A photovoltaic power prediction method and a photovoltaic power prediction model training method

Abstract

The application is a photovoltaic power prediction method, and particularly relates to the technical field of photovoltaic power. The method comprises the following steps: obtaining a target input sequence; performing sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term; performing time feature extraction on the target input sequence to obtain a first time vector and a second time vector; processing the target input sequence and the first time vector according to an encoder in a photovoltaic power prediction model to obtain an encoding result; inputting the initialization period term, the second time vector and the encoding result into a first branch of a decoder in the photovoltaic power prediction model to obtain a first sub-result output by the decoder; inputting the initialization trend term into a second branch of the decoder to obtain a second sub-result; and determining a photovoltaic power result based on a fusion result of the first sub-result and the second sub-result. The photovoltaic power result predicted based on the above scheme has high accuracy.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power technology, specifically to a photovoltaic power prediction method and a photovoltaic power prediction model training method. Background Technology

[0002] Photovoltaic power generation has developed rapidly in recent years due to its advantages such as being clean and pollution-free, having flexible application forms, and being safe and reliable.

[0003] However, photovoltaic (PV) power generation exhibits significant intermittent and random fluctuations. As the penetration rate of PV power generation in the power grid continues to increase, it poses a significant challenge to the real-time dynamic balance of power generation, transmission, and consumption in the power system, severely restricting the security of PV power generation. Existing technologies typically employ algorithms such as Long Short-Term Memory (LSTM) networks and gradient boosting tree algorithms for PV power generation prediction.

[0004] However, the above methods have poor accuracy in predicting photovoltaic power generation. Summary of the Invention

[0005] This application provides a photovoltaic power prediction method and a photovoltaic power prediction model training method, which improves the accuracy of photovoltaic power prediction. The technical solution is as follows.

[0006] On the one hand, a photovoltaic power prediction method is provided, the method comprising:

[0007] Obtain the target input sequence; the target input sequence is used to indicate the photovoltaic power generation status of the photovoltaic equipment under the corresponding weather conditions within the target time period;

[0008] The target input sequence is decomposed to obtain the initialization period term and the initialization trend term;

[0009] Temporal features are extracted from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data points in a specified interval of the target input sequence.

[0010] The encoder in the photovoltaic power prediction model processes the target input sequence and the first time vector to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0011] The initialization periodic term, the second time vector, and the encoding result are input into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence.

[0012] The initial trend term is input into the second branch of the decoder to obtain the second sub-result;

[0013] The photovoltaic power result is determined based on the fusion of the first sub-result and the second sub-result.

[0014] On another front, a method for training a photovoltaic power prediction model is provided, the method comprising:

[0015] Obtain the sample input sequence and the sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions within a first sample time period; the sample label is the photovoltaic power generation of the second sample time period after the first sample time period;

[0016] The sample input sequence is decomposed to obtain the initial sample period term and the initial sample trend term;

[0017] Temporal features are extracted from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data points in a specified interval of the sample input sequence.

[0018] The sample input sequence and the first sample time vector are processed by the encoder in the photovoltaic power prediction model to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0019] The initial sample periodic term, the second sample vector, and the sample encoding result are input into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence.

[0020] Input the initial sample trend term into the second branch of the decoder; obtain the second sample sub-result;

[0021] Based on the first sample sub-result and the second sample sub-result, the sample photovoltaic power result is determined;

[0022] Based on the sample photovoltaic power results and the sample labels, the photovoltaic power prediction model is updated to obtain the updated photovoltaic power prediction model.

[0023] In another aspect, a photovoltaic power prediction device is provided, the device comprising:

[0024] An input sequence acquisition module is used to acquire a target input sequence; the target input sequence is used to indicate the photovoltaic power generation status of the photovoltaic equipment under corresponding weather conditions within a target time period.

[0025] The sequence decomposition module is used to perform sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term.

[0026] The time extraction module is used to extract time features from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data points in a specified interval of the target input sequence.

[0027] The encoding module is used to process the target input sequence and the first time vector according to the encoder in the photovoltaic power prediction model to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0028] The decoding module is used to input the initialization periodic term, the second time vector, and the encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch of the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence; the initialization trend term is input into the second branch of the decoder to obtain the second sub-result;

[0029] The power prediction module is used to determine the photovoltaic power result based on the fusion result of the first sub-result and the second sub-result.

[0030] In one possible implementation, the sequence decomposition module is used to perform an average pooling operation on the target input sequence to obtain an initial trend term;

[0031] The initialization period term is generated based on the difference between the target input sequence and the initialization trend term.

[0032] In one possible implementation, the time extraction module is further configured to extract features from the timestamps of each data point in the target input sequence to obtain timestamp data corresponding to each data point; the timestamp data is used to indicate at least one time position; the time position includes the number of minutes in the current hour, the number of hours in the current day, the number of days in the current week, the number of days in the current month, and the number of days in the current year;

[0033] The first time vector is generated based on the timestamp data corresponding to each data point in the target input sequence.

[0034] The second time vector is generated based on the timestamp information corresponding to the data in the specified time interval of the target input sequence.

[0035] In one possible implementation, the input sequence acquisition module is further configured to acquire target power generation data and target weather data; the target power generation data is used to indicate the photovoltaic power generation status of the photovoltaic equipment within a target time period; the target weather data is used to indicate the weather conditions within the target time period.

[0036] Historical forecast irradiance and historical measured irradiance are obtained; the historical forecast irradiance is used to indicate the expected irradiance within the target time period; the historical measured irradiance is used to indicate the actual irradiance within the target time period.

[0037] An irradiance error is generated based on the difference between the historical predicted irradiance and the historical measured irradiance.

[0038] The target input sequence is generated based on the target power generation data, target weather data, and irradiance error.

[0039] In one possible implementation, the photovoltaic power prediction model is an Autoformer model; the decoding module is further configured to input the initialization periodic term and the second time vector into the first branch of the decoder, so as to process them through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector;

[0040] The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result.

[0041] In one possible implementation, the decoding module is further configured to input the initialization trend term into the second branch of the decoder, and sequentially fuse it with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit to obtain the second sub-result.

[0042] In another aspect, a photovoltaic power prediction model training device is provided, the device comprising:

[0043] The sample acquisition module is used to acquire a sample input sequence and a sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions within a first sample time period; the sample label is the photovoltaic power generation of the equipment in a second sample time period after the first sample time period.

[0044] The sample sequence decomposition module is used to decompose the sample input sequence to obtain the initial sample period term and the initial sample trend term.

[0045] The sample time extraction module is used to extract time features from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data points in a specified interval of the sample input sequence.

[0046] The sample encoding module is used to process the sample input sequence and the first sample time vector according to the encoder in the photovoltaic power prediction model to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0047] The sample decoding module is used to input the initial sample periodic term, the second sample vector, and the sample encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch of the decoder includes at least one pair of sequentially connected autocorrelation mechanism units and sequence decomposition units; the initial sample trend term is input into the second branch of the decoder; and the second sample sub-result is obtained.

[0048] The sample prediction module is used to determine the sample photovoltaic power result based on the first sample sub-result and the second sample sub-result;

[0049] The model update module is used to update the photovoltaic power prediction model based on the sample photovoltaic power results and the sample labels, so as to obtain the updated photovoltaic power prediction model.

[0050] In another aspect, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement the above-described photovoltaic power prediction method or photovoltaic power prediction model training method.

[0051] In another aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement the above-described photovoltaic power prediction method or photovoltaic power prediction model training method.

[0052] In another aspect, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform a photovoltaic power prediction method or a photovoltaic power prediction model training method.

[0053] The technical solution provided in this application may include the following beneficial effects:

[0054] To predict photovoltaic (PV) power, computer equipment can obtain the PV power generation of photovoltaic devices under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained from the second branch of the decoder by inputting the initialization trend term to determine the PV power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the structure of a photovoltaic power prediction system according to an exemplary embodiment.

[0057] Figure 2 This is a flowchart illustrating a photovoltaic power prediction method according to an exemplary embodiment.

[0058] Figure 3 This is a flowchart illustrating a photovoltaic power prediction model training method according to an exemplary embodiment.

[0059] Figure 4 This is a flowchart illustrating a photovoltaic power prediction method according to an exemplary embodiment.

[0060] Figure 5 This document illustrates an input data processing flowchart for a model input according to an embodiment of this application.

[0061] Figure 6 A schematic diagram of the structure of a photovoltaic power prediction model according to an embodiment of this application is shown.

[0062] Figure 7 This is a structural block diagram of a photovoltaic power prediction device according to an exemplary embodiment.

[0063] Figure 8 This is a structural block diagram of a photovoltaic power prediction model training device according to an exemplary embodiment.

[0064] Figure 9 A structural block diagram of a computer device illustrated in an exemplary embodiment of this application is shown. Detailed Implementation

[0065] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] It should be understood that the term "instruction" mentioned in the embodiments of this application can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0067] In the description of the embodiments of this application, the term "correspondence" may indicate that there is a direct or indirect correspondence between two things, or that there is an association between two things, or that there is a relationship of instruction and being instructed, configuration and being configured, etc.

[0068] In the embodiments of this application, "predefined" can be achieved by pre-storing corresponding codes, tables or other means that can be used to indicate relevant information in the device (e.g., including terminal devices and network devices). This application does not limit the specific implementation method.

[0069] Figure 1 This is a schematic diagram illustrating the structure of a photovoltaic power prediction system according to an exemplary embodiment. The photovoltaic power prediction system includes a data processing device 110 and a current acquisition device 120.

[0070] Optionally, the current acquisition device 120 includes a data storage device. After acquiring the target current and obtaining the target current data, the current acquisition device can store the current data in the data storage device. For example, the current acquisition device can be a current sensor, a current acquisition instrument, or a current acquisition unit.

[0071] Optionally, the data processing device 110 can be a computer device with high computing power, which is used to analyze the collected target current data to obtain the characteristics of the target current data.

[0072] Optionally, the data processing device 110 can be a terminal device with current analysis software installed. When the terminal device receives an instruction to analyze current data, it can read the corresponding current data from the data storage in the current acquisition device 120 and analyze the current data to obtain the characteristics of the target current data.

[0073] Optionally, the target current data can be historical photovoltaic power generation data within a certain period of time. The terminal equipment can process the target current data through a photovoltaic power prediction model to predict the power generation of the photovoltaic equipment in the future period.

[0074] Optionally, the data processing device 110 can also be a server with current analysis software installed. The current acquisition device can be a terminal device. After the terminal device acquires the target current data, it can transmit the target current data to the server to complete the prediction of the power generation of the photovoltaic equipment in the future time period.

[0075] Optionally, the data processing device 110 and the current acquisition device 120 can communicate via a wired or wireless network.

[0076] Optionally, the aforementioned server can be a server cluster or a distributed system consisting of multiple physical servers, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other technology computing services.

[0077] Optionally, the system may also include a management device for managing the system (such as managing the connection status between each module and the server), and the management device is connected to the server via a communication network. Optionally, the communication network may be a wired network or a wireless network.

[0078] Optionally, the aforementioned wireless or wired networks use standard communication technologies and / or protocols. The network is typically the Internet, but can also be any other network, including but not limited to any combination of local area networks (LANs), metropolitan area networks (MANs), wide area networks (WANs), mobile, wired or wireless networks, private networks, or virtual private networks (VPNs). In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network. Furthermore, conventional encryption technologies such as Secure Sockets Layer (SSL), Transport Layer Security (TLS), VPNs, and Internet Protocol (IP) security can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0079] Figure 2 This is a flowchart illustrating a photovoltaic power prediction method according to an exemplary embodiment. The method is executed by a computer device, which may be, for example... Figure 1 The data processing device 110 shown is an example. Figure 2 As shown, the photovoltaic power prediction method may include the following steps:

[0080] Step 201: Obtain the target input sequence.

[0081] This target input sequence is used to indicate the photovoltaic power generation of photovoltaic equipment under corresponding weather conditions within a target time period.

[0082] When it is necessary to predict the power generation of photovoltaic equipment in the future (e.g., 72 hours), the computer equipment can first obtain the photovoltaic power generation of the photovoltaic equipment in the corresponding weather during the target time period (e.g., the first 48 hours) and generate the target input sequence.

[0083] In other words, in this embodiment of the application, the target input sequence includes photovoltaic power generation data within the target time period, as well as weather data corresponding to the photovoltaic power generation data.

[0084] In one possible implementation, target power generation data and target weather data are acquired; the target power generation data is used to indicate the photovoltaic power generation status of the photovoltaic equipment within a target time period; the target weather data is used to indicate the weather conditions within the target time period.

[0085] Obtain historical forecast irradiance and historical measured irradiance; the historical forecast irradiance is used to indicate the expected irradiance within the target time period; the historical measured irradiance is used to indicate the actual irradiance within the target time period.

[0086] Irradiance error is generated based on the difference between the historical predicted irradiance and the historical measured irradiance.

[0087] The target input sequence is generated based on the target's power generation data, target weather data, and irradiance error.

[0088] In this embodiment, when the photovoltaic power prediction model processes target power generation data and target weather data to predict the power generation of photovoltaic equipment over a future period, the input to the photovoltaic power prediction model is the power generation data and weather data, and the corresponding output should also be the power generation data and weather data for the future period. That is to say, in this embodiment, the photovoltaic power prediction model can predict not only power generation data but also future weather data. Therefore, the relationship between weather forecast data and power generation is actually considered during the processing of the photovoltaic power prediction model.

[0089] Therefore, in this embodiment of the application, irradiance error, target power generation data, and target weather data can be concatenated into a target input sequence, that is, the feature of irradiance error is added to the target input sequence. The irradiance error is defined as the difference between the predicted irradiance (i.e., historical forecast irradiance) and the actual irradiance (historical measured irradiance). At this time, the irradiance error reflects whether a meteorological change has occurred within a certain period of time, and the irradiance fluctuation when photoelectric conversion is carried out under the environment of the meteorological change.

[0090] If the irradiance error is incorporated into the target input sequence, the photovoltaic power prediction model can take into account the impact of severe weather conditions such as rain and snow on photovoltaic power generation when processing the target input sequence to predict the photovoltaic power generation situation in the future. This improves the accuracy of photovoltaic power generation prediction under severe weather conditions and enhances the anti-interference and robustness of the photovoltaic power prediction model.

[0091] Step 202: Perform sequence decomposition on the target input sequence to obtain the initial periodic term and the initial trend term.

[0092] After obtaining the target input sequence, the target input sequence can be decomposed. For example, the initial trend term in the target input sequence can be obtained through average pooling. At this time, the error between the target input sequence and the initial trend term is the initial period term.

[0093] Step 203: Extract time features from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data in a specified interval of the target input sequence.

[0094] In this embodiment, time features can also be extracted from the target input sequence, and the timestamps of each data point in the target input sequence can be extracted as a first time vector. In this case, the first time vector represents the overall time features of the target input sequence. Simultaneously, timestamps can also be extracted from data within a specified interval of the target input sequence, and the extracted second time vector represents the features of that specified interval.

[0095] Optionally, the specified interval can be customized as needed. For example, the specified interval can be the specified position at the end of the target input sequence. In this case, the extracted second time vector represents the time characteristics of the most recent data of the photovoltaic device.

[0096] Step 204: Process the target input sequence and the first time vector according to the encoder in the photovoltaic power prediction model to obtain the encoding result.

[0097] The encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0098] After obtaining the target input sequence and the first time vector, the target input sequence and the first time vector can be encoded, and the encoded result can be input into the encoder in the photovoltaic power prediction model for processing to obtain the encoding result.

[0099] In this embodiment of the application, the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0100] Step 205: Input the initial periodic term, the second time vector, and the encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence.

[0101] In this embodiment of the application, the decoder includes at least two processing branches, wherein the first branch is used to process the initialization periodic term, the second time vector and the encoding result to obtain the first sub-result output by the decoder.

[0102] The first branch includes at least one pair of sequentially connected autocorrelation mechanism units and sequence decomposition units. Each time the autocorrelation mechanism unit performs feature extraction based on the autocorrelation mechanism, the sequence decomposition unit decomposes the periodic and trend terms. After several operations, the final first sub-result has more accurate periodic term features.

[0103] Step 206: Input the initial trend term into the second branch of the decoder to obtain the second sub-result.

[0104] Optionally, in the second branch of the decoder, the trend term decomposed by the sequence decomposition unit in the first branch can be concatenated with the initial trend term to finally obtain the second sub-result.

[0105] In this step, the second sub-result represents the extracted trend features.

[0106] Step 207: Based on the fusion result of the first sub-result and the second sub-result, determine the photovoltaic power result.

[0107] After fusing the first and second sub-results, the fused result can characterize the trend of photovoltaic power change over a period of time in the future, as well as the periodic changes of photovoltaic power caused by factors such as sunshine duration in a short period of time. Therefore, each data in the fused result can characterize the photovoltaic power result at each point in time in the future, which is the photovoltaic power generation situation in the future.

[0108] In summary, to achieve photovoltaic power prediction, the computer equipment can obtain the photovoltaic power generation of photovoltaic equipment under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained by inputting the initialization trend term into the second branch of the decoder to determine the photovoltaic power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy.

[0109] Figure 3 This is a flowchart illustrating a photovoltaic power prediction model training method according to an exemplary embodiment. The method is executed by a computer device, which may be, for example... Figure 1 The data processing device 110 shown is an example. Figure 3 As shown, the training method for this photovoltaic power prediction model may include the following steps:

[0110] Step 301: Obtain the sample input sequence and the sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions during the first sample time period; the sample label is the photovoltaic power generation during the second sample time period after the first sample time period.

[0111] In one possible implementation, sample power generation data and sample weather data are acquired; the target power generation data is used to indicate the photovoltaic power generation status of the photovoltaic equipment during a first sample time period; the sample weather data is used to indicate the weather conditions during the first sample time period.

[0112] Obtain the predicted irradiance and the measured irradiance of the sample; the predicted irradiance is used to indicate the expected irradiance during the first sample time period; the measured irradiance is used to indicate the actual irradiance during the first sample time period.

[0113] The sample irradiance error is generated based on the difference between the predicted irradiance and the measured irradiance of the sample.

[0114] The sample input sequence is generated based on the sample power generation data, sample weather data, and sample irradiance error.

[0115] The method of obtaining the sample input sequence is similar to that of... Figure 2 The method for obtaining the target input sequence shown is similar and will not be repeated here.

[0116] Step 302: Perform sequence decomposition on the input sequence of the sample to obtain the initial sample period term and the initial sample trend term.

[0117] Step 303: Extract time features from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data in a specified interval of the sample input sequence.

[0118] Step 304: The encoder in the photovoltaic power prediction model processes the sample input sequence and the first sample time vector to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0119] Step 305: Input the initial sample periodic term, the second sample vector, and the sample encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence.

[0120] Step 306: Input the initial sample trend term into the second branch of the decoder; obtain the second sample sub-result.

[0121] Step 307: Based on the first sample sub-result and the second sample sub-result, determine the sample photovoltaic power result.

[0122] Step 308: Based on the photovoltaic power results and the sample label, update the photovoltaic power prediction model to obtain the updated photovoltaic power prediction model.

[0123] The training process of the above photovoltaic power prediction model is as follows: Figure 2 The steps of the photovoltaic power prediction scheme shown are similar and will not be repeated in the embodiments of this application.

[0124] In summary, to achieve photovoltaic power prediction, the computer equipment can obtain the photovoltaic power generation of photovoltaic equipment under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained by inputting the initialization trend term into the second branch of the decoder to determine the photovoltaic power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy.

[0125] Figure 4 This is a flowchart illustrating a photovoltaic power prediction method according to an exemplary embodiment. The method is executed by a computer device, which may be, for example... Figure 1 The data processing device 110 shown is an example. Figure 4 As shown, the photovoltaic power prediction method may include the following steps:

[0126] Step 401: Obtain the target input sequence.

[0127] Please refer to Figure 5 This illustrates a flowchart of input data processing for a model input according to an embodiment of this application. Figure 5 As shown in this embodiment, the computer device can randomly extract time series from historical time data, setting the input sequence length seq_len to 96, the label sequence label_len to 48, and the sequence length to be predicted pred_len to 288. The label sequence is a part of the input sequence and is mainly used to help the encoder predict the required subsequent sequences. N is the size of each batch fed into the model, and M is the number of weather features.

[0128] Optionally, the target input sequence can also be obtained based on target power generation data, target weather data, and irradiance error. For specific implementation details, please refer to [example missing]. Figure 2 The embodiments shown are not described in detail here.

[0129] Optionally, in this embodiment, each data point in the target input sequence is standardized data. The standardization process involves applying the mean and standard deviation (std) of the data obtained statistically over a period of time to all data. Standardization is performed, and the standardization formula is as follows:

[0130]

[0131] Step 402: Perform average pooling on the target input sequence to obtain the initial trend term.

[0132] Step 403: Generate the initialization period term based on the difference between the target input sequence and the initialization trend term.

[0133] Once the computer device obtains the target input sequence, it can perform sequence decomposition on the target input sequence, namely seq_x(N,96,M), to obtain the trend term (N,96,M) and the period term (N,96,M).

[0134] Alternatively, the principle of sequence decomposition is as follows:

[0135] Based on the concept of moving average, a sequence can be decomposed into a trend term and a periodic term, reflecting the long-term and periodic nature of the sequence respectively, smoothing out the periodic term and highlighting the trend term.

[0136]

[0137]

[0138] Where X is the latent variable to be decomposed, X tX s These are the trend item and the cycle item, respectively.

[0139] Furthermore, the computer device takes the last label_len portion (N, 48, M) of the periodic term and concatenates it with a zero tensor (N, 288, M) of length pred_len to obtain a tensor of size (N, 336, M). Similarly, it takes the last label_len portion (N, 48, M) of the trend term and concatenates it with the mean tensor of size (N, 288, M) calculated by seq_x, thus completing the initialization of the periodic and trend terms.

[0140] Step 404: Extract features from the timestamps of each data point in the target input sequence to obtain the timestamp data corresponding to each data point.

[0141] The timestamp data is used to indicate at least one time location; the time location includes the number of minutes in the current hour, the number of hours in the current day, the number of days in the current week, the number of days in the current month, and the number of days in the current year.

[0142] Step 405: Generate the first time vector based on the timestamp data corresponding to each data point in the target input sequence.

[0143] Step 406: Generate the second time vector based on the timestamp information corresponding to the data in the specified time interval of the target input sequence.

[0144] In this embodiment of the application, the specified time interval can be the timestamp information corresponding to the last specified number of bits of data in the target input sequence.

[0145] In one possible implementation, such as Figure 5 As shown, in the computer device, the last label_len data (i.e., the data within the specified time interval) of the target input sequence seq_x is selected and concatenated with a 0 sequence of length pred_len to form seq_y, which serves as the input sequence for the decoder. Simultaneously, time features of seq_x and seq_y are extracted in minutes: the current timestamp is calculated, corresponding to the minute of the current hour, the hour of the current day, the day of the current week, the day of the current month, and the day of the current year, respectively, yielding time feature vectors seq_x_mark and seq_y_mark (i.e., the first time vector and the second time vector).

[0146] Step 407: The encoder in the photovoltaic power prediction model processes the target input sequence and the first time vector to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0147] In this embodiment, the photovoltaic power prediction model can be a trained Autoformer model. The training process of the Autoformer model is similar to the application process of the model described in this embodiment, so it will not be described in detail.

[0148] Please refer to Figure 6 This illustrates a schematic diagram of the structure of a photovoltaic power prediction model according to an embodiment of this application. Figure 6 As shown, the computer device can input the tensor (N,96,512) obtained by encoding seq_x(N,96,M) and seq_x_mark(N,96,5) into the encoder. It passes through the autocorrelation mechanism unit, sequence decomposition unit, feedforward network, sequence decomposition and two shortcut connections, and is repeated multiple times to finally obtain the output enc_out of (N,96,512).

[0149] Step 408: Input the initialization periodic term and the second time vector into the first branch of the decoder, so as to process them through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector.

[0150] like Figure 6 As shown, the computer device inputs the tensor (N,336,512) obtained by encoding the initialization periodic term (N,336,M) and seq_y_mark(N,336,5) into the first branch of the decoder, and processes it through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector. At this time, the intermediate vector can be used as the query of the next autocorrelation mechanism unit (that is, the second autocorrelation mechanism unit), and the key and value are taken from the encoder output (that is, the encoding result) enc_out.

[0151] Step 409: The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result.

[0152] After processing the intermediate vector and the encoding result through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit respectively, the first sub-result can be obtained. In this step, for the periodic term, the autocorrelation mechanism utilizes the periodicity of the sequence to aggregate sub-sequences with similar processes in different periods. Therefore, the first sub-result contains more accurate periodic term features.

[0153] Step 410: Input the initialization trend term into the second branch of the decoder, and merge it sequentially with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit to obtain the second sub-result.

[0154] When the first, second, and third sequence decomposition units in the first branch perform sequence decomposition, in addition to generating periodic terms, they also generate corresponding trend terms (i.e., the first sub-trend term, the second sub-trend term, and the third sub-trend term). At this time, in the second branch, the first and second sub-trend terms can be fused with the initial periodic terms in sequence. That is, the decoder uses an accumulation method to gradually extract trend information from the predicted latent variables. At this time, the second sub-result contains more accurate trend term features.

[0155] Step 411: Based on the fusion result of the first sub-result and the second sub-result, determine the photovoltaic power result.

[0156] The fusion result of the first sub-result and the second sub-result is the last dimension (N,288,1) of the last pred_len length of the decoder output dec_out. The computer device performs inverse normalization on this result and outputs the power prediction values ​​of 288 points to be predicted. These 288 power prediction values ​​represent the power prediction situation at each time point in the future.

[0157] In summary, to achieve photovoltaic power prediction, the computer equipment can obtain the photovoltaic power generation of the photovoltaic equipment under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained by inputting the initialization trend term into the second branch of the decoder to determine the photovoltaic power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy.

[0158] Figure 7 This is a structural block diagram illustrating a photovoltaic power prediction device according to an exemplary embodiment. The photovoltaic power prediction device includes:

[0159] The input sequence acquisition module 701 is used to acquire a target input sequence; the target input sequence is used to indicate the photovoltaic power generation status of the photovoltaic equipment under corresponding weather conditions within a target time period.

[0160] Sequence decomposition module 702 is used to perform sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term;

[0161] The time extraction module 703 is used to extract time features from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data points in a specified interval of the target input sequence.

[0162] The encoding module 704 is used to process the target input sequence and the first time vector according to the encoder in the photovoltaic power prediction model to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0163] The decoding module 705 is used to input the initialization periodic term, the second time vector, and the encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence; the initialization trend term is input into the second branch of the decoder to obtain the second sub-result;

[0164] The power prediction module 706 is used to determine the photovoltaic power result based on the fusion result of the first sub-result and the second sub-result.

[0165] In one possible implementation, the sequence decomposition module is used to perform an average pooling operation on the target input sequence to obtain an initial trend term;

[0166] The initialization period term is generated based on the difference between the target input sequence and the initialization trend term.

[0167] In one possible implementation, the time extraction module is further configured to extract features from the timestamps of each data point in the target input sequence to obtain timestamp data corresponding to each data point; the timestamp data is used to indicate at least one time position; the time position includes the number of minutes in the current hour, the number of hours in the current day, the number of days in the current week, the number of days in the current month, and the number of days in the current year;

[0168] The first time vector is generated based on the timestamp data corresponding to each data point in the target input sequence.

[0169] The second time vector is generated based on the timestamp information corresponding to the data in the specified time interval of the target input sequence.

[0170] In one possible implementation, the photovoltaic power prediction model is an Autoformer model; the decoding module is further configured to input the initialization periodic term and the second time vector into the first branch of the decoder, so as to process them through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector;

[0171] The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result.

[0172] In one possible implementation, the decoding module is further configured to input the initialization trend term into the second branch of the decoder, and sequentially fuse it with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit to obtain the second sub-result.

[0173] In summary, to achieve photovoltaic power prediction, the computer equipment can obtain the photovoltaic power generation of the photovoltaic equipment under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained by inputting the initialization trend term into the second branch of the decoder to determine the photovoltaic power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy.

[0174] Figure 8 This is a structural block diagram illustrating a photovoltaic power prediction model training device according to an exemplary embodiment. The photovoltaic power prediction device includes:

[0175] The sample acquisition module 801 is used to acquire a sample input sequence and a sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions within a first sample time period; the sample label is the photovoltaic power generation of the photovoltaic equipment in a second sample time period after the first sample time period.

[0176] The sample sequence decomposition module 802 is used to decompose the sample input sequence to obtain the initial sample period term and the initial sample trend term.

[0177] The sample time extraction module 803 is used to extract time features from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data points in a specified interval of the sample input sequence.

[0178] The sample encoding module 804 is used to process the sample input sequence and the first sample time vector according to the encoder in the photovoltaic power prediction model to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence.

[0179] The sample decoding module 805 is used to input the initial sample periodic term, the second sample vector, and the sample encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch of the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence; the initial sample trend term is input into the second branch of the decoder; and the second sample sub-result is obtained.

[0180] The sample prediction module 806 is used to determine the sample photovoltaic power result based on the first sample sub-result and the second sample sub-result;

[0181] The model update module 807 is used to update the photovoltaic power prediction model based on the sample photovoltaic power results and the sample labels to obtain the updated photovoltaic power prediction model.

[0182] In summary, to achieve photovoltaic power prediction, the computer equipment can obtain the photovoltaic power generation of the photovoltaic equipment under corresponding weather conditions during the target time period from historical power generation data and convert it into a target input sequence. The computer equipment then performs sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The computer equipment then extracts time features from the target input sequence to obtain a first time vector related to the timestamps of each data point in the target input sequence, and a second time vector related to the timestamps of data within a specified interval of the target input sequence. The target input sequence and its corresponding first time vector are processed by the autocorrelation mechanism unit and sequence decomposition unit in the encoder to extract the corresponding features and form the encoding result. The computer equipment then inputs the initialization period term, the second time vector, and the encoding result output by the encoder into the decoder. The decoder is processed by the autocorrelation mechanism unit and sequence decomposition unit to obtain a first sub-result. During the feature extraction process, the first sub-result continuously eliminates the trend term through the autocorrelation mechanism unit and sequence decomposition unit. Finally, the first sub-result is fused with the second sub-result obtained by inputting the initialization trend term into the second branch of the decoder to determine the photovoltaic power result. The above scheme fully considers the overall time characteristics and specific time characteristics of the target input sequence. Furthermore, the encoder and decoder gradually refine the periodic term and fuse it with the trend term, so that the first sub-result and the second sub-result can more accurately represent the period and trend of photovoltaic power generation in the future. Therefore, the fusion of the first sub-result and the second sub-result obtained by the photovoltaic power prediction model in this application embodiment can yield a fused result. The photovoltaic power result obtained based on this fused result has high accuracy. Moreover, the above photovoltaic power prediction model can greatly reduce the computational complexity and improve the prediction efficiency of photovoltaic power generation while ensuring prediction accuracy.

[0183] Figure 9 A structural block diagram of a computer device 900 illustrating an exemplary embodiment of this application is shown. This computer device can be implemented as a server as described above in this application. The computer device 900 includes a Central Processing Unit (CPU) 901, a system memory 904 including Random Access Memory (RAM) 902 and Read-Only Memory (ROM) 903, and a system bus 905 connecting the system memory 904 and the CPU 901. The computer device 900 also includes a mass storage device 906 for storing an operating system 909, application programs 910, and other program modules 911.

[0184] The mass storage device 906 is connected to the central processing unit 901 via a mass storage controller (not shown) connected to the system bus 905. The mass storage device 906 and its associated computer-readable media provide non-volatile storage for the computer device 900. That is, the mass storage device 906 may include computer-readable media (not shown) such as a hard disk or a compact disc read-only memory (CD-ROM) drive.

[0185] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state storage technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape cassettes, magnetic tape, disk storage, or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage media are not limited to the above-mentioned types. The system memory 904 and mass storage device 906 described above can be collectively referred to as memory.

[0186] According to various embodiments of this disclosure, the computer device 900 can also be connected to a remote computer on a network, such as the Internet. That is, the computer device 900 can be connected to a network 908 via a network interface unit 907 connected to the system bus 905, or it can use the network interface unit 907 to connect to other types of networks or remote computer systems (not shown).

[0187] The memory also includes at least one computer program stored in the memory, and the central processing unit 901 executes the at least one computer program to implement all or part of the steps in the methods shown in the above embodiments.

[0188] In one exemplary embodiment, a computer-readable storage medium is also provided for storing at least one computer program, which is loaded and executed by a processor to implement all or part of the steps in the above-described method. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage device, etc.

[0189] In one exemplary embodiment, a computer program product or computer program is also provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned actions. Figure 2 or Figure 3 All or part of the steps of the method shown in any embodiment.

[0190] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0191] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A photovoltaic power prediction method, characterized in that, The method includes: Obtain the target input sequence; the target input sequence is used to indicate the photovoltaic power generation status of the photovoltaic equipment under the corresponding weather conditions within the target time period; The target input sequence is decomposed to obtain the initialization period term and the initialization trend term; Temporal features are extracted from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data points in a specified interval of the target input sequence. The target input sequence and the first time vector are processed by the encoder in the photovoltaic power prediction model to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence, and the photovoltaic power prediction model is an Autoformer model; The initialization periodic term, the second time vector, and the encoding result are input into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence. The initial trend term is input into the second branch of the decoder to obtain the second sub-result; Based on the fusion result of the first sub-result and the second sub-result, the photovoltaic power result is determined; The step of inputting the initialization periodic term, the second time vector, and the encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder includes: The initialization periodic term and the second time vector are input into the first branch of the decoder to be processed by the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector; The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result; The step of inputting the initialized trend term into the second branch of the decoder to obtain the second sub-result includes: The initial trend term is input into the second branch of the decoder and then fused with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit in sequence to obtain the second sub-result.

2. The method according to claim 1, characterized in that, The target input sequence is decomposed to obtain an initial periodic term and an initial trend term, including: Perform average pooling on the target input sequence to obtain an initial trend term; The initialization period term is generated based on the difference between the target input sequence and the initialization trend term.

3. The method according to claim 1, characterized in that, The step of extracting time features from the target input sequence to obtain a first time vector and a second time vector includes: The timestamps of each data point in the target input sequence are used to extract features, thereby obtaining the timestamp data corresponding to each data point. The timestamp data is used to indicate at least one time position. The time position includes the number of minutes in the current hour, the number of hours in the current day, the number of days in the current week, the number of days in the current month, and the number of days in the current year. The first time vector is generated based on the timestamp data corresponding to each data point in the target input sequence. The second time vector is generated based on the timestamp information corresponding to the data in the specified time interval of the target input sequence.

4. The method according to claim 1, characterized in that, The acquisition of the target input sequence includes: Acquire target power generation data and target weather data; the target power generation data is used to indicate the photovoltaic power generation status of the photovoltaic equipment within a target time period; the target weather data is used to indicate the weather conditions within the target time period; Historical forecast irradiance and historical measured irradiance are obtained; the historical forecast irradiance is used to indicate the expected irradiance within the target time period; the historical measured irradiance is used to indicate the actual irradiance within the target time period. An irradiance error is generated based on the difference between the historical predicted irradiance and the historical measured irradiance. The target input sequence is generated based on the target power generation data, target weather data, and irradiance error.

5. A method for training a photovoltaic power prediction model, characterized in that, The method includes: Obtain the sample input sequence and the sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions within a first sample time period; the sample label is the photovoltaic power generation of the second sample time period after the first sample time period; The sample input sequence is decomposed to obtain the initial sample period term and the initial sample trend term; Temporal features are extracted from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data points in a specified interval of the sample input sequence. The sample input sequence and the first sample time vector are processed by the encoder in the photovoltaic power prediction model to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence, and the photovoltaic power prediction model is an Autoformer model; The initial sample periodic term, the second sample vector, and the sample encoding result are input into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch in the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence. Input the initial sample trend term into the second branch of the decoder; obtain the second sample sub-result; Based on the first sample sub-result and the second sample sub-result, the sample photovoltaic power result is determined; Based on the sample photovoltaic power results and the sample labels, the photovoltaic power prediction model is updated to obtain the updated photovoltaic power prediction model; The step of inputting the initial sample periodicity term, the second sample vector, and the sample encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder includes: The initialization periodic term and the second time vector are input into the first branch of the decoder to be processed by the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector; The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result; The step of inputting the initialized trend term into the second branch of the decoder to obtain the second sub-result includes: The initial trend term is input into the second branch of the decoder and then fused with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit in sequence to obtain the second sub-result.

6. A photovoltaic power prediction device, characterized in that, The device includes: An input sequence acquisition module is used to acquire a target input sequence; the target input sequence is used to indicate the photovoltaic power generation status of the photovoltaic equipment under corresponding weather conditions within a target time period. The sequence decomposition module is used to perform sequence decomposition on the target input sequence to obtain an initialization period term and an initialization trend term. The time extraction module is used to extract time features from the target input sequence to obtain a first time vector and a second time vector; the first time vector is used to indicate the timestamps of each data point in the target input sequence; the second time vector is used to indicate the timestamps of data points in a specified interval of the target input sequence. The encoding module is used to process the target input sequence and the first time vector according to the encoder in the photovoltaic power prediction model to obtain the encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence, and the photovoltaic power prediction model is an Autoformer model; The decoding module is used to input the initialization periodic term, the second time vector, and the encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sub-result output by the decoder; the first branch of the decoder includes at least one pair of autocorrelation mechanism units and sequence decomposition units connected in sequence; the initialization trend term is input into the second branch of the decoder to obtain the second sub-result; The power prediction module is used to determine the photovoltaic power result based on the fusion result of the first sub-result and the second sub-result; The decoding module is specifically used to input the initialization periodic term and the second time vector into the first branch of the decoder, and process them through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain an intermediate vector; process the intermediate vector and the encoding result sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain a first sub-result; input the initialization trend term into the second branch of the decoder, and fuse it sequentially with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit to obtain a second sub-result.

7. A photovoltaic power prediction model training device, characterized in that, The device includes: The sample acquisition module is used to acquire a sample input sequence and a sample label corresponding to the sample input sequence; the sample input sequence is used to indicate the photovoltaic power generation of the photovoltaic equipment under the corresponding weather conditions within a first sample time period; the sample label is the photovoltaic power generation of the equipment in a second sample time period after the first sample time period. The sample sequence decomposition module is used to decompose the sample input sequence to obtain the initial sample period term and the initial sample trend term. The sample time extraction module is used to extract time features from the sample input sequence to obtain a first sample vector and a second sample vector; the first sample vector is used to indicate the timestamps of each data point in the sample input sequence; the second sample vector is used to indicate the timestamps of data points in a specified interval of the sample input sequence. The sample encoding module is used to process the sample input sequence and the first sample time vector according to the encoder in the photovoltaic power prediction model to obtain the sample encoding result; the encoder includes an autocorrelation mechanism unit and a sequence decomposition unit connected in sequence, and the photovoltaic power prediction model is an Autoformer model; The sample decoding module is used to input the initial sample periodic term, the second sample vector, and the sample encoding result into the first branch of the decoder in the photovoltaic power prediction model to obtain the first sample sub-result output by the decoder; the first branch of the decoder includes at least one pair of sequentially connected autocorrelation mechanism units and sequence decomposition units; the initial sample trend term is input into the second branch of the decoder; and the second sample sub-result is obtained. The sample prediction module is used to determine the sample photovoltaic power result based on the first sample sub-result and the second sample sub-result; The model update module is used to update the photovoltaic power prediction model based on the sample photovoltaic power results and the sample labels to obtain the updated photovoltaic power prediction model. The sample decoding module is specifically used to: input the initialization periodic term and the second time vector into the first branch of the decoder, so as to process them through the first autocorrelation mechanism unit and the first sequence decomposition unit to obtain the intermediate vector; The intermediate vector and the encoding result are processed sequentially through the second autocorrelation mechanism unit, the second sequence decomposition unit, the first feedforward neural network, and the third sequence decomposition unit to obtain the first sub-result; the initialization trend term is input into the second branch of the decoder and sequentially fused with the first sub-trend term obtained by the first sequence decomposition unit, the second sub-trend term obtained by the second sequence decomposition unit, and the third sub-trend term obtained by the third sequence decomposition unit to obtain the second sub-result.

8. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one instruction, the at least one instruction being loaded and executed by the processor to implement the photovoltaic power prediction method as described in any one of claims 1 to 4; or, the at least one instruction being loaded and executed by the processor to implement the photovoltaic power prediction model training method as described in claim 5.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, which is loaded and executed by a processor to implement the photovoltaic power prediction method as described in any one of claims 1 to 4; or, the at least one instruction is loaded and executed by a processor to implement the photovoltaic power prediction model training method as described in claim 5.

Images