A method and system for predicting regional power generation capacity of a multi-stage power station

Abstract

The present application discloses a method and system for predicting regional power generation capacity of a multi-level power station, which involves the operation and maintenance of a hydropower station, including: determining the historical coverage area images of the reservoir areas of each level of power stations based on remote sensing images, and obtaining the actual power generation data of each historically defined time interval for the top-level power station, training the reservoir area prediction model, and fitting the power generation curve of the top-level power station; based on the trained reservoir area prediction model and weather forecast information, determining the reservoir area information of the prediction time, and matching the utilization hours of each generator of the top-level power station within the prediction time in segments; for the subsequent power stations, determining the corresponding water consumption according to the utilization hours of each generator of the previous power station, predicting the utilization hours of each generator of the subsequent power station, and then predicting the regional peak power generation capacity of the multi-level power station within the prediction time. The method of the present application is used to carry out regional power generation capacity prediction and solve the problem of difficulty in prediction due to insufficient power generation data of multi-level hydropower stations.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station operation and maintenance, and in particular to a method and system for predicting regional power generation capacity of a multi-stage power station. Background Art

[0002] The power generation capacity of small and medium-sized hydropower refers to the power generation potential of small and medium-sized hydropower under certain meteorological and hydrological conditions. The power generation capacity of small and medium-sized hydropower is related to many factors such as meteorology, hydrology, installed capacity, etc., and its power generation capacity prediction faces many difficulties.

[0003] Most small and medium-sized hydropower stations are run-of-river power stations with poor regulation capabilities. Power generation is largely determined by incoming water. In addition, the installed capacity of a single station is small, and power generation has strong uncertainty and volatility, making it difficult to predict.

[0004] Small and medium-sized hydropower stations also lack historical hydrological, meteorological, and power generation capacity observation and monitoring data for prediction, and the data accumulation is insufficient. This is mainly due to the fact that most small and medium-sized hydropower stations have weak management, difficult information collection, and lack of relevant hydrological stations and meteorological data accumulation. Small and medium-sized hydropower stations are widespread and multi-point, and the prediction of power generation capacity of individual power stations is not regular. Predicting one by one cannot obtain the accuracy that meets actual requirements. Summary of the invention

[0005] The embodiment of the present application provides a method and system for predicting regional power generation capacity of a multi-stage power station, which combines historical data of regional power stations to carry out regional power generation capacity prediction, and solves the problem of difficulty in prediction due to insufficient power generation data of small and medium-sized multi-stage hydropower stations.

[0006] The present application embodiment provides a method for predicting regional power generation capacity of a multi-level power station, comprising the following steps:

[0007] According to the sequence of the watershed locations of the multi-level power stations, the runoff information of any level of power station is obtained;

[0008] Determine the historical coverage area images of the reservoir areas of each level of power stations based on remote sensing images, and obtain the actual power generation data of each historically defined time interval for the top-level power station;

[0009] Based on the historical coverage area image of the reservoir area of ​​the top-level power station, the power generation data of each designated time interval, and the corresponding weather information of each designated time interval, the reservoir area prediction model is trained;

[0010] According to the weather information, the acquired historical coverage area image of the reservoir area is classified and fitted with the change relationship curve between the designated time interval, and the power generation curve of the uppermost power station is fitted in each designated time interval;

[0011] Estimate the peak power demand during the forecast period based on the dispatch data of the same period, and determine the reservoir area information during the forecast period based on the trained reservoir forecast model and weather forecast information;

[0012] According to the determined reservoir area information, weather forecast information and the fitted change relationship curve, the utilization hours of each generator of the uppermost power station within the forecast period are matched in sections;

[0013] For the downstream power station, the corresponding water consumption is determined according to the utilization hours of each generator of the upper power station as the first water inflow of the downstream power station, and the second water inflow of the runoff is obtained based on the runoff information;

[0014] Predicting the utilization hours of each generator of the subsequent power station according to the subsequent power station, the current reservoir coverage area image determined based on the remote sensing image, and the first water inflow and the second water inflow;

[0015] According to the utilization hours of generators at each level of power stations, the regional peak power generation capacity of multiple power stations within the predicted time period is predicted.

[0016] Optionally, based on the historical coverage area image of the reservoir area of ​​the uppermost power station, the power generation data of each designated time interval, and the weather information corresponding to each designated time interval, the training reservoir area prediction model includes:

[0017] According to the historical coverage area image, determine the coverage area mean of the historical coverage area image, and configure a fixed value area region based on the coverage area mean, wherein the fixed value area region is configured with a reference point included in all historical coverage area images, and the fixed value area region is 75%-90% of the coverage area mean;

[0018] Determine the reference point position, align the reference point position with the fixed value area region, and remove pixels of the fixed value area region;

[0019] Determine the reservoir boundary area of ​​any historical coverage area image under the remote sensing image;

[0020] The image features of the removed area image are extracted using a convolutional network;

[0021] The extracted image features are fused with the corresponding power generation trend features in chronological order and input into the LSTM model. The output of the LSTM model is decoded by a Decoder to train a reservoir area prediction model.

[0022] Optionally, determining the reservoir boundary area of ​​any historical coverage area image under the remote sensing image includes:

[0023] Determine a first pixel range of the water area based on any one of the historical coverage area images;

[0024] In the first pixel range, a second pixel range is set, wherein the second pixel range is larger than and covers the first pixel range;

[0025] According to the second pixel range, continuously identifying pixels of the water area to determine a possible water area;

[0026] Determining a third pixel range based on the first pixel distance outside the determined possible water area;

[0027] The pixel values ​​of the third pixel range are eliminated from the possible water area, and the obtained boundary range is used as the reservoir boundary area.

[0028] Optionally, the extracted image features are fused with corresponding power generation trend features in chronological order, input into an LSTM model, and the output of the LSTM model is decoded by a Decoder to train a reservoir area prediction model, including:

[0029] The extracted image features are used to construct an initial sequence in sequence within a set time window; and

[0030] For any image feature within the time window, extract the power generation curve feature of the previous time window length;

[0031] fusing the extracted power generation curve features with any of the image features to construct a training sequence;

[0032] Time and weather tags are added to the training sequence to construct training samples, and the constructed training samples are used to train the reservoir area prediction model.

[0033] Optionally, using the constructed training samples to train the reservoir prediction model includes:

[0034] Determining the predicted reservoir area boundary range for the reservoir area prediction image at a later time output by the reservoir area prediction model; and,

[0035] Obtaining the real reservoir area image at a later time described in the historical coverage area image, and determining the boundary range of the real reservoir area;

[0036] In the predicted reservoir area boundary range and the actual reservoir area boundary range, a portion of representative pixels are respectively selected within a specified pixel distance to construct a representative pixel range;

[0037] The predicted reservoir area boundary range and the real reservoir area boundary range identified are completed using the representative pixel range, so that the boundaries are continuous after completion;

[0038] For the completed boundary, the mismatch between the predicted reservoir boundary and the actual reservoir boundary is calculated according to the set pixel width;

[0039] The calculated boundary mismatch is used as a loss to train the reservoir area prediction model.

[0040] Optionally, according to the determined reservoir area information, weather forecast information and the fitted change relationship curve, the utilization hours of each generator of the uppermost power station within the forecast period are matched in sections, including:

[0041] According to the weather forecast information within a preset time period, fuzzy weather matching is performed on the change relationship curve fitted according to the weather classification, so that the fuzzily matched fitted change relationship curve and the rainfall intensity within the predicted time period are similar;

[0042] According to the determined reservoir area information, the various change relationship curves obtained by fuzzy matching are matched to determine the utilization hours of each generator of the adapted uppermost power station.

[0043] Optionally, according to the determined reservoir area information, the change relationship curves obtained by fuzzy matching are matched to determine the utilization hours of each generator of the uppermost power station to be adapted, including:

[0044] For each change relationship curve, calculate the similarity between the corresponding reservoir area change trend and the predicted reservoir area information;

[0045] Sort by similarity, and determine a plurality of change relationship curves whose similarity is greater than a preset threshold;

[0046] The utilization hours of each generator corresponding to the determined variation relationship curve;

[0047] The peak power generation capacity of the uppermost power station is adapted according to the maximum value of the total power generation capacity of the utilization hours of each generator.

[0048] Optionally, when any one of the power stations opens a flood control gate, based on the power station at the next stage, the current reservoir coverage image determined based on the remote sensing image, and the first water inflow and the second water inflow, predicting the utilization hours of each generator of the power station at the next stage further includes:

[0049] Acquiring the opening data of the flood control gate, and determining the third water volume for discharging flood water to the lower reservoir area based on the opening data of the flood control gate;

[0050] For the downstream power station, the utilization hours of each generator of the downstream power station are predicted based on the current reservoir coverage area image determined by the remote sensing image, and the first water inflow, the second water inflow and the third water inflow.

[0051] An embodiment of the present application also proposes a regional power generation capacity prediction system for a multi-stage power station, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps of the regional power generation capacity prediction method for the multi-stage power station as described above are implemented.

[0052] The method of the embodiment of the present application combines the historical data of the regional power station and makes a comprehensive prediction of the storage area of ​​the reservoir area, so as to estimate the regional power generation capacity according to the prediction results, and solve the problem of difficulty in prediction due to insufficient power generation data of small and medium-sized multi-stage hydropower stations.

[0053] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0055] Figure 1 The basic process diagram of the regional power generation capacity prediction method for multi-stage power stations is shown in FIG. DETAILED DESCRIPTION

[0056] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0057] The present application embodiment proposes a method for predicting regional power generation capacity of a multi-stage power station, such as Figure 1 As shown, the following steps are included:

[0058] In step S101, the runoff information of any power station is obtained according to the sequence of the watershed positions of the multi-level power stations. In some examples, the runoff information of each level of power station can be determined based on the actual terrain relationship, map, geographic model, etc.

[0059] In step S102, the historical coverage area images of the reservoir areas of power stations at all levels are determined based on the remote sensing images, and for the power station at the top level, the actual power generation data of each historically defined time interval is obtained. In some examples, the length of each historically defined time interval obtained can be set according to actual needs, such as one week, half a month, three weeks, etc., or other mixed time lengths, which can be set according to actual needs. The remote sensing image can also be a panoramic image of the reservoir area taken by a drone at a specified height at a specified location.

[0060] The embodiment of the present application is related to the generation capacity of small and medium-sized hydropower stations and many factors such as meteorology, hydrology, and installed capacity. The resulting generation capacity prediction faces many difficulties. In step S103, the present application trains a reservoir prediction model based on the historical coverage area image of the reservoir area of ​​the uppermost power station, the generation data of each demarcated time interval, and the weather information corresponding to each demarcated time interval. In some specific examples, the embodiment of the present application considers the main power generation conditions, that is, the reservoir capacity. The present application reflects the reservoir capacity by predicting the reservoir area, and combines the water conditions corresponding to the weather (such as rain) and the water conditions corresponding to power generation, so as to comprehensively train a reservoir prediction model for the upper power station of the multi-level power station. The reservoir capacity of the lower power station is greatly affected by the power generation and drainage of the upper power station, that is, by accurately predicting the changes in the reservoir area of ​​the uppermost power station, accurate predictions can be achieved for the lower power station.

[0061] In step S104, according to the weather information, the change relationship curve between the historical coverage area image of the reservoir area obtained and the demarcated time interval is classified and fitted, and the power generation curve of the top power station is fitted in each demarcated time interval. In some examples, according to the weather information or the different rainfall intensities corresponding to the weather information, the change relationship between the historical coverage area image of the reservoir area obtained and the demarcated time interval is classified and fitted. By fitting the curve and combining the area change, the power generation capacity of the power station can be predicted and judged. For details, please refer to the subsequent embodiments.

[0062] In step S105, the peak power demand in the later prediction time is estimated based on the scheduling data of the same period, and the reservoir area information of the prediction time is determined based on the trained reservoir prediction model and weather forecast information. In some examples, the possible power generation trend of the current period can be estimated based on the scheduling data of the same period in previous years, so as to estimate the peak power demand in the later prediction time, and the reservoir area change information within the prediction time can be determined through the trained reservoir prediction model.

[0063] In step S106, the utilization hours of each generator of the top-level power station within the forecast period are matched in segments according to the determined reservoir area information, weather forecast information, and the fitted change relationship curve. In some examples, the basis for segmentation can be the rainfall corresponding to the weather forecast information, such as partitioning the rainfall, or partitioning the rainfall intensity, or in the case of no rainfall during the forecast period, the segmentation can be performed in a way of equal duration. Through segmented prediction, the accuracy of the prediction can be further improved. After segmentation, the utilization hours of the generators in each segment can be matched according to the aforementioned fitted curve relationship, so that the power generation capacity of the top-level power station during the forecast period can be determined.

[0064] In step S107, for the downstream power station, the corresponding water consumption is determined according to the utilization hours of each generator of the upper power station as the first water inflow of the downstream power station, and the second water inflow of the runoff is obtained based on the runoff information. After the prediction of the upper power station is completed, the corresponding water consumption can be determined according to the utilization hours of its generator, that is, the discharge to the lower power station, as the water inflow of the lower power station, thereby completing the power generation capacity of the lower power station.

[0065] In step S108, the utilization hours of each generator of the subsequent power station are predicted according to the subsequent power station, the current reservoir coverage area image determined based on the remote sensing image, and the first water inflow and the second water inflow.

[0066] In step S109, the regional peak power generation capacity of the multi-stage power station within the predicted time period is predicted based on the utilization hours of the generators of each stage of the power station.

[0067] The method of the embodiment of the present application combines the historical data of the regional power station and makes a comprehensive prediction of the storage area of ​​the reservoir area, so as to estimate the regional power generation capacity according to the prediction results, and solve the problem of difficulty in prediction due to insufficient power generation data of small and medium-sized multi-stage hydropower stations.

[0068] In some embodiments, based on the historical coverage area image of the reservoir area of ​​the uppermost power station, the power generation data of each designated time interval, and the weather information corresponding to each designated time interval, training the reservoir area prediction model includes:

[0069] According to the historical coverage area image, the mean coverage area of ​​the historical coverage area image is determined, and a fixed area region is configured based on the coverage area mean, wherein the fixed area region is configured with reference points contained in all historical coverage area images, and the fixed area region is 75%-90% of the coverage area mean. By determining a fixed area region to eliminate image areas that will not participate in the calculation, the amount of data processing in the future can be reduced and the data processing efficiency can be improved. On the other hand, the noise introduced into the image features can be reduced, so that the model focuses on learning the boundary change features of the reservoir area. Fixed objects that are easily identifiable in the coverage area image can be selected as reference points.

[0070] Determine the reference point position, then align the reference point position with the fixed value area, and remove pixels of the fixed value area. Set the fixed value area, and remove the part of the area covered by the reference point after alignment.

[0071] Determine the reservoir boundary area of ​​any historical coverage area image under the remote sensing image.

[0072] The convolutional network is used to extract the image features of the regional image within the removed reservoir boundary area, and the extracted image features are fused with the corresponding power generation trend features in chronological order and input into the LSTM model. The output of the LSTM model is decoded by Decoder to train the reservoir area prediction model.

[0073] In some embodiments, determining a reservoir boundary area of ​​any historical coverage area image under the remote sensing image includes:

[0074] Based on any of the historical coverage area images, the first pixel range of the water area is determined. In some examples, the first pixel range may be a pixel value at a specified pixel distance from a fixed area region, and multiple pixel values ​​in the circumferential direction at the specified pixel distance from the fixed area region are taken to construct the first pixel range.

[0075] In the first pixel range, a second pixel range is set, wherein the second pixel range is larger than and covers the first pixel range. The second pixel range is set to expand the first pixel range, thereby improving the recognition capability of the water surface boundary.

[0076] Based on the second pixel range, continuous pixel recognition is performed on the water area to determine the possible water area.

[0077] Based on the first pixel distance outside the determined possible water area, a third pixel range is determined. In some examples, the third pixel range is used to characterize the pixel interval of the ground or woodland near the water boundary. The pixel values ​​of the third pixel range are eliminated from the possible water area, and the obtained boundary range is used as the reservoir boundary area. Through two-level pixel recognition and pixel filtering, the accurate reservoir boundary area can be determined.

[0078] In some embodiments, the extracted image features are fused with corresponding power generation trend features in chronological order, input into the LSTM model, and the output of the LSTM model is decoded by a Decoder to train the reservoir area prediction model, including:

[0079] The extracted image features are used to construct an initial sequence in sequence within a set time window; and

[0080] For any image feature within the time window, extract the power generation curve feature of the previous time window length.

[0081] The extracted power generation curve features are fused with any of the image features to construct a training sequence. In the embodiment of the present application, water used for power generation is considered to be the main factor consuming the reservoir area, while runoff and rainfall are the main factors of water inflow to the reservoir. The prediction of water inflow can be implied by predicting the area in segmented time, and the power curve features are further fused with any of the image features, so that the water consumption factor of water used for power generation can be characterized in the features, thereby improving the prediction accuracy of the model.

[0082] Add time and weather tags to the training sequence to construct training samples, and use the constructed training samples to train the reservoir area prediction model. By adding time and weather tags, introducing runoff and rainfall as the main water inflow factors, the reservoir area prediction model is finally realized to comprehensively reflect the changes in the reservoir area caused by power generation.

[0083] In some embodiments, training the reservoir prediction model using the constructed training samples includes:

[0084] Determining the predicted reservoir area boundary range for the reservoir area prediction image at a later time output by the reservoir area prediction model; and,

[0085] The real reservoir area image at the later time in the historical coverage area image is obtained, and the boundary range of the real reservoir area is determined. The specific boundary range determination method can refer to the above embodiment.

[0086] In the predicted reservoir area boundary range and the actual reservoir area boundary range, a portion of representative pixels are respectively selected within the specified pixel distance to construct a representative pixel range, that is, representative pixels are determined for the actual reservoir area boundary and the predicted reservoir area boundary in the training sample respectively.

[0087] The predicted reservoir area boundary range and the real reservoir area boundary range identified by the representative pixel range are complemented so that the boundary has continuity after complementation. After complementation, the real reservoir area boundary and the predicted reservoir area boundary are guaranteed to have continuity to participate in subsequent calculations.

[0088] For the completed boundary, the mismatch between the predicted reservoir boundary and the actual reservoir boundary is calculated according to the set pixel width. The calculated boundary mismatch is used as a loss to train the reservoir prediction model. In some examples, in the embodiments of the present application, the boundary range is considered, so that the area within the boundary is the reservoir area or most of the reservoir area. The coverage reflects the changes in the reservoir due to power generation. By calculating the boundary mismatch in the training sample, the reservoir prediction model can learn a more accurate reservoir boundary, while other areas of the reservoir are not focused on and learned.

[0089] In some embodiments, according to the determined reservoir area information, weather forecast information and the fitted change relationship curve, segmentally matching the utilization hours of each generator of the uppermost power station within the forecast period includes:

[0090] According to the weather forecast information within the preset time period, the change relationship curve fitted according to the weather classification is fuzzy weather matched, so that the fuzzily matched fitted change relationship curve is similar to the rainfall intensity within the predicted time period. In a specific embodiment, the predicted time period can be further segmented or classified according to the aforementioned classification and the weather forecast information within the predicted time period, and after the classification, the change relationship curve fitted according to the weather classification is first fuzzy weather matched, so as to select the previously fitted change relationship curve with similar rainfall intensity within the segment.

[0091] According to the determined reservoir area information, the various change relationship curves obtained by fuzzy matching are matched to determine the utilization hours of each generator of the uppermost power station. By determining the above-mentioned fitted change relationship curves with similar rainfall intensity or rainfall, further matching is performed according to the reservoir area change information predicted by the model, thereby determining the utilization hours of each generator of the uppermost power station. In this way, it can be closer to the actual operating conditions of the power station and improve the prediction accuracy of power generation capacity.

[0092] In some embodiments, according to the determined reservoir area information, the change relationship curves obtained by fuzzy matching are matched to determine the utilization hours of each generator of the uppermost power station to be adapted, including:

[0093] For each change relationship curve, the similarity between the corresponding reservoir area change trend and the predicted reservoir area information is calculated. The specific similarity calculation method is not described here in detail, and reference may be made to the prior art.

[0094] The similarity is sorted to determine a plurality of change relationship curves whose similarity is greater than a preset threshold. For example, for change relationship curves whose similarity is greater than 90%, the next step of calculation is performed.

[0095] According to the utilization hours of each generator corresponding to the determined change relationship curve, that is, the utilization hours of the corresponding generator are determined from the change relationship curve, and the generator power generation condition is also matched and determined.

[0096] The peak power generation capacity of the uppermost power station is adapted according to the maximum value of the total power generation capacity of the utilization hours of each generator.

[0097] In some embodiments, when any one of the power stations opens the flood control gate, based on the power station at the next stage, the current reservoir coverage image determined based on the remote sensing image, and the first water inflow and the second water inflow, predicting the utilization hours of each generator of the power station at the next stage further includes:

[0098] Acquiring the opening data of the flood control gate, and determining the third water volume for discharging flood water to the lower reservoir area based on the opening data of the flood control gate;

[0099] For the downstream power station, the utilization hours of each generator of the downstream power station are predicted based on the current reservoir coverage area image determined by the remote sensing image, and the first water inflow, the second water inflow and the third water inflow.

[0100] For predicting the utilization hours of each generator in the downstream power station, you can refer to the method of the embodiment of the present application, further combine the opening status of the flood gate of the upstream power station, pay attention to several key water inflow factors, and then combine the current reservoir coverage area image to complete the power generation capacity prediction.

[0101] The method of the present application combines historical data of regional power stations, introduces water inflow factors such as runoff and rainfall, as well as water use factors such as power generation and flood discharge, and conducts a comprehensive forecast of the reservoir area, so as to estimate the regional power generation capacity based on the forecast results, solve the problem of difficult forecasting due to insufficient power generation data of small and medium-sized multi-stage hydropower stations, improve the accuracy of power generation capacity forecasting of small and medium-sized multi-stage hydropower stations, and provide an important reference for ensuring the full operation of the power grid.

[0102] An embodiment of the present application also proposes a regional power generation capacity prediction system for a multi-stage power station, wherein a computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps of the regional power generation capacity prediction method for the multi-stage power station as described above are implemented.

[0103] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present disclosure with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. It is not limited to the examples described in this specification or during the implementation of this application, and its examples will be interpreted as non-exclusive.

[0104] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) may be used in combination with each other. For example, a person of ordinary skill in the art may use other embodiments when reading the above description.

[0105] The above embodiments are merely exemplary embodiments of the present disclosure. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present disclosure, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A method for predicting regional power generation capacity of a multi-stage hydropower station, characterized in that: The steps include: According to the sequence of the basin locations of multi-level hydropower stations, the runoff information of any level of hydropower stations is obtained; Determine the historical coverage area images of the reservoirs of hydropower stations at all levels based on remote sensing images, and obtain the actual power generation data of each historically defined time interval for the top-level hydropower station; Training a reservoir prediction model based on the historical coverage area image of the reservoir area of ​​the uppermost hydropower station, the actual power generation data of each designated time interval, and the weather information corresponding to each designated time interval; According to the weather information, the acquired historical coverage area images of the reservoir areas of the hydropower stations at all levels are classified and fitted with the change relationship curves between the designated time intervals, and the power generation curves of the uppermost hydropower station are fitted in the designated time intervals; Estimate the peak power demand during the forecast period based on the dispatch data of the same period, and determine the reservoir area information during the forecast period based on the trained reservoir forecast model and weather forecast information; According to the determined reservoir area information, weather forecast information and the fitted change relationship curve, the utilization hours of each generator of the top-level hydropower station within the forecast period are matched in sections; For the downstream hydropower station, the corresponding water consumption is determined as the first water inflow of the downstream hydropower station according to the utilization hours of each generator of the upstream hydropower station, and the second water inflow of the runoff is obtained based on the runoff information; For the downstream hydropower station, based on the current reservoir coverage area image determined by the remote sensing image, the first water inflow and the second water inflow, predict the utilization hours of each generator of the downstream hydropower station; According to the utilization hours of each generator of each hydropower station, the regional peak power generation capacity of multiple hydropower stations within the forecast period is predicted; Wherein, based on the historical coverage area image of the reservoir area of ​​the uppermost hydropower station, the actual power generation data of each designated time interval and the weather information corresponding to each designated time interval, the training reservoir area prediction model includes: According to the historical coverage area image, determine the coverage area mean of the historical coverage area image, and configure a fixed value area region based on the coverage area mean, wherein the fixed value area region is configured with a reference point included in all historical coverage area images, and the area of ​​the fixed value area region is 75%-90% of the coverage area mean; Determine the position of the reference point, align the position of the reference point with the fixed value area, and remove pixels of the fixed value area; Determine the reservoir boundary area of ​​any historical coverage area image under the remote sensing image; The image features of the removed area image are extracted using a convolutional network; The extracted image features are fused with the corresponding power generation trend features in chronological order and input into the LSTM model. The output of the LSTM model is decoded by a Decoder to train a reservoir area prediction model.

2. The method for predicting regional power generation capacity of a multi-stage hydropower station according to claim 1, characterized in that: Determine the reservoir boundary area of ​​any historical coverage area image under remote sensing image including: Determine a first pixel range of the water area based on any one of the historical coverage area images; Setting a second pixel range according to the first pixel range, wherein the second pixel range is larger than and covers the first pixel range; According to the second pixel range, continuously identifying pixels of the water area to determine a possible water area; Determine a third pixel range within the first pixel distance outside the determined possible water area; The pixels in the third pixel range are eliminated from the possible water area, and the obtained boundary range is used as the reservoir boundary area.

3. The method for predicting regional power generation capacity of a multi-stage hydropower station according to claim 1, characterized in that: The extracted image features are fused with the corresponding power generation trend features in chronological order and input into the LSTM model. The output of the LSTM model is decoded by a Decoder to train the reservoir area prediction model, including: The extracted image features are used to construct an initial sequence in order within a set time window; For any image feature within the time window, extract the power generation curve feature of the previous time window length; fusing the extracted power generation curve features with any of the image features to construct a training sequence; Time and weather tags are added to the training sequence to construct training samples, and the constructed training samples are used to train the reservoir area prediction model.

4. The method for predicting regional power generation capacity of a multi-stage hydropower station as claimed in claim 3, characterized in that: The reservoir prediction model trained using the constructed training samples includes: Determine the predicted reservoir area boundary range for the reservoir area prediction image at a later time output by the reservoir area prediction model; Obtaining the real reservoir area image at a later time described in the historical coverage area image, and determining the boundary range of the real reservoir area; Selecting a portion of representative pixels within a specified pixel distance from the predicted reservoir area boundary range and the actual reservoir area boundary range to construct a representative pixel range; The predicted reservoir area boundary range and the real reservoir area boundary range are completed using the representative pixel range, so that the boundaries are continuous after completion; For the completed boundary, the mismatch between the predicted reservoir boundary range and the actual reservoir boundary range is calculated according to the set pixel width; The calculated mismatch degree is used as a loss to train the reservoir area prediction model.

5. The method for predicting regional power generation capacity of a multi-stage hydropower station as claimed in claim 1, characterized in that: According to the determined reservoir area information, weather forecast information and the fitted change relationship curve, the utilization hours of each generator of the top-level hydropower station within the forecast period are matched in sections, including: According to the weather forecast information within the forecast period, fuzzy weather matching is performed on the weather information corresponding to the change relationship curve classified and fitted according to the weather information, so that the rainfall intensity of each change relationship curve matched by the fuzzy weather is similar within the forecast period; According to the determined reservoir area information, the various change relationship curves obtained by fuzzy weather matching are matched to determine the utilization hours of each generator of the uppermost hydropower station.

6. The method for predicting regional power generation capacity of a multi-stage hydropower station as claimed in claim 5, characterized in that: According to the determined reservoir area information, the various change relationship curves obtained by fuzzy weather matching are matched to determine the utilization hours of each generator of the uppermost hydropower station, including: For each change relationship curve, calculate the similarity between the corresponding reservoir area change trend and the predicted reservoir area information; Sort by similarity, and determine a plurality of change relationship curves whose similarity is greater than a preset threshold; Determine the utilization hours of each corresponding generator according to the determined change relationship curve; According to the maximum total power generation capacity of each generator's utilization hours, the peak power generation capacity of the uppermost hydropower station is adapted.

7. The method for predicting regional power generation capacity of a multi-stage hydropower station according to claim 1, characterized in that: When any one of the first-stage hydropower stations opens the flood control gate, for the subsequent hydropower station, based on the current reservoir coverage image determined by the remote sensing image, the first water inflow, and the second water inflow, the predicted utilization hours of each generator of the subsequent hydropower station include: Acquiring the opening data of the flood control gate, and determining the third water volume for discharging flood water to the lower reservoir area based on the opening data of the flood control gate; For the downstream hydropower station, the utilization hours of each generator of the downstream hydropower station are predicted based on the current reservoir coverage area image determined by the remote sensing image, the first water inflow, the second water inflow and the third water inflow.

Images