Method and device for predicting steep drop of wind-solar power generation for wind-solar coupled hydrogen production system

Abstract

This specification relates to the field of wind and solar energy new energy technology, and specifically discloses a method and device for predicting a steep drop in wind and solar power generation for a wind and solar coupled hydrogen production system, the method comprising: generating a training sample set based on actual power generation, meteorological pattern data and ground-based cloud map data in a historical time period; performing repeated sampling with replacement on the training sample set to obtain multiple sub-training sets; using a time series algorithm based on trend extraction to learn multiple sub-training sets to obtain multiple time series algorithm models; inputting meteorological pattern data and ground-based cloud map data in a future target time period into each time series algorithm model to obtain a target power generation corresponding to each time series algorithm model; determining the power steep drop data of wind and solar power generation in the future target time period according to the target power generation of wind and solar power generation in the future target time period corresponding to each time series algorithm model. The above method can accurately predict the steep drop in power generation, thereby improving the operational safety of wind and solar coupled hydrogen production.

Description

Technical Field

[0001] The present invention relates to the field of wind and solar energy technologies, and in particular to a method and device for predicting a steep drop in wind and solar power generation power for a wind and solar coupled hydrogen production system. Background Art

[0002] Wind-solar coupling power generation to produce green hydrogen is an important measure to improve the efficiency of wind and solar resource utilization. However, the fluctuating characteristics of wind and solar power generation do not meet the hydrogen production system's demand for stable load. High-precision power prediction data is the key technology to resolve this contradiction.

[0003] At present, a lot of research has been done on the power prediction methods of new energy. From the time dimension, power prediction technology can be divided into ultra-short-term power prediction, short- and medium-term power prediction, and annual power generation prediction. From the algorithm dimension, it can be divided into physical prediction method, statistical prediction method, and artificial intelligence prediction method. Ultra-short-term power prediction mainly predicts the power of the next 0-4h, with a time resolution of 15min and a total of 16 points. Short- and medium-term power prediction mainly predicts the power of the next 240 hours, with a time resolution of 15min. Annual power generation prediction mainly predicts the power generation of the next year. The physical prediction method is to build a physical model based on the specific geographical location, topography and other geographical information of the site, and generate a mapping relationship between meteorological characteristics and power by using physical principles such as fluid mechanics. The statistical prediction method mainly focuses on the historical data of the site. Through relevant statistical methods, the historical data is analyzed to obtain the potential laws contained in the data, and the future power is summarized and summarized by historical data. The artificial intelligence prediction method is to learn the characteristics of historical data and learn potential laws to predict future power by using mainstream machine learning algorithms and deep learning algorithms.

[0004] At present, the research hotspot in the field of green electricity hydrogen production is to use the power prediction data of new energy as the input of the wind-solar coupled hydrogen production system control end to assist in improving the operating efficiency and safety of the wind-solar coupled hydrogen production system. However, although the existing wind-solar power prediction technology for wind-solar coupled hydrogen production systems has used the power prediction data of new energy as the input of the wind-solar coupled hydrogen production system control end, the relevant research has only simply combined the power prediction and hydrogen production systems. The conventional new energy power prediction system mainly serves the production management of new energy power stations and the needs of grid connection, and cannot meet the personalized needs of wind-solar coupled hydrogen production systems for power prediction technology, which greatly limits the effect of power prediction technology on improving the operating efficiency and safety of wind-solar coupled hydrogen production systems.

[0005] Currently, no effective solution has been proposed to the above problems. Summary of the invention

[0006] The embodiments of this specification provide a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system, so as to solve the problem that the wind-solar power generation prediction technology in the prior art has limited effect on improving the operating efficiency and safety of the wind-solar coupled hydrogen production system.

[0007] The embodiment of this specification provides a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system, including:

[0008] Acquire actual power generation, meteorological model data and ground-based cloud map data of wind-solar power generation for a wind-solar coupled hydrogen production system in a historical time period; generate a training sample set based on the actual power generation, meteorological model data and ground-based cloud map data in the historical time period;

[0009] Repeatedly sampling the data samples in the training sample set with replacement to obtain multiple sub-training sets; using a time series algorithm based on trend extraction to learn the data samples in the multiple sub-training sets to obtain multiple time series algorithm models;

[0010] Obtaining meteorological model data and ground-based cloud map data for wind and solar power generation in a future target time period; inputting the meteorological model data and ground-based cloud map data in the future target time period into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation in the future target time period corresponding to each of the time series algorithm models;

[0011] According to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model, the power drop data of wind and solar power generation in the future target time period is determined.

[0012] In one embodiment, a time series algorithm based on trend extraction is used to learn data samples in the multiple sub-training sets to obtain multiple time series algorithm models, including:

[0013] Performing multi-level downsampling on the sample data of each sub-training set in the multiple sub-training sets to obtain time series data of multiple different time granularities corresponding to each sub-training set;

[0014] Decomposing the time series data of each time granularity among the multiple time series data of different time granularities corresponding to each sub-training set to obtain multiple sub-sequences corresponding to each sub-training set;

[0015] A linear neural network is used to scale and sum a plurality of subsequences corresponding to each of the sub-training sets to obtain a plurality of sequence data of different time granularities corresponding to each of the sub-training sets;

[0016] A time series algorithm based on trend extraction is used to learn the sequence data of multiple different time granularities corresponding to each of the sub-training sets to obtain multiple time series algorithm models.

[0017] In one embodiment, decomposing the time series data of each time granularity among the time series data of multiple different time granularities corresponding to each sub-training set to obtain multiple sub-sequences corresponding to each sub-training set includes:

[0018] Using an empirical mode decomposition algorithm or a variational mode decomposition algorithm, the time series data of each time granularity in the time series data of multiple different time granularities corresponding to each sub-training set is decomposed to extract trend data features and seasonal data features, and obtain multiple subsequences corresponding to each sub-training set; the multiple subsequences include seasonal sequences and trend sequences.

[0019] In one embodiment, a linear neural network is used to scale and sum the multiple subsequences corresponding to each of the sub-training sets, including:

[0020] A linear neural network is used to scale the trend data features corresponding to each sub-training set from long to short, and the seasonal data features corresponding to each sub-training set from short to long, to obtain a plurality of scaled sub-sequences corresponding to each sub-training set;

[0021] The scaled multiple subsequences corresponding to the sub-training sets are summed up according to the multiple different time granularities to obtain sequence data of the multiple different time granularities corresponding to the sub-training sets.

[0022] In one embodiment, the time series algorithm based on trend extraction includes at least one of the following:

[0023] Long short-term memory network, TimeMixer model and autoregressive moving average model.

[0024] In one embodiment, according to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model, the power drop data of wind and solar power generation in the future target time period is determined, including:

[0025] According to the target power generation at each target moment in the multiple target moments within the future target time period corresponding to each time series algorithm model, multiple target power generation corresponding to each target moment is obtained;

[0026] Based on the multiple target power generation powers corresponding to the target moments, calculating the confidence interval corresponding to each confidence degree among the multiple confidence degrees at the target moments;

[0027] According to the confidence interval corresponding to each confidence degree among the multiple confidence degrees at the target moments, the steep drop probability and steep drop amplitude of wind and solar power generation at the target moments are determined.

[0028] In one embodiment, according to the confidence interval corresponding to each confidence degree among the multiple confidence degrees at each target time, determining the steep drop probability and steep drop amplitude of wind and solar power generation at each target time includes:

[0029] For each target moment, the difference between the generated power at the current moment and the upper limit closest to the generated power at the current moment in the confidence interval corresponding to each confidence level is taken as the steep drop amplitude, and the confidence level of the confidence interval corresponding to the steep drop amplitude is taken as the steep drop probability corresponding to the steep drop amplitude.

[0030] The embodiment of this specification also provides a wind-solar power generation power drop prediction device for a wind-solar coupled hydrogen production system, comprising:

[0031] An acquisition module, used to acquire actual power generation, meteorological model data and ground-based cloud map data of wind-solar power generation for a wind-solar coupled hydrogen production system in a historical time period; and also used to generate a training sample set based on the actual power generation, meteorological model data and ground-based cloud map data in the historical time period;

[0032] A learning module, used for repeatedly sampling the data samples in the training sample set with replacement to obtain multiple sub-training sets; and also used for learning the data samples in the multiple sub-training sets using a time series algorithm based on trend extraction to obtain multiple time series algorithm models;

[0033] A power prediction module, used to obtain meteorological model data and ground-based cloud map data for wind and solar power generation in a future target time period; and also used to input the meteorological model data and ground-based cloud map data in the future target time period into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation in the future target time period corresponding to each of the time series algorithm models;

[0034] The steep drop determination module is used to determine the power steep drop data of wind and solar power generation in the future target time period according to the target power generation power of wind and solar power generation in the future target time period corresponding to each timing algorithm model.

[0035] An embodiment of the present specification also provides a computer device, including a processor and a memory for storing processor executable instructions, wherein when the processor executes the instructions, the steps of the method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system described in any of the above embodiments are implemented.

[0036] An embodiment of the present specification also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, implement the steps of the method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system described in any of the above embodiments.

[0037] The embodiment of this specification provides a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system, which can obtain the actual power generation, meteorological pattern data and ground-based cloud map data of wind-solar power generation for a wind-solar coupled hydrogen production system in a historical time period, generate a training sample set based on the actual power generation, meteorological pattern data and ground-based cloud map data in the historical time period, repeatedly sample the data samples in the training sample set with replacement to obtain multiple sub-training sets, and use a time series algorithm based on trend extraction to learn the data samples in the multiple sub-training sets to obtain multiple time series algorithm models. Afterwards, the meteorological pattern data and ground-based cloud map data of wind-solar power generation in a future target time period can be obtained, and the meteorological pattern data and ground-based cloud map data in the future target time period can be input into each of the multiple time series algorithm models to obtain the target power generation of wind-solar power generation in the future target time period corresponding to each time series algorithm model. The power steep drop data of wind-solar power generation in the future target time period can be determined based on the target power generation of wind-solar power generation in the future target time period corresponding to each time series algorithm model. The above scheme will focus on the prediction of power drop values. It uses artificial intelligence algorithms and combines high-resolution meteorological data provided by all-sky imagers, wind measuring equipment and photometric equipment to achieve high-precision power drop prediction. It solves the problem that conventional new energy power prediction cannot meet the wind-solar coupled hydrogen production system's needs for high accuracy and high resolution prediction of wind-solar power drop values, and can greatly improve the operational safety of the wind-solar coupled hydrogen production system.

[0038] With reference to the following description and drawings, specific embodiments of the present invention are disclosed in detail, indicating the manner in which the principles of the present invention can be adopted. It should be understood that the embodiments of the present invention are not limited in scope. Features described and / or shown for one embodiment can be used in one or more other embodiments in the same or similar manner, combined with features in other embodiments, or replace features in other embodiments.

[0039] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, parts, steps or components, but does not exclude the presence or addition of one or more other features, parts, steps or components. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings described herein are only for explanation purposes and are not intended to limit the scope of the present invention in any way. In addition, the shapes and proportional dimensions of the components in the drawings are only for illustration purposes and are used to help understand the present invention. They do not specifically limit the shapes and proportional dimensions of the components of the present invention. Under the guidance of the present invention, those skilled in the art can select various possible shapes and proportional dimensions to implement the present invention according to specific circumstances. In the drawings:

[0041] Figure 1 A flow chart of a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system in accordance with an embodiment of this specification is shown;

[0042] Figure 2 A flow chart of a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system in accordance with an embodiment of this specification is shown;

[0043] Figure 3 A flowchart of processing a sub-training set into sequence data in one embodiment of this specification is shown;

[0044] Figure 4 A schematic diagram of the structure of a wind-solar power generation power drop prediction device for a wind-solar coupled hydrogen production system in an embodiment of this specification is shown;

[0045] Figure 5 A schematic diagram of the structure of a computer device in an embodiment of the present specification is shown. DETAILED DESCRIPTION

[0046] The principles and spirit of this specification will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement this specification, and are not intended to limit the scope of this specification in any way. On the contrary, these embodiments are provided to make this specification more thorough and complete, and to fully convey the scope of this disclosure to those skilled in the art.

[0047] Those skilled in the art know that the embodiments of this specification can be implemented as a system, device, method or computer program product. Therefore, this specification can be implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0048] The details of the present invention can be more clearly understood by combining the accompanying drawings and the description of the specific embodiments of the present invention. However, the specific embodiments of the present invention described herein are only used for the purpose of explaining the present invention and cannot be understood as limiting the present invention in any way. Under the guidance of the present invention, technicians can conceive of any possible variations based on the present invention, which should be regarded as belonging to the scope of the present invention. It should be noted that when an element is referred to as "arranged on" another element, it can be directly on another element or there can also be a central element. When an element is considered to be "connected" to another element, it can be directly connected to another element or there may be a central element at the same time. The terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a mechanical connection or an electrical connection, or it can be the communication between the two elements, it can be directly connected, or it can be indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances. The terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are only for illustrative purposes and do not represent the only implementation method.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this specification belongs. The terms used herein in this specification are only for the purpose of describing specific embodiments and are not intended to limit this specification. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0050] For the wind-solar coupled hydrogen production system, due to changes in weather conditions, such as a sudden drop in wind speed or clouds blocking sunlight, the wind-solar power generation may drop significantly in a short period of time, resulting in insufficient wind-solar power generation for hydrogen production, which may cause the electrolyzer to shut down instantly, and have a huge impact on the operating life of the electrolyzer equipment. At the same time, at a lower current density, the electrolyzer may produce more hydrogen-oxygen mixtures, which poses an explosion risk and has great safety hazards, endangering the safety of the entire hydrogen production system. Therefore, the inventors have found through research that predicting whether there will be a sharp drop in wind-solar power generation in the future is a rigid requirement for the long-term safe operation of the wind-solar coupled hydrogen production system. However, the current conventional power prediction technology is mostly trend prediction and prediction of time-series power data for a period of time in the future, which cannot meet the wind-solar coupled hydrogen production system's prediction requirements for high accuracy and high resolution of wind-solar power drop values.

[0051] Based on the above problems, the embodiments of this specification provide a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system. Figure 1A flow chart of a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system in an embodiment of this specification is shown. Although this specification provides method operation steps or device structures as shown in the following embodiments or drawings, more or fewer operation steps or module units may be included in the method or device based on routine or no creative labor. In steps or structures that do not logically have a necessary causal relationship, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure described in the embodiments of this specification and shown in the drawings. When the method or module structure described is applied to an actual device or terminal product, it can be connected according to the method or module structure shown in the embodiments or drawings for sequential execution or parallel execution (for example, a parallel processor or multi-threaded processing environment, or even a distributed processing environment).

[0052] Specifically, Figure 1 As shown, a method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system provided in an embodiment of the present specification may include the following steps.

[0053] Step S101, obtaining actual power generation, meteorological model data and ground-based cloud map data of wind-solar coupled hydrogen production system in a historical time period; generating a training sample set based on the actual power generation, meteorological model data and ground-based cloud map data in the historical time period.

[0054] The method in the embodiments of this specification can be applied to a computer device. The computer device can be a client or a server, etc. The actual power generation power, meteorological pattern data and ground-based cloud map data of wind-solar power generation for a wind-solar coupled hydrogen production system in a historical time period can be obtained. Among them, the historical time period refers to a preset time period before the current moment. The meteorological pattern data includes at least one of the following: wind speed, irradiance, temperature, humidity, wind direction and pressure data.

[0055] A training sample set can be generated based on the actual power generation, meteorological model data and ground-based cloud map data in a historical time period. The training sample set may include multiple sample data, each sample data may be a sample and a label, the sample may include the meteorological model data and ground-based cloud map data at a certain moment, and the label may be the actual power generation at that moment. In one embodiment, the actual power generation, meteorological model data and ground-based cloud map data in a historical time period may be preprocessed. Preprocessing may include cleaning the data, removing invalid data and correcting erroneous information. Then, fill in missing values. Subsequently, these data are integrated according to a unified standard and format to form a data set for analysis.

[0056] In one embodiment, the data set can be divided into a training sample set, a validation sample set and a test sample set. The training sample set can be used for model training, the validation sample set is used to verify and optimize the model, and the test sample set is used to evaluate the model.

[0057] Step S102, repeatedly sampling the data samples in the training sample set with replacement to obtain multiple sub-training sets; using a time series algorithm based on trend extraction to learn the data samples in the multiple sub-training sets to obtain multiple time series algorithm models.

[0058] After generating a training sample set, the data samples in the training sample set can be repeatedly sampled with replacement to obtain multiple sub-training sets. The number of sample data in multiple sub-training sets may be the same or different. In one embodiment, the Bootstrap method may be used to perform repeated sampling with replacement. After obtaining multiple training subsets, a time series algorithm based on trend extraction may be used to learn the data samples in the multiple sub-training sets to obtain multiple time series algorithm models. In some embodiments of the present specification, the time series algorithm based on trend extraction includes at least one of the following: a long short-term memory network, a TimeMixer model, and an autoregressive moving average model. By using a time series algorithm based on trend extraction to learn the data samples in each of the multiple sub-training sets, a time series algorithm model corresponding to each training set is obtained. The time series algorithm model can predict the changing trend of the power generation power of the wind-solar coupled hydrogen production system.

[0059] Step S103, obtaining meteorological pattern data and ground-based cloud map data for wind and solar power generation in a future target time period; inputting the meteorological pattern data and ground-based cloud map data in the future target time period into each of the multiple time series algorithm models, and obtaining the target power generation power of wind and solar power generation in the future target time period corresponding to each of the time series algorithm models.

[0060] Meteorological pattern data and ground-based cloud map data for wind and solar power generation in a future target time period may be obtained. The meteorological pattern data and ground-based cloud map data in the future target time period may be meteorological pattern prediction data and ground-based cloud map prediction data in the future target time period. In one embodiment, the meteorological pattern prediction data may be predicted based on meteorological pattern data in a historical time period, for example, by machine learning. The ground-based cloud map prediction data may be predicted based on ground-based cloud map prediction data in a historical time period, for example, by machine learning.

[0061] The meteorological model data and ground-based cloud map data within the future target time period can be input into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation within the future target time period corresponding to each of the multiple time series algorithm models. The future target time period may include multiple target moments. The meteorological model data and ground-based cloud map data corresponding to each of the multiple target moments can be input into each of the time series algorithm models to obtain multiple power generation corresponding to each of the target moments.

[0062] Step S104, determining the power drop data of wind and solar power generation in the future target time period according to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model.

[0063] After obtaining the target power generation of wind and solar power generation in the future target time period corresponding to each of the timing algorithm models, the power drop data of wind and solar power generation in the future target time period can be determined according to the target power generation of wind and solar power generation in the future target time period corresponding to each of the timing algorithm models. Specifically, the predicted power generation distribution at each target moment can be determined based on the multiple power generation corresponding to each target moment, thereby determining the power drop data of wind and solar power generation in the future target time period. The power drop data may include power drop conditions at multiple target moments in the future target time period. For example, whether a power drop occurs, the probability of a power drop, and / or the amplitude of a power drop.

[0064] In the above embodiment, power drops can be predicted. By using artificial intelligence algorithms and combining high-resolution meteorological data provided by all-sky imagers, wind measuring equipment, and photometric equipment, high-precision prediction of wind and solar power generation power drops can be achieved. This solves the problem that conventional new energy power predictions cannot meet the wind-solar coupled hydrogen production system's requirements for high accuracy and high resolution prediction of wind and solar power generation power drops, and can greatly improve the operational safety of the wind-solar coupled hydrogen production system.

[0065] In some embodiments of the present specification, using a trend extraction-based time series algorithm to learn data samples in the multiple sub-training sets to obtain multiple time series algorithm models can include: performing multi-level downsampling on the sample data of each sub-training set in the multiple sub-training sets to obtain multiple time series data of different time granularities corresponding to each sub-training set; decomposing the time series data of each time granularity in the multiple time series data of different time granularities corresponding to each sub-training set to obtain multiple sub-sequences corresponding to each sub-training set; using a linear neural network to scale and sum the multiple sub-sequences corresponding to each sub-training set to obtain multiple sequence data of different time granularities corresponding to each sub-training set; using a trend extraction-based time series algorithm to learn the multiple sequence data of different time granularities corresponding to each sub-training set to obtain multiple time series algorithm models.

[0066] Specifically, the sample data of each sub-training set in the multiple sub-training sets can be multi-level down-sampled to obtain time series data of multiple different time granularities corresponding to each sub-training set. Multi-level down-sampling can make the sample data show the regularity under different time periods. Multiple different time granularities can be preset according to actual needs. For example, multiple different time granularities can include: 5 minutes, 15 minutes, 30 minutes and 60 minutes. The time series data of each time granularity in the time series data of multiple different time granularities corresponding to each sub-training set can be decomposed to obtain multiple subsequences corresponding to each sub-training set, and a certain time series can be decomposed into multiple time series data with more regularity or easier to fit, and the sum of these decomposed time series is equal to the original sequence. A linear neural network can be used to scale and sum the multiple subsequences corresponding to each sub-training set to obtain the sequence data of multiple different time granularities corresponding to each sub-training set. Afterwards, the sequence data of multiple different time granularities corresponding to each sub-training set can be learned using a time series algorithm based on trend extraction to obtain multiple time series algorithm models. Through the above method, the data can show the rules in different time periods, thereby improving the accuracy of the time series algorithm model.

[0067] In some embodiments of the present specification, decomposing the time series data of each time granularity among the multiple time series data of different time granularities corresponding to the each sub-training set to obtain the multiple subsequences corresponding to the each sub-training set may include: using an empirical mode decomposition algorithm or a variational mode decomposition algorithm to decompose the time series data of each time granularity among the multiple time series data of different time granularities corresponding to the each sub-training set to extract trend data features and seasonal data features to obtain the multiple subsequences corresponding to the each sub-training set; the multiple subsequences may include seasonal sequences and trend sequences.

[0068] Specifically, historical power generation data is closely related to changes in global irradiance and global wind speed, and natural phenomena in nature are often caused by certain changes, and this type of change is often directly or indirectly related to the laws of social operation and the laws of the earth's rotation and revolution. Therefore, these changes often have regularities that can be disassembled and analyzed: trend and seasonality. The empirical mode decomposition algorithm or the variational mode decomposition algorithm can be used to decompose the time series data of each time granularity in the time series data of multiple different time granularities corresponding to each sub-training set to extract trend data features and seasonal data features, and obtain multiple subsequences corresponding to each sub-training set. In the above manner, trend data features and seasonal data features can be extracted.

[0069] In some embodiments of the present specification, scaling and summing the multiple subsequences corresponding to the sub-training sets using a linear neural network may include: scaling the trend data features corresponding to the sub-training sets from long to short, and scaling the seasonal data features corresponding to the sub-training sets from short to long using a linear neural network, to obtain the scaled multiple subsequences corresponding to the sub-training sets; and summing the scaled multiple subsequences corresponding to the sub-training sets according to the multiple different time granularities, to obtain the sequence data of the multiple different time granularities corresponding to the sub-training sets. In the above manner, the trend data and seasonal data may be scaled and summed to obtain the sequence data of multiple different time granularities.

[0070] In some embodiments of the present specification, determining the power drop data of wind and solar power generation in the future target time period according to the target power generation of wind and solar power generation in the future target time period corresponding to each timing algorithm model may include: obtaining multiple target power generation powers corresponding to each target moment according to the target power generation power at each target moment in the future target time period corresponding to each timing algorithm model; calculating the confidence interval corresponding to each of the multiple confidence levels at each target moment based on the multiple target power generation powers corresponding to each target moment; determining the probability and amplitude of the drop of wind and solar power generation at each target moment according to the confidence interval corresponding to each of the multiple confidence levels at each target moment.

[0071] Specifically, after obtaining the multiple target power generation corresponding to each target moment, the confidence interval corresponding to each of the multiple confidence levels at each target moment can be calculated based on the multiple target power generation corresponding to each target moment. In one embodiment, multiple confidence levels can be preset. For example, the multiple confidence levels can include multiple confidence levels of the following: 95%, 90%, 85%, 80%, 75%, 70%, 60%, 50%, 40%, 30%, 20%, 10%, etc. The confidence interval corresponding to each of the multiple confidence levels at each target moment can be calculated based on the multiple target power generation corresponding to each target moment. The confidence interval can include the upper and lower limits of the power generation. According to the confidence interval corresponding to each of the multiple confidence levels at each target moment, the steep drop probability and steep drop amplitude of wind and solar power generation at each target moment can be determined. In the above manner, confidence levels or confidence intervals of different power steep drop degrees can be obtained.

[0072] In some embodiments of the present specification, determining the probability and amplitude of a sudden drop in wind and solar power generation at each target moment based on the confidence interval corresponding to each of the multiple confidence levels at each target moment may include: for each target moment, taking the difference between the power generation at the current moment and the upper limit closest to the power generation at the current moment in the confidence interval corresponding to each confidence level as the amplitude of the sudden drop, and taking the confidence level of the confidence interval corresponding to the sudden drop amplitude as the probability of a sudden drop corresponding to the sudden drop amplitude.

[0073] Specifically, by using confidence to calculate the confidence interval, a set of x×2×m prediction results are finally generated, where x is the number of different confidence levels, 2 is the upper and lower limits of the confidence interval, and m is the length of the prediction data. According to the different degrees of steep drop combined with the upper and lower limits of the interval, the corresponding confidence of the interval closest to the actual / predicted power at the previous moment that meets the conditions is selected as the probability of a steep drop in wind and solar power generation at the current moment. At present, the definition of the degree of steep drop is vague, so the probability of different degrees of steep drop is output. For example, the degrees of steep drop are: Capacity×0.1, Capacity×0.2, Capacity×0.3. Capacity is the installed capacity. Assuming that the actual power generation at 10:15 is 100MW, the installed capacity is 200MW, the confidence interval of the predicted output at 10:30 with a confidence level of 90% is [10, 80], the confidence interval of 80% is [20, 60], and the confidence interval of 70% is [30, 40]. Then it can be determined that there is a 90% probability of a 10% drop at 10:30 in the future, an 80% probability of a 20% drop, and a 70% probability of a 30% drop. In this way, high-precision confidence or confidence interval predictions for different degrees of power drop in wind and solar power output can be achieved.

[0074] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For details, please refer to the description of the above-mentioned related processing related embodiments, and no further description is given here.

[0075] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0076] The above method is described below in conjunction with a specific embodiment. However, it should be noted that the specific embodiment is only for better illustrating the present specification and does not constitute an improper limitation on the present specification.

[0077] In this specific embodiment, a method for predicting a sudden drop in wind and solar power generation power for a wind-solar coupled hydrogen production system is provided. The wind-solar coupled hydrogen production system needs to predict in advance whether the wind and solar power generation will experience a sudden drop in power in the future, so that the electrolyzer control system can intervene in time and make corresponding control strategies to prevent danger. This is a rigid requirement for the long-term safe operation of the wind-solar coupled hydrogen production system. This embodiment constructs an innovative algorithm architecture by applying time series algorithm design and Boostrap ideas to predict the probability of a sudden drop with high quality and accuracy. This technology has two important components: the Boostrap main architecture and a time series neural network algorithm based on trend extraction. Please refer to Figure 2 , shows a flow chart of the method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system in this specific embodiment.

[0078] like Figure 2 As shown, the data required in this embodiment are historical actual power generation, historical meteorological model data and ground-based cloud maps. The existing data samples are divided into training sets, validation sets and test sets. Many sub-training sets with different distributions are constructed by repeatedly sampling the original divided training set data samples with replacement (Boostrap sampling). Then, a time series algorithm based on trend extraction (for example, a time series algorithm model based on the trend of time series data such as Arima, LSTM, TimeMixer, etc. (here it is not limited that all sub-samples need to use the same algorithm, and can be mixed in actual situations)) is used as a "weak learner" to combine the design optimization algorithm to continuously optimize and learn n samples of different distributions, and finally generate and save the best time series model for each different sample. The trained model is used to predict the test set, and finally a prediction value matrix of n×m is generated, where n is the number of sampling times and m is the length of the test set. By using confidence to calculate the confidence interval, a set of x×2×m prediction results is finally generated, where x is the number of different confidence levels, 2 is the upper and lower limits of the confidence interval, and m is the length of the prediction data. Finally, a judgment is made based on the different degrees of steep drop combined with the upper and lower limits of the interval, and the corresponding confidence level of the interval closest to the actual / predicted power at the previous moment that meets the conditions is selected as the probability of the steep drop at the current moment.

[0079] like Figure 2As shown in the figure, the left side is the architecture implementation of Boostrap, and the right side is the algorithm architecture implementation of the core time series algorithm. The purpose of Boostrap is to output probability prediction, and the time series algorithm is to complete the algorithm support of power prediction. Bootstrap is a statistical method based on the idea of ​​bootstrapping. It can generate the distribution of estimates by randomly extracting multiple samples from the samples of existing data with replacement, so as to perform statistical analysis such as interval estimation. Historical power generation data is closely related to global irradiance changes and global wind speed changes. Natural phenomena in nature are often caused by some changes, and this type of change is often directly or indirectly related to the laws of social operation and the laws of the earth's rotation and revolution. Therefore, these changes often have regularities that can be disassembled and analyzed: trend and seasonality. First, collect and organize the existing historical model meteorological and historical actual data, and then generate multiple downsampled samples with gradually decreasing data lengths through multi-level downsampling. Use the empirical mode decomposition algorithm (EMD) or the variational mode decomposition algorithm (VMD) to decompose these samples and extract their trend and seasonal potential data characteristics. By using two layers of linear layers with GELU as the activation function, the downsampled data of different lengths are scaled. Finally, these sequences are summed up to generate new sequences, which are then input into the feedforward neural network to output each new processed predicted sequence data. Finally, two layers of linear neural networks are connected as prediction layers to finally output predictions for each subsequence, and the predicted data of all subsequences are summed up to generate the final power prediction technology.

[0080] Please refer to Figure 3 , which shows a flowchart for processing the sub-training set into sequence data. Figure 3 As shown in the figure, we first collect and organize the existing historical model meteorological and historical actual data, and obtain multiple sub-training sets by resampling with replacement. For each sub-training set, M time series data with different time granularities can be generated by multi-level downsampling. Multi-level downsampling can make the data show the laws under different time periods. Mixing these different time scales can further enhance the model's ability to extract laws. Perform EMD\VMD decomposition on the time series of each granularity to generate M subsequences, and the sum of these subsequences is equal to the sequence before decomposition. The multiple subsequences generated by decomposing the sequences of different time granularities are scaled using a neural network to generate M sequences of different time granularities. Scaling is performed because the time granularity of the data after downsampling is inconsistent. Scaling it through a linear network is mainly to make the data dimension consistent, and it can also further integrate the time series laws of different sequences.

[0081] All sequences scaled by sequences of other time granularities are summed up accordingly, and sequence data extracted, scaled, and synthesized by data of different time granularities are summarized. It is input into the feedforward neural network to output each new processed predicted sequence data. Finally, two layers of linear neural network are connected as the prediction layer, and the final output is a measurement of each subsequence. The predicted data of all subsequences are summed to generate the final power prediction method. The predicted data of all subsequences are summed because the sum of all subsequences of EMD\VMD is equal to the original sequence, so the prediction results also need to be summed to restore the original prediction.

[0082] Figure 3 The predictor in Figure 2 Feedforward neural network in the algorithm flow on the right. The scaling rule here is: seasonal series scale from short to long, and trend series scale from long to short. This is because trend series easily introduces noise when refining the time granularity, so it is scaled from long to short.

[0083] The method in this specific embodiment may include the following steps.

[0084] Step 1, data processing.

[0085] Step 1.1, data set reading.

[0086] Step 1.2, data set segmentation: divide the existing data samples into training set, validation set, and test set.

[0087] Step 2: Model training.

[0088] Step 2.1, sample sampling: Repeated sampling with replacement of the original divided training set data samples to construct many sub-training sets with different distributions.

[0089] Step 2.2, model training: The time series algorithm based on trend extraction is used as a "weak learner" in combination with the design optimization algorithm to continuously optimize and learn n samples with different distributions, and finally generate and save the best time series model for each different sample.

[0090] Step 3, model architecture.

[0091] Step 3.1, model architecture: Based on the existing historical model meteorological and historical actual data, multiple downsampled samples with gradually decreasing data lengths are generated through multi-level downsampling. These samples are decomposed using EMD / VMD to extract their trend and seasonal potential data features. The downsampled data of different lengths are scaled by using a two-layer linear layer with a GELU activation function. Finally, these sequences are summed to generate a new sequence, which is then input into a feedforward neural network to output each new processed prediction sequence data. Finally, a two-layer linear neural network is connected as the prediction layer to finally output the prediction for each subsequence, sum the prediction data of all subsequences, and generate the final power prediction technology.

[0092] Step 4: Probability output processing

[0093] Step 4.1, test set prediction: The trained model is used to predict the test set, and finally an n×m prediction value matrix is ​​generated, where n is the number of sampling times and m is the length of the test set. By using the confidence level to calculate the confidence interval, a set of x×2×m prediction results is finally generated, where x is the number of different confidence levels, 2 is the upper and lower limits of the confidence interval, and m is the length of the prediction data.

[0094] Step 4.2, probability output: According to different degrees of steep drop combined with the upper and lower limits of the interval, the corresponding confidence level of the interval closest to the actual / predicted power at the previous moment is selected as the probability of steep drop at the current moment. The steep drop levels are: Capacity×0.1, Capacity×0.2, Capacity×0.3. Capacity is the installed capacity. Assuming that the actual power at 10:15 is 100MW, the installed capacity is 200MW, the confidence interval of the predicted output at 10:30 with a 90% confidence level is [10, 80], the confidence interval of 80% is [20, 60], and the confidence interval of 70% is [30, 40]. Then it can be determined that there is a 90% probability of a 10% steep drop, an 80% probability of a 20% steep drop, and a 70% probability of a 30% steep drop at 10:30 in the future.

[0095] The method in this specific embodiment solves the problem that conventional new energy power prediction cannot meet the prediction requirements of the wind-solar coupled hydrogen production system for high accuracy and high resolution of power drop values, which will greatly improve the operating safety of the wind-solar coupled hydrogen production system.

[0096] The hardware devices on which the method in this specific embodiment is based include: all-sky imager, wind measuring equipment, light measuring equipment, computer equipment, power prediction system, etc.

[0097] The model can be built by combining the historical actual data, ground-based cloud map data, historical wind speed, and historical irradiance data provided on site with meteorological model forecasts, and then deployed on site. The model will be updated and re-issued within a fixed time period to achieve on-site model updates. Based on the actual data and actual wind speed and irradiance data collected on site at multiple time scales, the on-site model will support the correction of the irradiance and wind speed predicted by the model based on these data. Finally, the probability of power drop at a future moment will be predicted based on the corrected data.

[0098] The method in this embodiment solves the problem that conventional new energy power prediction cannot meet the prediction requirements of wind-solar coupled hydrogen production system for high accuracy and high resolution of power drop value, which will greatly improve the operational safety of wind-solar coupled hydrogen production system. Abandoning the prediction of power trend changes in traditional prediction models, the focus is placed on the prediction of power drop values. By using artificial intelligence algorithms, combined with high-resolution meteorological data provided by all-sky imagers, wind measuring equipment, and photometric equipment, high-precision confidence or confidence interval predictions for different power drop degrees of wind and solar output are achieved. A series of meteorological observation equipment is used, including but not limited to all-sky imagers, laser wind radars, wind towers, and environmental monitors.

[0099] Based on the same inventive concept, the embodiments of this specification also provide a wind-solar power drop prediction device for a wind-solar coupled hydrogen production system, as described in the following embodiments. Since the principle of solving the problem by the wind-solar power drop prediction device for a wind-solar coupled hydrogen production system is similar to the wind-solar power drop prediction method for a wind-solar coupled hydrogen production system, the implementation of the wind-solar power drop prediction device for a wind-solar coupled hydrogen production system can refer to the implementation of the wind-solar power drop prediction method for a wind-solar coupled hydrogen production system, and the repeated parts will not be repeated. As used below, the terms "unit" or "module" can be a combination of software and / or hardware that implements predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and conceived. Figure 4 is a structural block diagram of a wind-solar power generation power drop prediction device for a wind-solar coupled hydrogen production system according to an embodiment of this specification, such as Figure 4 As shown, it includes: an acquisition module 401, a learning module 402, a power prediction module 403 and a steep drop determination module 404. The structure is described below.

[0100] The acquisition module 401 is used to acquire the actual power generation, meteorological model data and ground-based cloud map data of wind-solar power generation for the wind-solar coupled hydrogen production system in a historical time period; and is also used to generate a training sample set based on the actual power generation, meteorological model data and ground-based cloud map data in the historical time period;

[0101] The learning module 402 is used to repeatedly sample the data samples in the training sample set with replacement to obtain multiple sub-training sets; and is also used to learn the data samples in the multiple sub-training sets using a time series algorithm based on trend extraction to obtain multiple time series algorithm models;

[0102] The power prediction module 403 is used to obtain the meteorological model data and ground-based cloud map data of wind and solar power generation in the future target time period; and is also used to input the meteorological model data and ground-based cloud map data in the future target time period into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation in the future target time period corresponding to each time series algorithm model;

[0103] The steep drop determination module 404 is used to determine the power steep drop data of wind and solar power generation in the future target time period according to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model.

[0104] In some embodiments of the present specification, the learning module is specifically used to: perform multi-level downsampling on the sample data of each sub-training set among the multiple sub-training sets to obtain multiple time series data of different time granularities corresponding to each sub-training set; decompose the time series data of each time granularity among the multiple time series data of different time granularities corresponding to each sub-training set to obtain multiple sub-sequences corresponding to each sub-training set; use a linear neural network to scale and sum the multiple sub-sequences corresponding to each sub-training set to obtain multiple sequence data of different time granularities corresponding to each sub-training set; use a time series algorithm based on trend extraction to learn the multiple sequence data of different time granularities corresponding to each sub-training set to obtain multiple time series algorithm models.

[0105] In some embodiments of the present specification, time series data of each time granularity among multiple time series data of different time granularities corresponding to each sub-training set are decomposed to obtain multiple subsequences corresponding to each sub-training set, including: using an empirical mode decomposition algorithm or a variational mode decomposition algorithm to decompose the time series data of each time granularity among multiple time series data of different time granularities corresponding to each sub-training set to extract trend data features and seasonal data features to obtain multiple subsequences corresponding to each sub-training set; the multiple subsequences include seasonal sequences and trend sequences.

[0106] In some embodiments of the present specification, a linear neural network is used to scale and sum multiple subsequences corresponding to each sub-training set, including: using a linear neural network to scale the trend data features corresponding to each sub-training set from short to long, and to scale the seasonal data features corresponding to each sub-training set from long to short, to obtain the scaled multiple subsequences corresponding to each sub-training set; and summing the scaled multiple subsequences corresponding to each sub-training set according to the multiple different time granularities to obtain sequence data of multiple different time granularities corresponding to each sub-training set.

[0107] In some embodiments of the present specification, the time series algorithm based on trend extraction includes at least one of the following: a long short-term memory network, a TimeMixer model, and an autoregressive moving average model.

[0108] In some embodiments of the present specification, the power prediction module is specifically used to: obtain multiple target power generation powers corresponding to each target moment according to the target power generation powers at each target moment in the future target time period corresponding to each timing algorithm model; calculate the confidence interval corresponding to each confidence level in multiple confidence levels at each target moment based on the multiple target power generation powers corresponding to each target moment; determine the steep drop probability and steep drop amplitude of wind and solar power generation at each target moment according to the confidence interval corresponding to each confidence level in multiple confidence levels at each target moment.

[0109] In some embodiments of the present specification, the steep drop determination module is specifically used to: for each target moment, take the difference between the power generation at the current moment and the upper limit closest to the power generation at the current moment in the confidence interval corresponding to each confidence level as the steep drop amplitude, and take the confidence level of the confidence interval corresponding to the steep drop amplitude as the steep drop probability corresponding to the steep drop amplitude.

[0110] This specification also provides a computer device, which can be found in Figure 5 The schematic diagram of the structure of the computer device composition of the wind-solar power generation steep drop prediction method for the wind-solar coupled hydrogen production system provided in the embodiment of this specification is shown, and the computer device may specifically include an input device 51, a processor 52, and a memory 53. Among them, the memory 53 is used to store processor executable instructions. When the processor 52 executes the instructions, the steps of the wind-solar power generation steep drop prediction method for the wind-solar coupled hydrogen production system described in any of the above embodiments are implemented.

[0111] In this embodiment, the input device may specifically be one of the main devices for information exchange between the user and the computer system. The input device may include a keyboard, a mouse, a camera, a scanner, a light pen, a handwriting input board, a voice input device, etc.; the input device is used to input the original data and the program for processing these numbers into the computer. The input device can also obtain and receive data transmitted from other modules, units, and devices. The processor can be implemented in any appropriate manner. For example, the processor can take the form of a computer-readable medium, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc., such as a microprocessor or a processor and a computer-readable program code (such as software or firmware) that can be executed by the (micro) processor. The memory may specifically be a memory device used to store information in modern information technology. The memory may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function without a physical form is also called a memory, such as a RAM, a FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, a TF card, etc.

[0112] In this embodiment, the functions and effects specifically realized by the computer device can be explained in comparison with other embodiments and will not be described in detail here.

[0113] The implementation manner of this specification also provides a computer storage medium based on the wind-solar power generation power drop prediction method for a wind-solar coupled hydrogen production system, wherein the computer storage medium stores computer program instructions, and when the computer program instructions are executed, the steps of the wind-solar power generation power drop prediction method for a wind-solar coupled hydrogen production system described in any of the above embodiments are implemented.

[0114] In this embodiment, the storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk (HDD) or a memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface for network connection communication set in accordance with the standard specified by the communication protocol.

[0115] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained in comparison with other embodiments and will not be described in detail here.

[0116] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of this specification can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of this specification are not limited to any specific combination of hardware and software.

[0117] It should be understood that the above description is for illustration and not for limitation. Many embodiments and many applications beyond the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this specification should not be determined with reference to the above description, but should be determined with reference to the preceding claims and the full scope of equivalents to which these claims belong.

[0118] The above description is only the preferred embodiment of this specification and is not intended to limit this specification. For those skilled in the art, the embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included in the protection scope of this specification.

Claims

1. A method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system, characterized in that: include: Obtain the actual power generation, meteorological model data and ground-based cloud map data of wind and solar power generation used in the wind-solar coupled hydrogen production system in the historical period; Generate a training sample set based on actual power generation, meteorological model data and ground-based cloud map data during the historical time period; Repeatedly sampling the data samples in the training sample set with replacement to obtain multiple sub-training sets; Using a time series algorithm based on trend extraction to learn the data samples in the multiple sub-training sets, to obtain multiple time series algorithm models; Obtain meteorological model data and ground-based cloud map data for wind and solar power generation in the future target time period; Input the meteorological model data and the ground-based cloud map data within the future target time period into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation within the future target time period corresponding to each time series algorithm model; Determine the power drop data of wind and solar power generation in the future target time period according to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model; The data samples in the plurality of sub-training sets are learned using a time series algorithm based on trend extraction to obtain a plurality of time series algorithm models, including: Using a linear neural network to scale and sum a plurality of subsequences corresponding to each of the plurality of sub-training sets, to obtain a plurality of sequence data of different time granularities corresponding to each of the sub-training sets; Using a time series algorithm based on trend extraction to learn a plurality of sequence data of different time granularities corresponding to each of the sub-training sets, to obtain a plurality of time series algorithm models; The linear neural network is used to scale and sum the multiple subsequences corresponding to each sub-training set, including: A linear neural network is used to scale the trend data features corresponding to each sub-training set from long to short, and the seasonal data features corresponding to each sub-training set from short to long, to obtain a plurality of scaled sub-sequences corresponding to each sub-training set; The scaled multiple subsequences corresponding to the sub-training sets are summed up according to the multiple different time granularities to obtain sequence data of the multiple different time granularities corresponding to the sub-training sets.

2. The method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system according to claim 1, characterized in that: Using a time series algorithm based on trend extraction to learn the data samples in the multiple sub-training sets to obtain multiple time series algorithm models, further comprising: Performing multi-level downsampling on the sample data of each sub-training set in the multiple sub-training sets to obtain time series data of multiple different time granularities corresponding to each sub-training set; The time series data of each time granularity among the multiple time series data of different time granularities corresponding to each of the sub-training sets are decomposed to obtain multiple sub-sequences corresponding to each of the sub-training sets.

3. The method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system according to claim 2 is characterized in that: Decomposing the time series data of each time granularity in the multiple time series data of different time granularities corresponding to each sub-training set to obtain multiple sub-sequences corresponding to each sub-training set, including: Using an empirical mode decomposition algorithm or a variational mode decomposition algorithm, the time series data of each time granularity in the time series data of multiple different time granularities corresponding to each sub-training set is decomposed to extract trend data features and seasonal data features, and obtain multiple subsequences corresponding to each sub-training set; the multiple subsequences include seasonal sequences and trend sequences.

4. The method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system according to claim 2, characterized in that: The time series algorithm based on trend extraction includes at least one of the following: Long short-term memory network, TimeMixer model and autoregressive moving average model.

5. The method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system according to claim 1, characterized in that: According to the target power generation of wind and solar power generation in the future target time period corresponding to each time series algorithm model, the power drop data of wind and solar power generation in the future target time period is determined, including: According to the target power generation at each target moment in the multiple target moments within the future target time period corresponding to each time series algorithm model, multiple target power generation corresponding to each target moment is obtained; Based on the multiple target power generation powers corresponding to the target moments, calculating the confidence interval corresponding to each confidence degree among the multiple confidence degrees at the target moments; According to the confidence interval corresponding to each confidence degree among the multiple confidence degrees at the target moments, the steep drop probability and steep drop amplitude of wind and solar power generation at the target moments are determined.

6. The method for predicting a steep drop in wind-solar power generation for a wind-solar coupled hydrogen production system according to claim 5, characterized in that: According to the confidence interval corresponding to each confidence degree among the multiple confidence degrees at each target moment, the steep drop probability and steep drop amplitude of wind and solar power generation at each target moment are determined, including: For each target moment, the difference between the generated power at the current moment and the upper limit closest to the generated power at the current moment in the confidence interval corresponding to each confidence level is taken as the steep drop amplitude, and the confidence level of the confidence interval corresponding to the steep drop amplitude is taken as the steep drop probability corresponding to the steep drop amplitude.

7. A wind-solar power generation power drop prediction device for a wind-solar coupled hydrogen production system, characterized in that: include: An acquisition module, used to acquire actual power generation, meteorological model data and ground-based cloud map data of wind-solar power generation used for a wind-solar coupled hydrogen production system in a historical period; It is also used to generate a training sample set based on the actual power generation, meteorological model data and ground-based cloud map data during the historical time period; A learning module, used for repeatedly sampling the data samples in the training sample set with replacement to obtain multiple sub-training sets; It is also used to learn the data samples in the multiple sub-training sets using a time series algorithm based on trend extraction to obtain multiple time series algorithm models; A power prediction module, used to obtain meteorological model data and ground-based cloud map data for wind and solar power generation in a future target time period; and also used to input the meteorological model data and ground-based cloud map data in the future target time period into each of the multiple time series algorithm models to obtain the target power generation of wind and solar power generation in the future target time period corresponding to each of the time series algorithm models; A steep drop determination module, used to determine the power steep drop data of wind and solar power generation in the future target time period according to the target power generation power of wind and solar power generation in the future target time period corresponding to each time series algorithm model; Among them, the learning module is specifically used to: use a linear neural network to scale and sum the multiple subsequences corresponding to each of the multiple subtraining sets to obtain multiple sequence data of different time granularities corresponding to each of the subtraining sets; use a time series algorithm based on trend extraction to learn the multiple sequence data of different time granularities corresponding to each of the subtraining sets to obtain multiple time series algorithm models; wherein, using a linear neural network to scale and sum the multiple subsequences corresponding to each of the subtraining sets includes: using a linear neural network to scale the trend data features corresponding to each of the subtraining sets from long to short, and scaling the seasonal data features corresponding to each of the subtraining sets from short to long, to obtain the scaled multiple subsequences corresponding to each of the subtraining sets; summing the scaled multiple subsequences corresponding to each of the subtraining sets according to the multiple different time granularities to obtain the multiple sequence data of different time granularities corresponding to each of the subtraining sets.

8. A computer device, characterized in that: The method comprises a processor and a memory for storing processor-executable instructions, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the instructions.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images