High-capacity ratio photovoltaic short-term power prediction method, device, equipment and storage medium

Abstract

The present disclosure provides a high-capacity ratio photovoltaic short-term power prediction method, device, equipment and storage medium, which relates to the technical fields of photovoltaic power generation and power prediction. Based on deep learning methods and combined with the business background, this method divides the station dataset into different weather types by means of historical output data and meteorological data, extracts the ramp characteristics of the station under different weather conditions, and thus learns and mines the accurate correspondence relationship between the actual weather, time period and power of the high-capacity ratio photovoltaic power station through the training of deep learning ideas. Furthermore, the accuracy of the photovoltaic short-term prediction results is higher, and the prediction error in the output ramp stage of the high-capacity ratio station is effectively reduced.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing, specifically to the technical fields of photovoltaic power generation and power prediction technology, and particularly to a short-term power prediction method, device, electronic device, computer-readable storage medium and computer program product for high-capacity ratio photovoltaic power generation. Background Art

[0002] When the weather condition of a high-capacity ratio photovoltaic power station is sunny, its output characteristics are significantly different from those of a normal power station, mainly reflected in the ramp-up stages in the morning and afternoon.

[0003] Since the DC-side capacity increase design is the best choice for the capacity expansion ratio of a photovoltaic power station, the output of a high-capacity ratio power station can usually reach about 85% of the grid connection capacity of the power station before 9:00 in the morning. Compared with the traditional power prediction method based on irradiance, this process is difficult to be accurately captured.

[0004] How to accurately capture and predict the actual power is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] Embodiments of the present disclosure propose a short-term power prediction method, device, electronic device and computer-readable storage medium for high-capacity ratio photovoltaic power generation.

[0006] In a first aspect, embodiments of the present disclosure propose a short-term power prediction method for high-capacity ratio photovoltaic power generation, including: for sunny days, obtaining the historical actual power and corresponding timestamps of a high-capacity ratio photovoltaic power generation station, and respectively performing normalization processing on the historical actual power and timestamps to obtain normalized data; using a convolutional neural network with convolutional layers and pooling layers of different sizes to extract features on different time scales in the normalized data to obtain local features; using an attention mechanism to weight the local features processed by a long short-term memory network to obtain target features represented in the form of tensors and the predicted power output by the long short-term memory network, and respectively constructing a rising rate feature and a time interval feature according to the predicted power; obtaining historical weather forecast data corresponding to the time period of the historical actual power, and after standardizing the historical weather forecast data, superimposing it with the target features, the rising rate features and the time interval features, and constructing a first training sample with the historical actual power of the corresponding time period as the target truth value; predicting the actual power based on the network framework of the long short-term memory network for the first training sample, and training to obtain a sunny day power prediction model; predicting the power of a target time period within a preset duration from the current based on the sunny day power prediction model to obtain a prediction result.

[0007] In some other embodiments of the first aspect, respectively performing normalization processing on the historical actual power and timestamps to obtain normalized data includes:

[0008] Convert the time stamp with the preset duration as the time resolution into a numerical feature;

[0009] Perform sine and cosine processing on the numerical feature to obtain normalized time data;

[0010] Perform normalization processing on the historical actual power to obtain normalized power data.

[0011] In some other embodiments of the first aspect, construct a rising rate feature and a time interval feature according to the predicted power, including:

[0012] Calculate the difference of the predicted power and divide the difference calculation result by the time interval to obtain a rising rate feature representing the power change speed;

[0013] Calculate the difference between the predicted powers in adjacent time intervals to obtain a time interval feature representing the power ramp speed.

[0014] In some other embodiments of the first aspect, a high-capacity ratio photovoltaic power station refers to a photovoltaic power station in which the ratio of the installed capacity of photovoltaic modules to the rated capacity of the inverter is higher than a preset threshold.

[0015] In some other embodiments of the first aspect, the method further includes:

[0016] For cloudy days, obtain the historical meteorological data and historical actual power of the high-capacity ratio photovoltaic power station, calculate the clarity index and direct ratio according to the historical meteorological data and historical actual power, and calculate the correlation coefficient corresponding to the historical actual power after standardizing the clarity index and direct ratio;

[0017] Construct a clustering index as the classification index of the clustering algorithm for the three meteorological variables with the highest correlation coefficient calculated with the historical actual power, and divide the weather types into three categories according to the classification index;

[0018] According to the correlation between meteorological variables and the actual output of power generation equipment in a high-capacity ratio photovoltaic power station under different weather types, remove meteorological variables with low or no correlation to obtain high-correlation meteorological variables;

[0019] Construct second training samples for the high-correlation meteorological variables and the corresponding historical actual power under different weather types respectively, and train the second training samples under different weather types based on the network framework of the long short-term memory network to train a cloud day power prediction model;

[0020] Correspondingly, based on the clear day power prediction model, predict the power of the target time period not exceeding the preset duration from the current time to obtain a prediction result, including:

[0021] Predict the power of sunny periods in the target period according to the sunny-day power prediction model to obtain the sunny-day period prediction result;

[0022] Predict the power of cloudy and overcast periods under the corresponding weather types in the target period according to the cloudy and overcast day power prediction model to obtain the cloudy and overcast day period prediction result;

[0023] Summarize the prediction results according to the sunny-day period prediction result and the cloudy and overcast day period prediction result.

[0024] In some other embodiments of the first aspect, the historical meteorological data includes:

[0025] Predict at least one of global radiation, diffuse radiation, direct radiation, total cloud cover, wind speed, temperature, relative humidity, precipitation, air pressure, visibility.

[0026] In some other embodiments of the first aspect, the method further includes:

[0027] Obtain the historical predicted global radiation and historical predicted total cloud cover of the high-capacity ratio photovoltaic power station;

[0028] Construct a weather classification comprehensive index according to the historical predicted global radiation, historical predicted total cloud cover and the corresponding weighting coefficients;

[0029] Classify the weather at the moments in the target period that exceed the preset threshold as cloudy and overcast;

[0030] Classify the weather at the moments in the target period that do not exceed the preset threshold as sunny.

[0031] In a second aspect, embodiments of the present disclosure propose a high-capacity ratio photovoltaic short-term power prediction device, including: a normalization processing unit configured to, for sunny days, obtain the historical actual power and corresponding timestamps of a high-capacity ratio photovoltaic power station, and perform normalization processing on the historical actual power and timestamps respectively to obtain normalized data; a local feature extraction unit configured to extract features on different time scales in the normalized data using a convolutional neural network with convolutional layers and pooling layers of different sizes to obtain local features; a feature construction unit configured to use an attention mechanism to weight the local features processed by a long short-term memory network to obtain target features represented in the form of tensors and the predicted power output by the long short-term memory network, and respectively construct an ascending rate feature and a time interval feature based on the predicted power; a first training sample construction unit configured to obtain historical weather forecast data corresponding to the time period of the historical actual power, and superimpose the historical weather forecast data after standardization processing with the target features, the ascending rate feature, and the time interval feature, and construct a first training sample with the historical actual power of the corresponding time period as the target true value; a sunny-day power prediction model construction unit configured to predict the actual power based on the network framework of the long short-term memory network using the first training sample, and train to obtain a sunny-day power prediction model; and a prediction unit configured to predict the power of a target time period not exceeding a preset duration from the current based on the sunny-day power prediction model to obtain a prediction result.

[0032] In some other embodiments of the second aspect, the normalization processing unit is further configured to:

[0033] Convert the timestamps with a preset duration as the time resolution into numerical features;

[0034] Perform sine-cosine processing on the numerical features to obtain normalized time data;

[0035] Perform normalization processing on the historical actual power to obtain normalized power data.

[0036] In some other embodiments of the second aspect, the feature construction unit includes a feature construction subunit configured to respectively construct an ascending rate feature and a time interval feature based on the predicted power. The feature construction subunit is further configured to:

[0037] Calculate the difference of the predicted power, and divide the difference calculation result by the time interval to obtain an ascending rate feature representing the power change speed;

[0038] Calculate the difference between the predicted powers in adjacent time intervals to obtain a time interval feature representing the power ramp-up speed.

[0039] In some other embodiments of the second aspect, a high-capacity ratio photovoltaic power station refers to a photovoltaic power station in which the ratio between the installed capacity of photovoltaic modules and the rated capacity of an inverter is higher than a preset threshold.

[0040] In some other embodiments of the second aspect, the apparatus further comprises:

[0041] A correlation coefficient calculation unit, configured to obtain historical meteorological data and historical actual power of a high-capacity ratio photovoltaic power station for cloud sky, calculate a clarity index and a direct ratio based on the historical meteorological data and the historical actual power, and calculate a correlation coefficient corresponding to the historical actual power after standardizing the clarity index and the direct ratio;

[0042] A weather type classification unit, configured to construct a clustering index as a classification index of a clustering algorithm with three meteorological variables having the highest correlation coefficient with the historical actual power calculated, and classify the weather type into three categories according to the classification index;

[0043] A high-correlation meteorological variable determination unit, configured to remove meteorological variables with low or no correlation according to the correlation between meteorological variables and the actual output of power generation equipment in a high-capacity ratio photovoltaic power station under different weather types, to obtain high-correlation meteorological variables;

[0044] A second training sample construction and cloud sky power prediction model construction unit, configured to construct second training samples for the high-correlation meteorological variables and the corresponding historical actual power under different weather types respectively, and train the second training samples under different weather types based on the network framework of a long short-term memory network respectively, to train and obtain a cloud sky power prediction model;

[0045] Correspondingly, the prediction unit is further configured to:

[0046] Predict the power during sunny periods in a target period according to a sunny day power prediction model, to obtain a sunny period prediction result;

[0047] Predict the power during cloud sky periods of the corresponding weather type in the target period according to the cloud sky power prediction model, to obtain a cloud sky period prediction result;

[0048] Summarize the sunny period prediction result and the cloud sky period prediction result to obtain a prediction result.

[0049] In some other embodiments of the second aspect, the historical meteorological data includes:

[0050] At least one of predicted global radiation, diffuse radiation, direct radiation, total cloud cover, wind speed, temperature, relative humidity, precipitation, air pressure, visibility.

[0051] In some other embodiments of the second aspect, the apparatus further includes:

[0052] Obtain the historical predicted total radiation and historical predicted total cloud amount of a high-capacity ratio photovoltaic power station;

[0053] Construct a weather classification comprehensive index according to the historical predicted total radiation, historical predicted total cloud amount, and corresponding weighting coefficients;

[0054] Classify the weather at the moments in the target period that exceed a preset threshold as cloudy days;

[0055] Classify the weather at the moments in the target period that do not exceed a preset threshold as sunny days.

[0056] In a third aspect, an embodiment of the present disclosure provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to implement the high-capacity ratio photovoltaic short-term power prediction method described in any implementation manner of the first aspect.

[0057] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions, and the computer instructions are used to enable a computer to implement the high-capacity ratio photovoltaic short-term power prediction method described in any implementation manner of the first aspect when executed.

[0058] To solve the problem that the output of a high-capacity ratio power station increases rapidly and the power is difficult to predict during the ramp-up stage, the high-capacity ratio photovoltaic short-term power prediction solution provided by the present disclosure is based on a deep learning method and combines the business background, and divides the power station data set into different weather types through historical output data and meteorological data, and extracts the ramp-up characteristics of the power station under different weather conditions, so as to learn and mine the accurate corresponding relationship between the actual weather, time period, and power of the high-capacity ratio photovoltaic power station through the deep learning idea, thereby making the accuracy of the photovoltaic short-term prediction result higher and effectively reducing the prediction error during the ramp-up stage of the output of the high-capacity ratio power station.

[0059] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives, and advantages of the present disclosure will become more apparent:

[0061] Figure 1Flowchart of a high-capacity ratio photovoltaic short-term power prediction method provided by an embodiment of the present disclosure;

[0062] Figure 2 Flowchart of a cloud-sky power prediction model and a prediction result determination method provided by an embodiment of the present disclosure;

[0063] Figure 3 Schematic flowchart of another high-capacity ratio photovoltaic short-term power prediction method provided by an embodiment of the present disclosure;

[0064] Figure 4 Structural block diagram of a high-capacity ratio photovoltaic short-term power prediction device provided by an embodiment of the present disclosure;

[0065] Figure 5 Schematic structural diagram of an electronic device suitable for executing the high-capacity ratio photovoltaic short-term power prediction method provided by an embodiment of the present disclosure. Detailed implementation manners

[0066] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted below. It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0067] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0068] Please refer to Figure 1 , Figure 1 Flowchart of a high-capacity ratio photovoltaic short-term power prediction method provided by an embodiment of the present disclosure, which includes the following steps:

[0069] Step 101: For sunny days, obtain the historical actual power and the corresponding timestamps of the high-capacity ratio photovoltaic power station, and perform normalization processing on the historical actual power and the timestamps respectively to obtain normalized data;

[0070] This step aims to obtain the historical actual power and the corresponding timestamps of the high-capacity ratio photovoltaic power station as the target object under sunny days by the execution subject of the high-capacity ratio photovoltaic short-term power prediction method (such as a local computing terminal or a remote computing terminal, such as a local workstation or a remote server), and perform normalization processing on the historical actual power and the timestamps respectively to obtain normalized data.

[0071] Among them, the capacity ratio of a photovoltaic power station refers to the ratio between the installed capacity of photovoltaic modules and the rated capacity of an inverter. This ratio reflects the matching relationship between photovoltaic modules and inverters in a photovoltaic power station and has an important impact on the power generation efficiency and operation safety of the power station. An overly small inverter capacity leads to power loss and unstable operation. Therefore, when selecting an inverter, ensure that its rated capacity can meet the maximum output power of the system; an overly large inverter capacity will result in low efficiency when the system operates at low load. Therefore, the capacity of the inverter should be reasonably selected according to the actual situation. A high-capacity-ratio photovoltaic power station refers to a photovoltaic power station in which the ratio between the installed capacity of photovoltaic modules and the rated capacity of an inverter is higher than a preset threshold.

[0072] Specifically, the historical actual power and corresponding timestamps of a high-capacity-ratio photovoltaic power station as the target object can also be recorded and represented in the form of a power sequence. Therefore, the subsequent normalization process can also be regarded as a process of normalizing the power information and time information in the power sequence respectively. One normalization implementation method including but not limited to can be:

[0073] Convert the timestamps with a preset duration as the time resolution into numerical features, then perform sine-cosine processing on the numerical features to obtain normalized time data; and perform normalization processing on the historical actual power to obtain normalized power data. Furthermore, through the normalization processing, various numerical values within a preset range are obtained to facilitate subsequent unified processing.

[0074] Step 102: Use a convolutional neural network with convolutional layers and pooling layers of different sizes to extract features at different time scales from the normalized data to obtain local features;

[0075] Based on step 101, the purpose of this step is for the above-mentioned execution entity to extract features at different time scales from the normalized data by means of a specific convolutional neural network (this network is designed to process features at different time scales in the normalized data) to obtain local features for characterizing the ramping situation.

[0076] Among them, convolutional layers of different sizes mean that convolutional kernels of different sizes are adopted in the convolutional layers of the network. A convolutional kernel is a small matrix that slides on the input data and extracts features through convolutional operations on the input data. Using convolutional kernels of different sizes helps to capture features at different scales, thereby improving the network's ability to understand the input data; pooling layers of different sizes mean that the network may use pooling windows of different sizes, which helps to abstract features at different levels, making the network more sensitive to features at different time scales of the input data. Therefore, the main objective of the network under this structural setting is to learn and extract features from the normalized time series data. By adopting convolutional kernels and pooling windows of different sizes, the network can model features at different time scales, thereby more comprehensively understanding the structure and pattern of the input data.

[0077] The advantage of this architecture is that it can adapt to changes at different time scales, making the network more versatile and not just applicable to a specific time scale. This is very useful for dealing with multi-scale features (such as fluctuations and trends) in time series data.

[0078] Step 103: Use the attention mechanism to weight the local features processed by the long short-term memory network to obtain the target features represented in tensor form and the predicted power output by the long short-term memory network, and respectively construct the rising rate feature and the time interval feature according to the predicted power.

[0079] Based on step 102, this step aims to have the above-mentioned execution entity use the output local features as the input of the long short-term memory network (LSTM), and further screen through the three gates (input gate, forget gate, and output gate) and memory unit of the LSTM to retain important features and forget unimportant features, and then use the attention mechanism to weight the output of the LSTM to highlight important features, and save the weighted features in tensor form and record them as the target features and finally output the predicted power, and the ramp characteristics of the power of the power station can be analyzed using the predicted power. Specifically, the rising rate feature and the time interval feature are respectively constructed. The rising rate feature is obtained by calculating the difference of the predicted power and dividing it by the time interval. This index represents the change speed of the power, while the time interval feature is the difference between adjacent time steps, and the value size represents the ramp speed of the power.

[0080] Step 104: Obtain the historical weather forecast data corresponding to the time period of the historical actual power, and after normalizing the historical weather forecast data, superimpose it with the target features, the rising rate feature, and the time interval feature, and construct the first training sample with the historical actual power of the corresponding time period as the target truth value.

[0081] Based on step 103, this step aims to obtain historical weather forecast data corresponding to the time period of the historical actual power from the above-mentioned execution entity, and after standardizing the historical weather forecast data (for example, specifically using Z-score standardization), superimpose it with the target feature, the rising rate feature, and the time interval feature, and construct the first training sample with the historical actual power of the corresponding time period as the target truth value.

[0082] It should be noted that the reason for superimposing the above four features is that the historical weather forecast data is used as the actual weather data of the corresponding time period, the target feature is used as the ramp feature represented at the global level, and the rising rate feature and the time interval feature exist as two different dimensional features representing the ramp respectively. That is, the superposition of the above four features actually makes the finally formed multi-dimensional features cover all dimensions as much as possible.

[0083] Step 105: Predict the actual power based on the network framework of the long short-term memory network for the first training sample, and train to obtain a sunny-day power prediction model;

[0084] Based on step 104, this step aims to predict the actual power based on the network framework of the long short-term memory network for the first training sample by the above-mentioned execution entity, and train to obtain a sunny-day power prediction model. Further, a homologous first test sample can be divided while constructing the first training sample, and then the effectiveness of the preliminarily trained sunny-day power prediction model is tested through the first test sample, and finally the model that passes the effectiveness test is output as an available sunny-day power prediction model.

[0085] Step 106: Predict the power of the target time period within a preset duration from the current time based on the sunny-day power prediction model to obtain a prediction result.

[0086] Based on step 105, this step aims to predict the power of the target time period within a preset duration from the current time based on the sunny-day power prediction model by the above-mentioned execution entity to obtain a prediction result.

[0087] Among them, the target time period is the time period within a preset duration from the current time, so as to complete the accurate prediction of the power in the near future based on the periodicity and relevance in time, and avoid making long-term inaccurate power predictions.

[0088] To solve the problem that the output of a high-capacity-ratio power station increases rapidly during the ramp-up stage and the power is difficult to predict, the short-term power prediction method for high-capacity-ratio photovoltaic power generation provided by this disclosure is based on deep learning methods and combines the business background. The power station dataset is divided into different weather types through historical output data and meteorological data. By extracting the ramp-up characteristics of the power station under different weather conditions, the accurate corresponding relationship between the actual weather, time period, and power of the high-capacity-ratio photovoltaic power generation station is learned and mined with the help of deep learning ideas. Furthermore, the accuracy of the short-term photovoltaic prediction results is higher, and the prediction error during the output ramp-up stage of the high-capacity-ratio power station is effectively reduced.

[0089] Considering that the target time period may not only include sunny days with less cloud cover, but also cloudy days may occur. Therefore, in order to obtain more accurate prediction results, please also refer to Figure 2 , Figure 2 which is a flowchart of a cloudy-day power prediction model and a prediction result determination method provided by an embodiment of this disclosure, including the following steps:

[0090] Step 201: For cloudy days, obtain the historical meteorological data and historical actual power of the high-capacity-ratio photovoltaic power generation station, calculate the clarity index and direct ratio according to the historical meteorological data and historical actual power, and calculate the correlation coefficient corresponding to the historical actual power after standardizing the clarity index and direct ratio;

[0091] The purpose of this step is for the above-mentioned execution subject to obtain the historical meteorological data and historical actual power of the high-capacity-ratio photovoltaic power generation station under cloudy-day conditions, calculate the clarity index and direct ratio according to the historical meteorological data and historical actual power, and calculate the correlation coefficient corresponding to the historical actual power after standardizing the clarity index and direct ratio.

[0092] Among them, considering the complexity under cloudy-day conditions, it is also possible to only obtain the meteorological data and actual power within the historical time period not earlier than the preset time period from the current time, so as to show that the meteorological data and actual power in the too early time period have no reference value. For example, only use the meteorological data and actual power within no more than half a year from the current time.

[0093] Specifically, the meteorological data may include predicted global horizontal irradiance ghi, diffuse horizontal irradiance dhi, direct normal irradiance dni, total cloud cover C, wind speed W, temperature T, relative humidity RH, precipitation RA, air pressure p, visibility V and other data, and calculate the clarity index k T and direct ratio B d . Standardize the above and calculate the correlation coefficient P with the actual power. Among them, the clarity index k T The calculation formula is: Where I represents the total hourly solar radiation on the horizontal plane, and I0 represents the hourly solar radiation on the horizontal plane outside the atmosphere. In the formula, γ is the solar azimuth angle, δ is the solar declination angle, and E sc is the solar constant, and E sc = 1367 W / m 2 .

[0094] Step 202: Construct a clustering index as the classification index of the clustering algorithm from the three meteorological variables with the highest correlation coefficient calculated with the historical actual power, and divide the weather types into three categories according to the classification index.

[0095] Based on Step 201, this step aims to have the above-mentioned execution entity calculate the three meteorological variables with the highest correlation coefficient with the historical actual power (such as the horizontal total radiation ghi, the clarity index k T and the total cloud cover C), and construct a clustering index by combining the respective weight coefficients of these three meteorological variables. At the same time, use it as the classification index of the clustering algorithm, and divide the weather types into three categories according to the classification index.

[0096] Step 203: Remove the meteorological variables with low or no correlation according to the correlation between the meteorological variables and the actual output of the power generation equipment in the high-capacity ratio photovoltaic power station under different weather types, and obtain the meteorological variables with high correlation.

[0097] Based on Step 202, this step aims to have the above-mentioned execution entity remove the meteorological variables with low or no correlation according to the correlation between the meteorological variables and the actual output of the power generation equipment in the high-capacity ratio photovoltaic power station under different weather types, and obtain the meteorological variables with high correlation, so as to remove the correlated variables and only retain the meteorological variables with high correlation to simplify the calculation.

[0098] Step 204: Construct second training samples for the meteorological variables with high correlation and the corresponding historical actual power under different weather types respectively, and train the second training samples under different weather types based on the network framework of the long short-term memory network to obtain a cloud power prediction model.

[0099] Based on Step 203, this step aims to have the above-mentioned execution entity construct second training samples for the meteorological variables with high correlation and the corresponding historical actual power under different weather types respectively (since there are three weather types, three second training samples will be constructed), and train the second training samples under different weather types based on the network framework of the long short-term memory network to obtain a cloud power prediction model.

[0100] Further, while constructing the second training samples, homologous second test samples can be divided out. Then, the cloud sky power prediction model obtained through preliminary training is tested for effectiveness using the second test samples, and finally, the model that passes the effectiveness test is output as the available cloud sky power prediction model.

[0101] Step 205: Predict the power during sunny periods in the target period according to the sunny day power prediction model to obtain the sunny period prediction result.

[0102] Step 206: Predict the power during cloud sky periods under the corresponding weather type in the target period according to the cloud sky power prediction model to obtain the cloud sky period prediction result.

[0103] Step 207: Aggregate the sunny period prediction result and the cloud sky period prediction result to obtain the prediction result.

[0104] Steps 205 - 207 respectively use the sunny day power prediction model and the cloud sky power prediction model to predict the power during sunny periods and cloud sky periods that make up the target period, and aggregate the prediction results output by each to facilitate obtaining a more accurate power prediction result jointly output by prediction models adapted to different weathers.

[0105] Further, in order to distinguish between sunny days and cloud skies, the historical predicted total radiation and historical predicted total cloud cover of the high-capacity ratio photovoltaic power station can also be obtained in advance. Then, according to the historical predicted total radiation, the historical predicted total cloud cover, and the corresponding weighting coefficients, a weather classification comprehensive index is constructed. Next, the weather at the moments in the target period that exceed the preset threshold is classified as cloud sky, and the weather at the moments in the target period that do not exceed the preset threshold is classified as sunny day. Furthermore, only the actual weather classification comprehensive index at one moment needs to be calculated and compared with the preset threshold to determine whether it belongs to the sunny day state or the cloud sky state.

[0106] For better understanding, the present disclosure also combines a specific application scenario and gives a complete and specific implementation solution. Please refer to the Figure 3 flow schematic diagram shown as follows:

[0107] Step 1: Obtain the historical predicted total radiation ghi and historical predicted total cloud cover C of the photovoltaic power station.

[0108] Define the comprehensive weather classification index W = predicted global horizontal irradiance ghi * predicted global horizontal irradiance weight W1 + ((1 - predicted total cloud cover C) * 100) * predicted total cloud cover weight W2, where W1 and W2 can be set according to experience. Since the output characteristics of the PV power station are relatively obvious, only the data during the daytime period is selected. According to the actual situation and requirements, the threshold of the comprehensive weather classification index W can be set to 350. If the comprehensive index W is higher than 350, it indicates that the weather type is sunny. When this value is lower than or equal to 350, it is classified as cloudy sky.

[0109] Step 2: Establish different models for sunny days and cloudy sky respectively:

[0110] Under sunny conditions, obtain the historical actual power data of the station and the corresponding timestamps (time resolution 15 min). Convert the timestamps into numerical features of each time unit such as month, day, hour, and minute. And perform sine and cosine processing on the month, day, hour, and minute numbers, and normalize the actual power data.

[0111] Step 3: Feature extraction:

[0112] Take the preprocessed data as the input, use the CNN model, and extract features on different time scales by using convolutional layers and pooling layers of different sizes, and extract local features in the power sequence.

[0113] The ramp features extracted after the power sequence is recognized by the CNN are used as the input of the LSTM, and are further screened through the three gates (input gate, forget gate, and output gate) and memory unit of the LSTM to retain important features and forget unimportant features. Then, use the attention mechanism to weight the output of the LSTM to highlight important features, and save the weighted features in the form of a tensor and record it as CL_F, and finally output the predicted power PF1.

[0114] Using the predicted power PF1, the ramp features of the power of the station can be analyzed, and the rising rate feature and time interval feature are constructed respectively. The rising rate feature is obtained by calculating the difference of PF1 and dividing it by the time interval, and this index represents the change speed of the power. The time interval feature is the difference between adjacent time steps, and the value of this represents the ramp speed of the power.

[0115] Step 6: Obtain the weather forecast data for the time period corresponding to the above historical actual power data, including predicted meteorological data such as predicted total radiation, temperature, humidity, and pressure, as well as the transformed timestamp data. Perform Z-score standardization on the above data, and superimpose the above features with CL_F, the rising rate, and the time interval. Using the historical actual power as the target value, divide the training set and the test set, and construct a new LSTM model to predict the actual power, so as to obtain the predicted power under sunny conditions at high-capacity ratio stations.

[0116] Step 7: Under cloudy and overcast conditions, this type needs to be further subdivided. Select meteorological and measured power data with a 15-minute resolution for about half a year. The meteorological data includes predicted total radiation ghi, diffuse radiation dhi, direct radiation dni, total cloud cover C, wind speed W, temperature T, relative humidity RH, precipitation RA, air pressure p, visibility V, etc. Calculate the clarity index k T and the direct ratio B d . Perform standardization processing on the above, and calculate the correlation coefficient P with the actual power. The calculation formula for the clarity index k T is as follows: where I represents the total hourly solar radiation on the horizontal plane, and I0 represents the hourly solar radiation on the horizontal plane outside the atmosphere. In the formula, γ is the solar azimuth angle, δ is the solar declination angle, E sc is the solar constant, and E sc = 1367 W / m 2 .

[0117] Step 8: Finally, select the three variables with the highest correlation, namely horizontal total radiation ghi, clarity index k T and total cloud cover C, and construct the clustering index H = ω1ghi + ω2k T + ω3C, where ω1, ω2, and ω3 are weight coefficients, which are proportional to the correlation coefficient P and sum to 1. As the classification index of the K-means clustering algorithm, the weather types are finally divided into three categories.

[0118] Step 9: Under different weather types, the influence degrees of input variables on photovoltaic power output are different. According to the correlation between meteorological variables and actual output under different weather types, eliminate variables with low or no correlation. The input variables corresponding to the three types of weather are shown in Table 1 below:

[0119] Table 1 Selection of input variables for different weather types

[0120] Weather type Input variable Weather type 1 <![CDATA[ghi,C,k T ,]]> Cloudy Weather type 2 <![CDATA[ghi,C,k T ,RH]]> Overcast Weather type 3 ghi, RH, T Rainy

[0121] Step 10: Finally, using the historical actual power as the target value, divide the training set and the test set, and construct a new LSTM model to predict the actual power, so as to obtain the predicted power under the cloud sky conditions of the high-capacity ratio power station.

[0122] The solution provided in this embodiment is based on the deep learning method and combines the business background. Through the historical output data and meteorological data, the power station dataset is divided into two weather types: sunny days and cloud sky days. By extracting the ramp characteristics of the power station under sunny day conditions and further subdividing the cloud sky type, datasets with different input variables are constructed, and the predicted power under sunny day conditions and cloud sky conditions are obtained respectively. Finally, a stable and reliable short-term predicted power can be obtained.

[0123] Further referring to Figure 4 , as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a high-capacity ratio photovoltaic short-term power prediction device. This device embodiment corresponds to the Figure 1 method embodiment shown, and this device can be specifically applied to various electronic devices.

[0124] As shown in Figure 4 , the high-capacity ratio photovoltaic short-term power prediction device 400 of this embodiment may include: a normalization processing unit 401, a local feature extraction unit 402, a feature construction unit 403, a first training sample construction unit 404, a sunny day power prediction model construction unit 405, and a prediction unit 406. Among them, the normalization processing unit 401 is configured to, for sunny days, obtain the historical actual power and the corresponding time stamps of the high-capacity ratio photovoltaic power station, and perform normalization processing on the historical actual power and the time stamps respectively to obtain normalized data; the local feature extraction unit 402 is configured to use a convolutional neural network with convolutional layers and pooling layers of different sizes to extract features on different time scales in the normalized data to obtain local features; the feature construction unit 403 is configured to use the attention mechanism to weight the local features processed by the long short-term memory network to obtain the target features represented in the form of tensors and the predicted power output by the long short-term memory network, and respectively construct the rising rate feature and the time interval feature according to the predicted power; the first training sample construction unit 404 is configured to obtain the historical weather forecast data corresponding to the time period of the historical actual power, and after standardizing the historical weather forecast data, superimpose it with the target features, the rising rate features, and the time interval features, and construct the first training sample with the historical actual power of the corresponding time period as the target true value; the sunny day power prediction model construction unit 405 is configured to predict the actual power based on the network framework of the long short-term memory network for the first training sample, and train to obtain a sunny day power prediction model; the prediction unit 406 is configured to predict the power of the target time period not exceeding the preset duration from the current based on the sunny day power prediction model to obtain a prediction result.

[0125] In this embodiment, in the high-capacity ratio photovoltaic short-term power prediction device 400: For the specific processing of the normalization processing unit 401, the local feature extraction unit 402, the feature construction unit 403, the first training sample construction unit 404, the sunny day power prediction model construction unit 405, and the prediction unit 406 and the technical effects brought thereby, reference can be made respectively to Figure 2 the relevant descriptions of steps 101-106 in the corresponding embodiment, which will not be elaborated here.

[0126] In some alternative implementation manners of this embodiment, the normalization processing unit 401 is further configured to:

[0127] Convert the time stamp with a preset duration as the time resolution into a numerical feature;

[0128] Perform sine-cosine processing on the numerical feature to obtain normalized time data;

[0129] Perform normalization processing on the historical actual power to obtain normalized power data.

[0130] In some alternative implementation manners of this embodiment, the feature construction unit 403 includes a feature construction subunit configured to respectively construct an ascending rate feature and a time interval feature according to the predicted power, and the feature construction subunit is further configured to:

[0131] Calculate the difference of the predicted power, and divide the difference calculation result by the time interval to obtain an ascending rate feature representing the power change speed;

[0132] Calculate the difference between the predicted powers in adjacent time intervals to obtain a time interval feature representing the power ramp-up speed.

[0133] In some alternative implementation manners of this embodiment, a high-capacity ratio photovoltaic power generation station refers to a photovoltaic power generation station in which the ratio of the installed capacity of photovoltaic modules to the rated capacity of an inverter is higher than a preset threshold.

[0134] In some alternative implementation manners of this embodiment, the high-capacity ratio photovoltaic short-term power prediction device 400 further includes:

[0135] A correlation coefficient calculation unit, configured to obtain the historical meteorological data and the historical actual power of the high-capacity ratio photovoltaic power generation station for cloudy days, calculate the clarity index and the direct ratio according to the historical meteorological data and the historical actual power, and calculate the correlation coefficient corresponding to the historical actual power after performing standardization processing on the clarity index and the direct ratio;

[0136] A weather type classification unit, configured to calculate three meteorological variables with the highest correlation coefficient with the historical actual power, construct a clustering index as a classification index for a clustering algorithm, and classify weather types into three categories according to the classification index;

[0137] A high-correlation meteorological variable determination unit, configured to remove meteorological variables with low or no correlation according to the correlation between meteorological variables and the actual output of power generation equipment in a high-capacity ratio photovoltaic power station under different weather types, and obtain high-correlation meteorological variables;

[0138] A second training sample construction and cloud-day power prediction model construction unit, configured to construct second training samples for the high-correlation meteorological variables and the corresponding historical actual power under different weather types respectively, and train the second training samples under different weather types based on the network framework of a long short-term memory network to obtain a cloud-day power prediction model;

[0139] Correspondingly, the prediction unit 406 can be further configured to:

[0140] Predict the power of sunny periods in the target period according to the sunny-day power prediction model to obtain a sunny-day period prediction result;

[0141] Predict the power of cloud-day periods under the corresponding weather types in the target period according to the cloud-day power prediction model to obtain a cloud-day period prediction result;

[0142] Summarize the prediction results according to the sunny-day period prediction result and the cloud-day period prediction result.

[0143] In some optional implementation manners of this embodiment, the historical meteorological data includes:

[0144] At least one of predicted global radiation, diffuse radiation, direct radiation, total cloud cover, wind speed, temperature, relative humidity, precipitation, air pressure, and visibility.

[0145] In some optional implementation manners of this embodiment, the high-capacity ratio photovoltaic short-term power prediction device 400 further includes:

[0146] Obtain the historical predicted global radiation and historical predicted total cloud cover of the high-capacity ratio photovoltaic power station;

[0147] Construct a weather classification comprehensive index according to the historical predicted global radiation, historical predicted total cloud cover, and the corresponding weighting coefficients;

[0148] Classify the weather at the moments in the target period that exceed a preset threshold as cloud days;

[0149] Classify the weather at the moments in the target period that do not exceed the preset threshold as sunny days.

[0150] This embodiment exists as a device embodiment corresponding to the above method embodiment.

[0151] To solve the problem that the output of a high-capacity-ratio power station increases rapidly and the power is difficult to predict during the ramp-up stage, the high-capacity-ratio photovoltaic short-term power prediction device provided by the present disclosure is based on a deep learning method and combines the business background. The station data set is divided into different weather types by historical output data and meteorological data. By extracting the ramp-up characteristics of the station under different weather conditions, the learning and mining of the accurate correspondence relationship between the actual weather, time period, and power of the high-capacity-ratio photovoltaic power station trained by means of deep learning ideas are carried out. Furthermore, the accuracy of the photovoltaic short-term prediction result is higher, and the prediction error during the output ramp-up stage of the high-capacity-ratio power station is effectively reduced.

[0152] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor can implement the high-capacity-ratio photovoltaic short-term power prediction method described in any of the above embodiments.

[0153] According to an embodiment of the present disclosure, the present disclosure also provides a readable storage medium, which stores computer instructions for enabling a computer to implement the high-capacity-ratio photovoltaic short-term power prediction method described in any of the above embodiments when executed.

[0154] Figure 5 FIG. shows a schematic block diagram of an exemplary electronic device 500 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0155] As Figure 5As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0156] Multiple components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, an optical disc, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0157] The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 501 executes the various methods and processes described above, such as the high-capacity ratio photovoltaic short-term power prediction method. For example, in some embodiments, the high-capacity ratio photovoltaic short-term power prediction method can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as the storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the high-capacity ratio photovoltaic short-term power prediction method described above can be executed. Alternatively, in other embodiments, the computing unit 501 can be configured to execute the high-capacity ratio photovoltaic short-term power prediction method in any other appropriate manner (e.g., by means of firmware).

[0158] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0159] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine, or entirely on the remote machine or server.

[0160] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0161] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and a pointing device (e.g., a mouse or a trackball), by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and the input received from the user can be in any form (including acoustic input, speech input, or tactile input).

[0162] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0163] A computer system can include a client and a server. The client and the server are generally far from each other and usually interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the problems of high management difficulty and weak business scalability existing in traditional physical hosts and virtual private server (VPS, Virtual Private Server) services.

[0164] To solve the problems of rapid power output and difficult power prediction during the ramp-up stage of a high-capacity-ratio power station, the short-term power prediction scheme for high-capacity-ratio photovoltaic power generation provided by the present disclosure is based on a deep learning method and combines the business background, and divides the power station dataset into different weather types by means of historical power output data and meteorological data. By extracting the ramp-up characteristics of the power station under different weather conditions, the learning and mining of the accurate correspondence relationship between the actual weather, time period, and power of the high-capacity-ratio photovoltaic power generation station trained by means of the deep learning idea are carried out, so that the accuracy of the short-term photovoltaic prediction result is higher, and the prediction error during the ramp-up stage of the power output of the high-capacity-ratio power station is effectively reduced.

[0165] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and no limitations are imposed herein.

[0166] The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub - combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A high-capacity ratio photovoltaic short-term power prediction method, characterized in that Including: For sunny days, obtain the historical actual power and corresponding timestamps of a high-capacity-ratio photovoltaic power station, and perform normalization processing on the historical actual power and the timestamps respectively to obtain normalized data. Specifically including: Convert the timestamps with a preset duration as the time resolution into numerical features; Perform sine-cosine processing on the numerical features to obtain normalized time data; Perform normalization processing on the historical actual power to obtain normalized power data; Use a convolutional neural network with convolutional layers and pooling layers of different sizes to extract features at different time scales from the normalized data to obtain local features; Use the attention mechanism to weight the local features processed by the long short-term memory network to obtain the target features represented in the form of tensors and the predicted power output by the long short-term memory network, and construct the rising rate feature and the time interval feature respectively according to the predicted power. Specifically including: Calculate the difference of the predicted power, and divide the difference calculation result by the time interval to obtain the rising rate feature representing the power change speed; Calculate the difference between the predicted powers at adjacent time steps to obtain the time interval feature representing the power ramp-up speed; Obtain the historical weather forecast data corresponding to the time period of the historical actual power, and after standardizing the historical weather forecast data, superimpose it with the target feature, the rising rate feature, and the time interval feature, and construct the first training sample with the historical actual power of the corresponding time period as the target truth value; Predict the actual power based on the network framework of the long short-term memory network for the first training sample, and train to obtain a sunny-day power prediction model; Predict the power of the target time period within a preset duration from the current based on the sunny-day power prediction model to obtain a prediction result.

2. The method according to claim 1, wherein The high-capacity-ratio photovoltaic power station refers to a photovoltaic power station in which the ratio of the installed capacity of the photovoltaic modules to the rated capacity of the inverter is higher than a preset threshold.

3. The method according to claim 1 or 2, characterized in that, Also including: For cloudy days, obtain the historical meteorological data and historical actual power of the high-capacity-ratio photovoltaic power station, calculate the clarity index and the direct ratio according to the historical meteorological data and the historical actual power, and calculate the correlation coefficient corresponding to the historical actual power after standardizing the clarity index and the direct ratio; Construct a clustering index as the classification index of the clustering algorithm from the three meteorological variables with the highest correlation coefficient calculated with the historical actual power, and divide the weather types into three categories according to the classification index; Remove the meteorological variables with low or no correlation according to the correlation between the meteorological variables and the actual output of the power generation equipment in the high-capacity-ratio photovoltaic power station under different weather types to obtain high-correlation meteorological variables; Construct second training samples for the high-correlation meteorological variables and the corresponding historical actual power under different weather types respectively, and train the second training samples under different weather types based on the network framework of the long short-term memory network respectively to obtain a cloudy-day power prediction model; Correspondingly, predicting the power of a target time period within a preset duration from the current time based on the sunny-day power prediction model to obtain a prediction result includes: Predicting the power of the sunny-day period in the target time period according to the sunny-day power prediction model to obtain a sunny-day period prediction result; Predicting the power of the cloud-day period under the corresponding weather type in the target time period according to the cloud-day power prediction model to obtain a cloud-day period prediction result; Summarizing the prediction result based on the sunny-day period prediction result and the cloud-day period prediction result.

4. The method according to claim 3, characterized in that, The historical meteorological data includes: At least one of predicted global radiation, diffuse radiation, direct radiation, total cloud cover, wind speed, temperature, relative humidity, precipitation, air pressure, and visibility.

5. The method according to claim 4, wherein It further includes: Obtaining the historical predicted global radiation and historical predicted total cloud cover of the high-capacity ratio photovoltaic power station; Constructing a weather classification comprehensive index according to the historical predicted global radiation, the historical predicted total cloud cover, and the corresponding weighting coefficients; Classifying the weather at the moments in the target time period that exceed a preset threshold as the cloud-day; Classifying the weather at the moments in the target time period that do not exceed the preset threshold as the sunny-day.

6. A high-capacity ratio photovoltaic short-term power prediction device, characterized in that, It includes: A normalization processing unit configured to, for sunny days, obtain the historical actual power and the corresponding timestamps of the high-capacity ratio photovoltaic power station, and perform normalization processing on the historical actual power and the timestamps respectively to obtain normalized data. It is also used to convert the timestamps with a preset duration as the time resolution into numerical features; perform sine-cosine processing on the numerical features to obtain normalized time data; perform normalization processing on the historical actual power to obtain normalized power data; A local feature extraction unit configured to extract features on different time scales in the normalized data using a convolutional neural network with convolutional layers and pooling layers of different sizes to obtain local features; A feature construction unit configured to weight the local features processed by a long short-term memory network using an attention mechanism to obtain a target feature represented in the form of a tensor and the predicted power output by the long short-term memory network, and respectively construct an ascending rate feature and a time interval feature according to the predicted power. It is also used to calculate the difference of the predicted power, and divide the difference calculation result by the time interval to obtain an ascending rate feature representing the power change speed; calculate the difference between the predicted powers at adjacent time steps to obtain a time interval feature representing the power ramp-up speed; A first training sample construction unit configured to obtain historical weather forecast data corresponding to the time period of the historical actual power, and superimpose the historical weather forecast data after standardization processing with the target feature, the ascending rate feature, and the time interval feature, and construct a first training sample with the historical actual power of the corresponding time period as the target truth value; A sunny-day power prediction model construction unit configured to predict the actual power based on the network framework of the long short-term memory network for the first training sample, and train to obtain a sunny-day power prediction model; A prediction unit, configured to predict the power of a target period within a preset duration from the current time based on the sunny-day power prediction model, and obtain a prediction result.

7. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the high-capacity ratio photovoltaic short-term power prediction method according to any one of claims 1-5.

8. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the high-capacity ratio photovoltaic short-term power prediction method according to any one of claims 1-5.

Images