or more processors implement the incremental photovoltaic power generation prediction method provided by any embodiment of the present invention.

[0032] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the incremental photovoltaic power generation prediction method provided by any embodiment of the present invention.

[0033] In a fifth aspect, an embodiment of the present invention further provides a computer program product, which includes a computer program, and when the program is executed by a processor, the incremental photovoltaic power generation prediction method provided by any embodiment of the present invention is implemented.

[0034] The embodiment of the present invention provides an incremental photovoltaic power generation prediction method, firstly, the newly added sequence data of photovoltaic power generation is obtained in real time, then the target sequence subset is screened from the old sequence data based on the first distribution difference between the existing old sequence data and the second distribution difference between the newly added sequence data, and then the target training sample is generated according to the target sequence subset and the newly added sequence data, and the current original prediction model is trained according to the target training sample to update the incremental prediction model, so as to use the incremental prediction model to predict the future photovoltaic power generation power. The incremental photovoltaic power generation prediction method provided by the embodiment of the present invention operates at the data level, screens the previous data based on the distribution difference between the old sequence data and the newly added sequence data, and continuously updates the prediction model in combination with the newly arrived data, thereby avoiding dependence on the task boundary, thereby meeting the time series prediction characteristics of photovoltaic power generation, and being able to adapt to the ever-changing data environment, and at the same time simplifies the sampling process of the previous data, improves the representativeness of the sampling samples and the generalization of the sampling method, thereby saving computing resources, improving the accuracy and timeliness of the prediction, and can also be combined with other online learning methods to achieve better prediction performance, and will not affect which algorithm the model uses to process the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flowchart of an incremental photovoltaic power generation prediction method provided in Embodiment 1 of the present invention;

[0036] Figure 2 A schematic diagram of SMAPE comparison results of four strategies under four models provided in Example 1 of the present invention;

[0037] Figure 3 A schematic diagram showing the comparison results of the training time of three strategies under four models provided in Example 1 of the present invention;

[0038] Figure 4 A schematic diagram of the structure of an incremental photovoltaic power generation prediction device provided in Embodiment 2 of the present invention;

[0039] Figure 5 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0041] It should be mentioned before discussing the exemplary embodiments in more detail that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0042] Embodiment 1

[0043] Figure 1 This is a flowchart of the incremental photovoltaic power generation prediction method provided in the first embodiment of the present invention. This embodiment can be applied to the situation where the prediction model is updated at any time to ensure the prediction accuracy during the process of the factors affecting the photovoltaic power generation prediction constantly changing. It can also be applied to other application scenarios with similar characteristics. It is independent of the task type and has strong scalability. The method can be executed by the incremental photovoltaic power generation prediction device provided in the embodiment of the present invention. The device can be implemented by hardware and / or software and can generally be integrated into a computer device. Figure 1 As shown, the specific steps include:

[0044] S11. Acquire new sequence data of photovoltaic power generation in real time.

[0045] S12: Filter the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the newly added sequence data.

[0046] S13: Generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model.

[0047] S14. Predicting future photovoltaic power generation power according to the incremental prediction model.

[0048] In order to adapt to the photovoltaic power generation prediction scenario, online experience accumulation can be used to enable certain experiences to be stored for a long time in limited memory, thereby incorporating prior knowledge into the current learning process. Specifically, a replay buffer can be used as a long-term storage mechanism, that is, within limited resources, selectively store past experiences to alleviate the problem of catastrophic forgetting. Further, a data sampling-based method can be used to fill the replay buffer, so that only a small number of previous data samples are used instead of the whole, and combined with incremental data to achieve online incremental learning without retraining the model from scratch, thereby saving computing resources and ensuring the timeliness of the prediction.

[0049] For the photovoltaic power generation prediction process, for example, given a photovoltaic task dataset Where n represents the data sampling time point, and m represents the user station area. Photovoltaic power generation power P t =[x t,1 ,x t,2 ,…,x t,m ] is a time-varying series, whose prediction is defined as [P t ,P t+1 ,…,P t+p ]=f(P t-q ,P t-q+1 ,…,P t-1 ), where f represents the photovoltaic power generation prediction network model, p is the time step required for prediction, and q is the time step of the input sequence. The update of the prediction model based on the sequence data of time step D at time t can be defined as Here, update is the update strategy of incremental learning. There is no restriction on the specific prediction model architecture used here.

[0050] Specifically, the latest relevant data for prediction model training can be obtained in real time, and the current newly added sequence data H = [P T ,P T+1 ,…,P T+L ], and can determine the existing old sequence data G = [P1,P t ,…,P T], wherein T represents the length of the old sequence data, which can be a preset value or increase with time, L represents the length of the newly added sequence data, that is, the preset time step, and T≥L. After obtaining the newly added sequence data, the samples that are most different from the newly added sequence data and can represent the old sequence data as much as possible can be screened out from the corresponding old sequence data to form a target sequence subset to fill the replay buffer, thereby matching the global distribution and fully considering the uncertainty of the data, so that the prediction model can obtain more comprehensive data. Specifically, the first distribution difference between the distribution of streaming data within a period of time in the old sequence data and the entire old sequence data can be compared, in order to retain more features of the old sequence data, that is, features that can better represent the old sequence data as a whole, and at the same time, the second distribution difference between the distribution of streaming data in each period of time and the newly added sequence data is compared accordingly, in order to consider the diversity of continuous learning data. Optionally, the first distribution difference and the second distribution difference are both maximum mean discrepancy (MMD), and accordingly, the strategy adopted to fill the replay buffer can be recorded as IL-MMDS. That is, MMD can be used as a consistency measurement method between two distributions. The smaller the MMD value, the more consistent the probability distribution between the two random samples. The calculation formula is as follows, where K represents the Gaussian kernel function, σ 2 Represents the nuclear radius:

[0051]

[0052] After obtaining the target sequence subset S, the target sequence subset S and the newly added sequence data H can be randomly combined into batch data and input into the current original prediction model for continued training to update the model parameters and obtain the incremental prediction model. For the updated incremental prediction model, it not only learns the previous data characteristics, but also learns the subsequent data knowledge, thereby improving the prediction ability of the model. In addition, only a small number of samples from the old sequence data are used to replace the whole, which greatly shortens the update training time. After the prediction model is updated, the updated incremental prediction model can be used to predict the future photovoltaic power generation. At the same time, you can still continue to obtain new additional sequence data and perform the above model update process again, that is, use the incremental prediction model obtained above as the original prediction model in the re-update process and update it again to obtain a new incremental prediction model to ensure the prediction ability of the model at any time.

[0053] On the basis of the above technical solution, optionally, the method of screening the target sequence subset from the old sequence data based on the first distribution difference with the existing old sequence data and the second distribution difference with the newly added sequence data includes: dividing the old sequence data into subsequences with a preset step length in equal amounts; respectively calculating the first distribution difference and the second distribution difference of each subsequence; calculating a screening score for each subsequence according to the first distribution difference and the second distribution difference of each subsequence, the screening score being used to indicate the degree of closeness to the distribution of the old sequence data and deviation from the distribution of the newly added sequence data; and selecting and combining the subsequences according to the screening scores to obtain the target sequence subset.

[0054] Specifically, the old sequence data G can be equally divided into subsequences S with a preset step length e. i To traverse the entire old sequence data G, that is, G = [S1, S2, ..., S E ], where E represents the number of subsequences, E = T / e. Thus, each subsequence S can be calculated i The MMD between the entire old sequence data G is denoted as C1 i =MMD(S i ,G), and each subsequence S i The MMD between the newly added sequence data H is denoted as C2 i =MMD(S i ,H), where i∈[1,E] represents the subsequence index. At the same time, a screening scoring algorithm can be designed, and the C1 of each subsequence i Value and C2 i The respective screening scores are calculated based on the values ​​to balance the selection of data close to the old sequence data and deviating from the new sequence data distribution, thereby selecting a target sequence subset that is closer to the old sequence data distribution and more different from the new sequence data distribution.

[0055] Optionally, the calculation formula of the screening score is:

[0056] C i =C2 i -λC1 i ;

[0057] λ = L / T;

[0058] Among them, C i represents the screening score of the ith subsequence, C1 i represents the first distribution difference of the i-th subsequence, C2 irepresents the second distribution difference of the i-th subsequence, λ represents the weight coefficient, L represents the length of the newly added sequence data, and T represents the length of the old sequence data. Specifically, the smaller the first distribution difference is and the larger the second distribution difference is, the larger the corresponding screening score is, which means that the distribution is closer to the old sequence data and deviates from the distribution of the newly added sequence data. At the same time, if the newly added sequence data is less than the old sequence data, the weight coefficient is smaller, which is relatively more inclined to value the impact of the second distribution difference.

[0059] Further optionally, the selecting and combining the target sequence subset from the subsequences according to the screening scores includes: sorting the screening scores, and selecting L / e corresponding subsequences with the largest size to combine to obtain the target sequence subset, where e represents the preset step length. This can make the length of the target sequence subset more reasonable, shortening the training time while ensuring the model update effect.

[0060] Further optionally, the preset step size is the greatest common divisor between the length of the newly added sequence data and the length of the old sequence data, that is, e=gcd(T,L).

[0061] The effect of the above method is evaluated by a specific example. The existing photovoltaic power generation data set can be used for experimental verification. The data set contains 10872 data, each of which contains weather data of 68 substations, and the sampling frequency of the data is 1h. At the same time, four different prediction models are used to verify the effect, including linear Dlinear, Nlinear, MLP and nonlinear RNN models. Short-term (next 24 hours) power prediction of 68 substations is performed on these four models. Specifically, 45%, 5%, and 5% divisions are selected to obtain the original training data, validation set, and test set, and 5% of the new sequence data is added in turn in 9 incremental simulations for incremental learning (recorded as increments 1-9). At the same time, three baselines are selected, namely: (1) the model trained with a fixed amount of data, which is used as the lower bound of accuracy comparison (recorded as Base); (2) the model trained with the full amount of data, which has the largest training set size, so it is theoretically the upper bound of the accuracy of all models (recorded as Expand); (3) the Greedy method based on gradient sampling. The symmetric mean absolute percentage error (SMAPE) is then used to measure the performance of each model. All experiments use Pytorch and run on a single Nvidia RTX3090 24G.

[0062] The SMAPE comparison results of the four strategies under the four models are as follows: Figure 2As shown, the red mark is the optimal value. Among them, the Base strategy represents the basic model trained with a data set of the lower limit size, and only switches different test sets as benchmarks to compare the SMAPE values ​​of other methods after training. The Expand strategy directly splices the newly added sequence data with the previous training set, and then retrains all the data for model learning. Both the IL-MMDS strategy and the Greedy strategy use the training set of the previous round of Expand increments as past data, sample from them to fill the replay buffer, and then train the model with the newly added sequence data. It can be seen that when the first 7 scales of data sets are incrementally trained, the SMAPE value of the IL-MMDS strategy is slightly better than that of the Greedy strategy, and is also better than the Expand strategy that uses the entire data set as the training set. The reason is that when making short-term predictions, by combining the newly added sequence data with the replay buffer data, the model can focus more on the current newly added data and the most important past experience, so better results are obtained than training with all the data. When performing incremental training on the latter two datasets, the IL-MMDS strategy sometimes cannot outperform the Expand strategy. The reason is that for short-term predictions, when the model size is small, when new sequence data arrives, the model needs to learn less past experience again, so it can better grasp the key features of past data. As a result, when the dataset size increases to a certain extent, the IL-MMDS strategy cannot outperform the Expand strategy. Therefore, the IL-MMDS strategy has the best effect of incremental training on small datasets.

[0063] The training time comparison results of three strategies (excluding Base strategy) under four models are as follows: Figure 3 As shown in the figure, it can be seen that the training time of the IL-MMDS strategy on different models is much shorter than that of the Expand strategy and the Greedy strategy. In summary, the IL-MMDS strategy is an efficient update strategy with good prediction performance. At the same time, the experiment based on photovoltaic power generation data also verifies the effectiveness of this update strategy in the photovoltaic power generation prediction scenario.

[0064] The technical solution provided by the embodiment of the present invention first obtains the newly added sequence data of photovoltaic power generation in real time, then selects the target sequence subset from the old sequence data based on the first distribution difference between the existing old sequence data and the second distribution difference between the newly added sequence data, and then generates the target training sample according to the target sequence subset and the newly added sequence data, and trains the current original prediction model according to the target training sample to update the incremental prediction model, so as to use the incremental prediction model to predict the future photovoltaic power generation power. By operating at the data level, the previous data is selected based on the distribution difference between the old sequence data and the newly added sequence data, and the prediction model is continuously updated in combination with the newly arrived data, avoiding dependence on the task boundary, thereby meeting the time series prediction characteristics of photovoltaic power generation, and being able to adapt to the ever-changing data environment. At the same time, it also simplifies the sampling process of the previous data, improves the representativeness of the sampling samples and the generalization of the sampling method, thereby saving computing resources, improving the accuracy and timeliness of the prediction, and can also be combined with other online learning methods to achieve better prediction performance, and will not affect which algorithm the model uses to process the data.

[0065] Embodiment 2

[0066] Figure 4 This is a schematic diagram of the structure of the incremental photovoltaic power generation prediction device provided in the second embodiment of the present invention. The device can be implemented by hardware and / or software, and can generally be integrated into a computer device to execute the incremental photovoltaic power generation prediction method provided in any embodiment of the present invention. Figure 4 As shown, the device comprises:

[0067] The incremental data acquisition module 41 is used to acquire the newly added sequence data of photovoltaic power generation in real time;

[0068] A sequence subset screening module 42, configured to screen the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data;

[0069] A prediction model updating module 43, configured to generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model;

[0070] The power generation prediction module 44 is used to predict the future photovoltaic power generation according to the incremental prediction model.

[0071] The technical solution provided by the embodiment of the present invention first obtains the newly added sequence data of photovoltaic power generation in real time, then selects the target sequence subset from the old sequence data based on the first distribution difference between the existing old sequence data and the second distribution difference between the newly added sequence data, and then generates the target training sample according to the target sequence subset and the newly added sequence data, and trains the current original prediction model according to the target training sample to update the incremental prediction model, so as to use the incremental prediction model to predict the future photovoltaic power generation power. By operating at the data level, the previous data is selected based on the distribution difference between the old sequence data and the newly added sequence data, and the prediction model is continuously updated in combination with the newly arrived data, avoiding dependence on the task boundary, thereby meeting the time series prediction characteristics of photovoltaic power generation, and being able to adapt to the ever-changing data environment. At the same time, it also simplifies the sampling process of the previous data, improves the representativeness of the sampling samples and the generalization of the sampling method, thereby saving computing resources, improving the accuracy and timeliness of the prediction, and can also be combined with other online learning methods to achieve better prediction performance, and will not affect which algorithm the model uses to process the data.

[0072] On the basis of the above technical solution, optionally, the sequence subset screening module 42 includes:

[0073] A subsequence division unit, used for dividing the old sequence data into subsequences with a preset step length in equal amounts;

[0074] a distribution difference calculation unit, configured to respectively calculate the first distribution difference and the second distribution difference of each of the subsequences;

[0075] a screening score calculation unit, configured to calculate a screening score for each of the subsequences according to the first distribution difference and the second distribution difference of each of the subsequences, wherein the screening score is used to indicate a degree of closeness to the distribution of the old sequence data and deviation from the distribution of the newly added sequence data;

[0076] A sequence subset determination unit is used to select and combine the subsequences according to the screening scores to obtain the target sequence subset.

[0077] On the basis of the above technical solution, optionally, the calculation formula of the screening score is:

[0078] C i =C2 i -λC1 i ;

[0079] λ = L / T;

[0080] Among them, C i represents the screening score of the ith subsequence, C1 irepresents the first distribution difference of the i-th subsequence, C2 i represents the second distribution difference of the i-th subsequence, λ represents the weight coefficient, L represents the length of the newly added sequence data, and T represents the length of the old sequence data.

[0081] Based on the above technical solution, optionally, the sequence subset determination unit is specifically used to:

[0082] The screening scores are sorted, and the larger L / e corresponding subsequence combinations are selected to obtain the target sequence subset, where e represents the preset step length.

[0083] Based on the above technical solution, optionally, the preset step size is the greatest common divisor between the length of the newly added sequence data and the length of the old sequence data.

[0084] Based on the above technical solution, optionally, the first distribution difference and the second distribution difference are both maximum mean differences.

[0085] The incremental photovoltaic power generation prediction device provided in the embodiment of the present invention can execute the incremental photovoltaic power generation prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0086] It is worth noting that in the above-mentioned embodiment of the incremental photovoltaic power generation prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0087] Embodiment 3

[0088] Figure 5 The schematic diagram of the structure of the computer device provided for the third embodiment of the present invention shows a block diagram of an exemplary computer device suitable for implementing the implementation mode of the present invention. Figure 5 The computer device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention. Figure 5 As shown, the computer device includes a processor 51, a memory 52, an input device 53 and an output device 54; the number of processors 51 in the computer device can be one or more. Figure 5 Taking a processor 51 as an example, the processor 51, memory 52, input device 53 and output device 54 in the computer device can be connected through a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0089] The memory 52, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the incremental photovoltaic power generation prediction method in the embodiment of the present invention (for example, the incremental data acquisition module 41, the sequence subset screening module 42, the prediction model update module 43 and the power generation prediction module 44 in the incremental photovoltaic power generation prediction device). The processor 51 executes various functional applications and data processing of the computer device by running the software programs, instructions and modules stored in the memory 52, that is, realizing the above-mentioned incremental photovoltaic power generation prediction method.

[0090] The memory 52 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 52 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 52 may further include a memory remotely arranged relative to the processor 51, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0091] The input device 53 can be used to obtain new sequence data of photovoltaic power generation in real time, and generate key signal input related to user settings and function control of computer equipment, etc. The output device 54 can include a display screen, which can be used to show the prediction results of photovoltaic power generation to the user, etc.

[0092] Embodiment 4

[0093] Embodiment 4 of the present invention further provides a storage medium comprising computer executable instructions, wherein the computer executable instructions are used to execute an incremental photovoltaic power generation prediction method when executed by a computer processor, the method comprising:

[0094] Acquire new series data of photovoltaic power generation in real time;

[0095] Screening the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data;

[0096] Generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model;

[0097] The future photovoltaic power generation power is predicted according to the incremental prediction model.

[0098] The storage medium may be any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROM, floppy disk or tape device; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disk or optical storage); registers or other similar types of memory elements, etc. The storage medium may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the computer system in which the program is executed, or may be located in a different second computer system, which is connected to the computer system via a network (such as the Internet). The second computer system may provide program instructions to the computer for execution. The term "storage medium" may include two or more storage media that may reside in different locations (e.g., in different computer systems connected via a network). The storage medium may store program instructions (e.g., embodied as a computer program) that may be executed by one or more processors.

[0099] Of course, the computer executable instructions of a storage medium containing computer executable instructions provided in an embodiment of the present invention are not limited to the method operations described above, and can also execute related operations in the incremental photovoltaic power generation prediction method provided in any embodiment of the present invention.

[0100] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0101] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0102] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0103] Embodiment 5

[0104] Embodiment 5 of the present invention also provides a computer program product, which includes a computer program (also referred to as code, instruction), which can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it is used to execute the incremental photovoltaic power generation prediction method provided in any of the above embodiments, and has the corresponding beneficial effects of the execution method.

[0105] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. An incremental photovoltaic power generation prediction method, characterized in that: include: Acquire new series data of photovoltaic power generation in real time; Screening the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data; Generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model; The future photovoltaic power generation power is predicted according to the incremental prediction model.

2. The incremental photovoltaic power generation prediction method according to claim 1, characterized in that: The step of selecting a target sequence subset from the old sequence data based on a first distribution difference with the existing old sequence data and a second distribution difference with the newly added sequence data includes: Dividing the old sequence data into subsequences with equal amounts and with a preset step length; respectively calculating the first distribution difference and the second distribution difference of each of the subsequences; Calculating a screening score for each of the subsequences according to the first distribution difference and the second distribution difference of each of the subsequences, wherein the screening score is used to indicate the degree of being close to the distribution of the old sequence data and deviating from the distribution of the newly added sequence data; The target sequence subset is obtained by selecting and combining the subsequences according to the screening scores.

3. The incremental photovoltaic power generation prediction method according to claim 2, characterized in that: The calculation formula of the screening score is: C i =C2 i -λC1 i ; λ = L / T; Among them, C i represents the screening score of the ith subsequence, C1 i represents the first distribution difference of the i-th subsequence, C2 i represents the second distribution difference of the i-th subsequence, λ represents the weight coefficient, L represents the length of the newly added sequence data, and T represents the length of the old sequence data.

4. The incremental photovoltaic power generation prediction method according to claim 3, characterized in that: The step of selecting and combining the subsequences according to the screening scores to obtain the target sequence subset comprises: The screening scores are sorted, and the larger L / e corresponding subsequence combinations are selected to obtain the target sequence subset, where e represents the preset step length.

5. The incremental photovoltaic power generation prediction method according to any one of claims 2 to 4, characterized in that: The preset step size is the greatest common divisor between the length of the newly added sequence data and the length of the old sequence data.

6. The incremental photovoltaic power generation prediction method according to claim 1, characterized in that: The first distribution difference and the second distribution difference are both maximum mean differences.

7. An incremental photovoltaic power generation prediction device, characterized in that: include: Incremental data acquisition module, used to acquire new sequence data of photovoltaic power generation in real time; A sequence subset screening module, configured to screen the target sequence subset from the old sequence data based on a first distribution difference between the old sequence data and the existing old sequence data and a second distribution difference between the new sequence data and the old sequence data; A prediction model updating module, used to generate a target training sample according to the target sequence subset and the newly added sequence data, and train the current original prediction model according to the target training sample to update the incremental prediction model; The power generation prediction module is used to predict the future photovoltaic power generation according to the incremental prediction model.

8. A computer device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the incremental photovoltaic power generation prediction method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the incremental photovoltaic power generation prediction method as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the incremental photovoltaic power generation prediction method as described in any one of claims 1 to 6.

Images