A cascade optimization scheduling method and system for flood control power generation

Abstract

This application discloses a cascade optimization scheduling method and system for flood control and power generation, involving hydropower station scheduling and data processing technology, including: grouping historical storage data, discharge data, and water level data according to a set time period to obtain an initial sample; expanding the initial sample based on the distribution of water level data in the data group to obtain a training sample; constructing an LSTM-based scheduling optimization model; adding labels to each group of training sample data based on historical rainfall data to train the scheduling optimization model; obtaining current data of the cascade basin, and performing predictions based on the trained scheduling optimization model, and performing scheduling optimization according to the prediction results. The method of this application is used to give full play to the joint interception and storage function of cascade reservoirs, which can ensure flood control safety in the basin and improve power generation efficiency.

Description

Technical Field

[0001] The present application relates to the technical fields of hydropower station scheduling and data processing, and in particular to a cascade optimization scheduling method and system for flood control power generation. Background Art

[0002] Currently, flood control storage capacity at water conservancy hubs is designed to ensure flood safety for dams or downstream protected areas. This is primarily achieved through the independent operation of water conservancy hubs to protect against standard floods and ensure flood safety for specific protected areas. Domestic and international research has also largely focused on analyzing the operational modes of individual or cascade reservoirs for specific flood control tasks, with less quantitative research on the role of sub-regional cascade reservoir flood control operations in basin-wide flood prevention.

[0003] While existing technologies are attempting to develop integrated defense plans for basin-wide floods, the coordinated operation of clusters of water conservancy hubs remains to be refined, making it difficult to fully utilize the regulating and storage functions of regional controlled reservoirs. Existing technologies, when addressing basin-wide flood control, still rely on lowering reservoir operating levels to flood control limits and reserving planned flood control storage capacity. This approach fails to balance the comprehensive benefits of cascade reservoirs and prevents efficient use of basin water resources. In practical applications, existing technologies still require each water conservancy hub to first reach a pre-set flood control limit to prevent floods. This makes it difficult to fully utilize the regulating and storage functions of regional controlled reservoirs when addressing different types of basin-wide floods. Summary of the Invention

[0004] The embodiments of the present application provide a cascade optimization scheduling method and system for flood control and power generation, which is used to optimize the scheduling of cascade river basins by combining flood control and power generation, give full play to the joint interception and storage role of cascade reservoirs, ensure flood control safety in the river basin, and improve power generation efficiency.

[0005] The present application embodiment provides a cascade optimization scheduling method for flood control power generation, including:

[0006] Obtain historical inflow data, discharge data, and water level data for cascade river basins during flood season;

[0007] According to the set time period, historical storage data, discharge data and water level data are grouped to obtain the initial sample;

[0008] Determining a data group exceeding a warning water level from various types of data in the initial sample, and expanding the initial sample based on the distribution of water level data in the data group to obtain a training sample;

[0009] Constructing an LSTM-based scheduling optimization model, wherein the scheduling optimization model includes a three-way branch model, and each branch model includes an LSTM module, wherein each branch model is used to independently perform prediction based on storage data, drainage data, or water level data in a set of input data; and

[0010] An attention mechanism is set between the three-way branch models. The attention mechanism is used to compare the peak deviations of the feature information extracted by the corresponding branch models, and to assign weights to the extracted feature information according to the comparison results to splice the extracted feature information. The greater the peak deviation of the feature information extracted by any branch, the greater the weight assigned to the branch.

[0011] adding labels to each group of training sample data according to historical rainfall data to train the scheduling optimization model;

[0012] The current inflow data, outflow data and water level data of the cascade basin are obtained, and prediction is performed based on the trained scheduling optimization model to perform scheduling optimization according to the prediction results.

[0013] Optionally, historical inflow data, outflow data, and water level data can be grouped according to a set time period, including:

[0014] Set the inflow limit, water level deviation limit and discharge limit of the inflow data respectively;

[0015] In any period of time, when one of the inflow, water level deviation and discharge reaches the corresponding limit, the other data are intercepted based on the same time and grouped as a group of data.

[0016] Optionally, determining a data group exceeding a warning water level from various types of data in the initial sample, and expanding the initial sample based on the distribution of water level data in the data group includes:

[0017] Detecting the water level distribution in the data group containing data exceeding the warning water level, and eliminating the data group in which the proportion of the time when the water level data exceeds the warning water level is greater than a preset threshold;

[0018] According to the detected water level distribution, the drainage interval and the maximum inflow of the cascade basin are used as constraints to generate data points for the inflow data and drainage data in the remaining data groups;

[0019] Fit the water level at the corresponding moment according to the generated inflow data and outflow data to obtain an extended sample;

[0020] Repeat the expansion until the proportion of time that the water level data in the remaining groups exceeds the warning water level is greater than the preset threshold, or the number of samples reaches the preset number.

[0021] Optionally, generating data points for the incoming data and outgoing data in the remaining groups includes:

[0022] Select inflow data points or discharge data points from the data groups whose time proportion is greater than a preset threshold, wherein the larger the single point value of the inflow data point or the discharge data point, the greater the probability of selection;

[0023] Based on the selected inflow data points or discharge data points, discharge data points or inflow data points are generated according to corresponding constraints.

[0024] Optionally, adding labels to each group of training sample data based on historical rainfall data includes:

[0025] Obtain historical rainfall data of the main stream and tributaries of the cascade watershed;

[0026] Divide the main stream and its tributaries into sections according to their geographical locations;

[0027] According to the segment relationship, the time delay of the historical rainfall data in the corresponding segment is configured. The farther the distance of any tributary is from the current basin, the longer the time delay;

[0028] According to the historical rainfall data of each segment and the time delay of the segment, each group data of the training sample is marked.

[0029] Optionally, the constructed LSTM-based scheduling optimization model also includes a fully connected layer set after the three-way branch model and the attention mechanism, wherein the fully connected layer has an output node number of 3, which is used to combine and output the predicted data group results.

[0030] Optionally, performing prediction based on the trained scheduling optimization model to perform scheduling optimization according to the prediction result includes:

[0031] Obtaining the most recent prediction result and calculating the deviation relationship between the prediction result and the current inflow data, outflow data, and water level data;

[0032] According to the deviation relationship, mapping is performed based on a preset scheduling relationship to obtain a scheduling direction;

[0033] According to the scheduling direction, combined with the current power generation information and gate opening information, optimized scheduling is performed.

[0034] Optionally, the scheduling direction includes increasing the drainage volume, maintaining the drainage volume, and reducing the drainage volume;

[0035] According to the scheduling direction, combined with the current power generation information and gate opening information, performing optimized scheduling includes:

[0036] Multiple sets of gate and / or power generation scheduling data are generated based on the deviation relationship, and a data set closest to the current power generation data is determined from the multiple sets of gate and / or power generation scheduling data as recommended scheduling data.

[0037] An embodiment of the present application also proposes a cascade optimization scheduling system for flood control power generation, 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 cascade optimization scheduling method for flood control power generation as described above are implemented.

[0038] The embodiment of the present application combines flood control and power generation to optimize the scheduling of cascade river basins, giving full play to the joint interception and storage function of cascade reservoirs, and can ensure the flood control safety of the river basin and improve the overall power generation efficiency of the cascade river basin.

[0039] 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

[0040] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0041] Figure 1 Schematic diagram of the basic process of the cascade optimization scheduling method for flood control power generation in this embodiment. DETAILED DESCRIPTION

[0042] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although 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. Rather, 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.

[0043] The embodiment of the present application proposes a cascade optimization scheduling method for flood control power generation, such as Figure 1 As shown, the following steps are included:

[0044] In step S101, historical inflow data, drainage data, and water level data for the cascade basin during the flood season are obtained. In some embodiments, the inflow data refers to the inflow of the basin at the current level. In specific examples, this data can be determined by combining drainage data from tributaries and the upper basin. The drainage data in the embodiments of the present application includes water used for power generation and flood gate discharge data. In specific examples, a relationship between the flood gate opening and the discharge volume can be pre-established to determine the discharge volume data. The corresponding guide vane opening can be found based on the power generation volume to determine the power generation drainage volume, and the drainage data can be accumulated and calculated. The water level data can be measured, for example, by a water level sensor.

[0045] In step S102, historical storage data, drainage data, and water level data are grouped according to a set time period to obtain an initial sample.

[0046] In step S103, data groups exceeding the warning water level are identified from the various types of data in the initial sample, and the initial sample is expanded based on the distribution of water level data in the data group to obtain training samples. In a specific example, using data groups exceeding the warning water level in the initial sample for expansion can increase the number of samples during the flood season, better simulate the flood conditions that may occur during the flood season, and improve the expected effect of optimized scheduling.

[0047] In step S104, a scheduling optimization model based on LSTM is constructed, wherein the scheduling optimization model includes a three-way branch model, and any branch model includes an LSTM module, wherein any branch model is used to perform prediction separately based on the incoming data, drainage data or water level data in a set of input data. In a specific example, the embodiment of the present application sets an attention mechanism between the three-way branch models. The attention mechanism is used to compare the peak deviations of the feature information extracted by the corresponding branch models, and to assign weights to the extracted feature information according to the comparison results to splice the extracted feature information. The greater the peak deviation of the feature information extracted by any branch, the greater the weight assigned to any branch. The peak deviation referred to in the embodiment of the present application refers to the deviation between the maximum values ​​in the data group. For example, in a 6-hour long data group, the peak (maximum and minimum) deviation of the segmented incoming data is greater than the incoming threshold, while other deviations are within the normal range. It is considered that the current incoming data belongs to the main factor affecting flood control scheduling, so a higher weight is assigned. The scheduling optimization model of the three-way branch LSTM of the embodiment of the present application is jointly trained and outputs the scheduling optimization results. In some examples, the prediction can be performed step by step according to the interception and storage effect of the cascade power station at the predicted moment. In some embodiments, the constructed LSTM-based scheduling optimization model also includes a fully connected layer set after the three-way branch model and the attention mechanism, wherein the fully connected layer has an output node number of 3, which is used to combine and output the predicted data group results.

[0048] In the specific example, this application makes full use of the characteristics of LSTM to predict the incoming data, drainage data and water level data respectively, and introduces changing weights to highlight the proportion of a certain type of data in the current data group to reflect the dominant data in the data. For example, when the incoming data is high, the incoming data has a greater weight, which is convenient for outputting the final prediction results based on the predictions of each LSTM, thereby improving the practical application effect of the model.

[0049] In step S105, each group of training sample data is labeled according to historical rainfall data to train the scheduling optimization model.

[0050] In step S106, the current inflow data, outflow data and water level data of the cascade basin are obtained, and prediction is performed based on the trained scheduling optimization model to perform scheduling optimization according to the prediction results.

[0051] The embodiment of the present application combines flood control and power generation to optimize the scheduling of cascade river basins, giving full play to the joint interception and storage function of cascade reservoirs, and can ensure the flood control safety of the river basin and improve the overall power generation efficiency of the cascade river basin.

[0052] In some embodiments, grouping historical storage data, drainage data, and water level data according to a set time period includes:

[0053] Set the inflow limit, water level deviation limit and discharge limit of the inflow data respectively;

[0054] In any period of time, when one of the inflow, water level deviation and discharge reaches the corresponding limit, the other data are intercepted based on the same time and grouped as a group of data.

[0055] That is, in the embodiment of the present application, the time intervals of any data group may be different. For example, the upper limit of the inflow flow, the upper limit of the water level deviation, and the limit of the drainage rate may be set, so that when any of the values ​​reaches, for example, the upper limit of the inflow flow in a certain period of time, it is intercepted as a data group.

[0056] In some embodiments, determining a data group exceeding a warning water level from various types of data in the initial sample, and expanding the initial sample based on the distribution of water level data in the data group includes:

[0057] The water level distribution in the data groups containing data exceeding the warning water level is detected, and data groups in which the proportion of time the water level data exceeds the warning water level is greater than a preset threshold are eliminated. In some embodiments, for example, if the water level data in a three-hour data group continuously exceeds the warning water level for 70% of the time, this data group is considered to have no room for expansion and is eliminated in this example. Sample expansion is then performed based on the remaining data groups. This approach can greatly improve the adaptability and rationality of the expanded sample to the local watershed and greatly avoid data distortion.

[0058] According to the detected water level distribution, data points are generated for the inflow data and outflow data in the remaining data groups, with the discharge range and maximum inflow of the cascade basin as constraints.

[0059] The water level at the corresponding moment is fitted according to the generated inflow data and outflow data to obtain an extended sample, that is, the new water level data is obtained by fitting the replaced inflow data and outflow data to improve the extended data set.

[0060] Repeat the expansion until the proportion of time that the water level data in the remaining groups exceeds the warning water level is greater than the preset threshold, or the number of samples reaches the preset number.

[0061] In a specific example, generating data points for the incoming data and outgoing data in the remaining groups includes:

[0062] An inflow data point or a discharge data point is selected from the data group whose time proportion is greater than a preset threshold, wherein the larger the single point value of the inflow data point or the discharge data point is, the greater the probability of selection is.

[0063] Based on the selected inflow data points or discharge data points, discharge data points or inflow data points are generated according to corresponding constraints. In some examples, this application utilizes the fact that data groups with a time percentage greater than a preset threshold are excluded because they contain more flood information. By selecting data points and generating discharge data points or inflow data points according to corresponding constraints, the authenticity of the expanded sample can be further improved.

[0064] In some embodiments, adding labels to each group of training sample data according to historical rainfall data includes:

[0065] Obtain historical rainfall data of the main stream and tributaries of the cascade watershed;

[0066] The main stream and its tributaries are segmented based on their geographical location. For example, in some examples, some cascade basins cover a large area or are long, so they can be divided into multiple segments.

[0067] Based on the segment relationships, configure a time delay for historical rainfall data within the corresponding segment. The farther a tributary is from the current watershed, the longer the time delay. By configuring a time delay, rainfall inflows can be reflected as close to the time period that directly affects water levels as possible.

[0068] Based on the historical rainfall data of each segment and the time delay of the segment, each group data of the training sample is marked. For example, if there is heavy rainfall in a certain area and the corresponding time delay is 1 hour, the flow caused by the corresponding heavy rainfall will be added to the inflow flow 1 hour later.

[0069] In some embodiments, performing prediction based on the trained scheduling optimization model to perform scheduling optimization according to the prediction result includes:

[0070] Obtain the prediction result at the most recent moment and calculate the deviation relationship between it and the current incoming data, drainage data and water level data. For example, the deviation relationship may include the positive and negative deviations of the incoming data, the positive and negative deviations of the drainage data and the positive and negative deviations of the water level data in the prediction result.

[0071] Based on the deviation relationship, a mapping is performed based on a preset scheduling relationship to obtain a scheduling direction. Since the scheduling method is essentially to control power generation and flood gates, that is, to control the discharge volume, in some examples, the scheduling direction includes increasing the discharge volume, maintaining the discharge volume, and decreasing the discharge volume.

[0072] According to the scheduling direction, combined with the current power generation information and gate opening information, optimized scheduling is performed.

[0073] In some embodiments, according to the scheduling direction, combined with current power generation information and gate opening information, performing optimized scheduling includes:

[0074] Based on the deviation relationship, multiple sets of gate and / or power generation scheduling data are generated. The data set that most closely matches the current power generation data is then identified from these multiple sets of gate and / or power generation scheduling data as the recommended scheduling data. In some specific examples, flood control gate scheduling data can be further recommended to the user based on the current power generation plan, thereby achieving optimized scheduling of flood control and power generation during the flood season.

[0075] An embodiment of the present application also proposes a cascade optimization scheduling system for flood control power generation, 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 cascade optimization scheduling method for flood control power generation as described above are implemented.

[0076] Furthermore, 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., solutions that intersect various embodiments), adaptations, or changes. The present invention is not limited to the examples described in this specification or during the practice of this application, which examples are to be construed as non-exclusive.

[0077] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art may use other embodiments when reading the above description.

[0078] 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 should also be deemed to fall within the protection scope of the present invention.

Claims

1. A cascade optimization scheduling method for flood control power generation, characterized in that: include: Obtain historical inflow data, discharge data, and water level data for cascade river basins during flood season; According to the set time period, historical storage data, discharge data and water level data are grouped to obtain the initial sample; Determining a data group exceeding a warning water level from various types of data in the initial sample, and expanding the initial sample based on the distribution of water level data in the data group to obtain a training sample; Constructing an LSTM-based scheduling optimization model, wherein the scheduling optimization model includes a three-way branch model, and each branch model includes an LSTM module, wherein each branch model is used to independently perform prediction based on storage data, drainage data, or water level data in a set of input data; as well as, An attention mechanism is set between the three-way branch models. The attention mechanism is used to compare the peak deviations of the feature information extracted by the corresponding branch models, and to assign weights to the extracted feature information according to the comparison results to splice the extracted feature information. The greater the peak deviation of the feature information extracted by any branch, the greater the weight assigned to the branch. adding labels to each group of training sample data according to historical rainfall data to train the scheduling optimization model; Obtaining current inflow data, outflow data, and water level data of the cascade basin, and performing prediction based on the trained scheduling optimization model, so as to perform scheduling optimization according to the prediction results; Determining a data group exceeding a warning water level from various types of data in the initial sample, and expanding the initial sample based on the distribution of water level data in the data group includes: Detecting the water level distribution in the data group containing data exceeding the warning water level, and eliminating the data group in which the proportion of the time when the water level data exceeds the warning water level is greater than a preset threshold; According to the detected water level distribution, the drainage interval and the maximum inflow of the cascade basin are used as constraints to generate data points for the inflow data and drainage data in the remaining data groups; Fit the water level at the corresponding moment according to the generated inflow data and outflow data to obtain an extended sample; Repeat the expansion until the proportion of time that the water level data in the remaining groups exceeds the warning water level is greater than the preset threshold, or the number of samples reaches the preset number; Generating data points for the incoming data and outgoing data in the remaining groups includes: Select inflow data points or discharge data points from the data groups whose time proportion is greater than a preset threshold, wherein the larger the single point value of the inflow data point or the discharge data point, the greater the probability of selection; Based on the selected inflow data points or discharge data points, discharge data points or inflow data points are generated according to corresponding constraints.

2. The cascade optimization scheduling method for flood control power generation according to claim 1, characterized in that: According to the set time period, the historical storage data, drainage data and water level data are grouped as follows: Set the inflow limit, water level deviation limit and discharge limit of the inflow data respectively; In any period of time, when one of the inflow, water level deviation and discharge reaches the corresponding limit, the other data are intercepted based on the same time and grouped as a group of data.

3. The cascade optimization scheduling method for flood control power generation according to claim 1, characterized in that: Adding labels to each group of training sample data based on historical rainfall data includes: Obtain historical rainfall data of the main stream and tributaries of the cascade watershed; Divide the main stream and its tributaries into sections according to their geographical locations; According to the segment relationship, the time delay of the historical rainfall data in the corresponding segment is configured. The farther the distance of any tributary is from the current basin, the longer the time delay; According to the historical rainfall data of each segment and the time delay of the segment, each group data of the training sample is marked.

4. The cascade optimization scheduling method for flood control power generation according to claim 1, characterized in that: The constructed LSTM-based scheduling optimization model also includes a fully connected layer set after the three-way branch model and the attention mechanism, wherein the fully connected layer has an output node number of 3, which is used to combine and output the predicted data group results.

5. The cascade optimization scheduling method for flood control power generation according to claim 4, characterized in that: Performing prediction based on the trained scheduling optimization model to perform scheduling optimization according to the prediction result includes: Obtaining the most recent prediction result and calculating the deviation relationship between the prediction result and the current inflow data, outflow data, and water level data; According to the deviation relationship, mapping is performed based on a preset scheduling relationship to obtain a scheduling direction; According to the scheduling direction, combined with the current power generation information and gate opening information, optimized scheduling is performed.

6. The cascade optimization scheduling method for flood control power generation according to claim 5, characterized in that: The dispatching direction includes increasing the displacement, maintaining the displacement and reducing the displacement; According to the scheduling direction, combined with the current power generation information and gate opening information, performing optimized scheduling includes: Multiple sets of gate and / or power generation scheduling data are generated based on the deviation relationship, and a data set closest to the current power generation data is determined from the multiple sets of gate and / or power generation scheduling data as recommended scheduling data.

7. A cascade optimization scheduling system for flood control power generation, characterized in that: A computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the cascade optimization scheduling method for flood control power generation according to any one of claims 1 to 6 are implemented.

Images