Small hydropower station hydrological data real-time prediction and unit start-stop control method

Abstract

The invention relates to the technical field of intelligent water conservancy, in particular to a small hydropower station hydrological data real-time prediction and unit start-stop control method. The method comprises the following steps: collecting hydrological data of the small hydropower station, carrying out abnormal value detection and filtering processing, and outputting basic hydrological data; acquiring a historical hydrological record; constructing a forebay water level dynamic prediction model according to the historical hydrological record and the basic hydrological data, and generating a forebay water level prediction result; performing water level parameter correction on the forebay water level prediction result by using the basic hydrological data, and outputting a forebay water level correction prediction result and prediction uncertainty; obtaining small hydropower station water head scheduling data; and establishing a start-stop trigger condition based on the forebay water level correction prediction result, the prediction uncertainty and the small hydropower water head scheduling data. According to the invention, the problem of unstable power generation caused by strong inflow randomness and poor regulation capability of small hydropower stations is solved based on a hydropower control technology, and the operation efficiency and the automation level are remarkably improved.

Description

Technical Field

[0001] This invention relates to the technical field of smart water conservancy, and in particular to a method for real-time prediction of hydrological data and start-up and shutdown control of small hydropower units. Background Technology

[0002] Existing small hydropower operation management and unit start-up and shutdown control technologies typically rely on manual experience or rule-based control methods based on fixed thresholds to respond to hydrological changes. For example, start-up and shutdown conditions are manually set based on real-time inflow, forebay water level, or empirical head range. While these methods can maintain basic operation in scenarios with relatively stable hydrological conditions, slow inflow changes, or a small number of units, they are insufficient in mountainous small hydropower stations where inflow fluctuations are frequent and hydrological changes exhibit significant randomness and lag. Due to the lack of ability to predict hydrological trends in advance, existing technologies struggle to provide timely and accurate guidance for unit start-up and shutdown decisions. Although existing technologies can achieve real-time acquisition of hydrological data through simple statistical analysis or short-term monitoring, they generally suffer from the following problems: First, most methods focus on judging the current hydrological state and lack joint modeling of historical hydrological data and real-time data, making it difficult to dynamically predict changes in the forebay water level and easily leading to delayed start-up and shutdown decisions; Second, start-up and shutdown control often uses fixed water level or head thresholds, without fully considering prediction uncertainty and head scheduling constraints, which can easily lead to frequent unit start-up and shutdown or load shocks; Third, there is a lack of effective linkage between the unit start-up and shutdown process and load protection and operation result evaluation, and the unit operating status cannot be fed back to the hydrological prediction model for correction. The overall control process lacks a closed-loop optimization mechanism, making it difficult to meet the needs of safe, stable and refined operation management of small hydropower. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a method for real-time prediction of hydrological data and start-up and shutdown control of small hydropower units, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units includes the following steps: Step S1: Collect hydrological data from small hydropower stations, perform outlier detection and filtering, and output basic hydrological data; Step S2: Obtain historical hydrological records; construct a dynamic prediction model for the forebay water level based on historical hydrological records and basic hydrological data, and generate the forebay water level prediction results; use basic hydrological data to correct the water level parameters of the forebay water level prediction results, and output the corrected forebay water level prediction results and prediction uncertainty. Step S3: Obtain small hydropower head scheduling data; establish start-stop trigger conditions based on the forebay water level correction prediction results, prediction uncertainty, and small hydropower head scheduling data; Step S4: Determine the unit coordination priority based on the start-stop trigger conditions; perform load protection based on the unit coordination priority data and record the unit operation results; transmit the unit operation results to the forebay water level dynamic prediction model and iterate to output a small hydropower start-stop control report.

[0005] The beneficial effects of this invention are as follows: (1) By jointly modeling historical hydrological records and real-time hydrological data, a dynamic prediction model for the forebay water level is constructed. Water level parameter correction and prediction uncertainty assessment are introduced to realize early perception and refined prediction of the forebay water level change trend, improve the temporal continuity and credibility of hydrological prediction results, and provide a stable data foundation for unit start-up and shutdown decisions.

[0006] (2) During the start-up and shutdown control process, a start-up and shutdown threshold sequence is generated based on the prediction results of the forebay water level correction, the prediction uncertainty and the head scheduling data. Through the segmented screening mechanism of positive trigger segment and reverse trigger segment, the frequent start-up and shutdown problem caused by a single fixed threshold is avoided, and the smoothness and adaptability of the start-up and shutdown decision of the unit are enhanced.

[0007] (3) By combining unit coordination priority, load protection simulation and turbine radial hydrodynamic assessment, the load change and load fluctuation during unit operation are quantitatively analyzed, and the operation results are fed back to the hydrological prediction model for iterative optimization, forming a closed-loop regulation mechanism of hydrological prediction and unit control, thereby improving the safety, stability and overall scheduling efficiency of small hydropower unit operation. Attached Figure Description

[0008] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the steps of a method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to the present invention. Figure 2 This is a schematic diagram of the execution flow of a method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to the present invention. Figure 3 This is a schematic diagram of the architecture of the dynamic prediction model for the forebay water level in this invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0009] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0010] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0011] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0012] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units, the method comprising the following steps: Step S1: Collect hydrological data from small hydropower stations, perform outlier detection and filtering, and output basic hydrological data; In one embodiment, hydrological sensors deployed at the inlet, forebay, and tailrace area collect water level, flow rate, rainfall, and water temperature data at a sampling frequency of 1 Hz. A sliding window anomaly detection is performed on each hydrological data sequence, with a window length of 60 seconds. The mean and standard deviation are calculated within the window, and values ​​exceeding the mean ± 3 times the standard deviation are marked as anomalies. Subsequently, linear interpolation is used to correct the anomalies, and a second-order Butterworth low-pass filter with a cutoff frequency of 0.1 Hz is applied to the corrected data sequence to eliminate high-frequency noise. Finally, basic hydrological data is output, providing reliable input data for subsequent forebay water level dynamic prediction and unit start-up / shutdown decisions.

[0013] In another embodiment, assuming 24-hour hydrological data was collected, a total of 86,400 sampling points were obtained for water level and 86,400 sampling points for flow rate. Anomaly detection identified 137 flow rate anomalies and 92 water level anomalies, accounting for less than 0.2% of the total data. After filtering, the mean square fluctuation amplitude of water level decreased from 3.6 cm to 1.1 cm, generating a basic hydrological data sequence for subsequent training and real-time updating of the dynamic prediction model.

[0014] Step S2: Obtain historical hydrological records; construct a dynamic prediction model for the forebay water level based on historical hydrological records and basic hydrological data, and generate the forebay water level prediction results; use basic hydrological data to correct the water level parameters of the forebay water level prediction results, and output the corrected forebay water level prediction results and prediction uncertainty. In one embodiment, historical hydrological records for three consecutive years, including water level, flow rate, and power output information, are obtained from a historical database and time-aligned with real-time basic hydrological data. A dynamic prediction model for the forebay water level is constructed, comprising an input normalization layer, a hydrological feature extraction layer, and a water level time series prediction layer. This model uses hydrological data from the past 30 minutes to predict water level changes for the next 10 minutes. Real-time water level values ​​are extracted to calculate the prediction residual sequence. Based on the residuals, prediction fusion weights are determined, water level parameters are corrected on the prediction results, and prediction uncertainty is calculated, providing accurate reference for start-up / shutdown triggering conditions and unit scheduling.

[0015] In another embodiment, assuming a historical record length of 1095 days, a prediction window of 10 minutes, and a prediction step size of 1 minute, the model outputs the predicted forebay water level as [312.45m, 312.47m, 312.52m], with a real-time water level of 312.50m, a residual mean of 0.06m, a standard deviation of 0.03m, and a corrected forebay water level range of 312.48~312.58m. The prediction uncertainty is ±0.05m, providing a quantifiable basis for the subsequent generation of start-stop trigger conditions.

[0016] Step S3: Obtain small hydropower head scheduling data; establish start-stop trigger conditions based on the forebay water level correction prediction results, prediction uncertainty, and small hydropower head scheduling data; In one embodiment, small hydropower head scheduling data is acquired, and the target head sequence for the corresponding time period is extracted. The predicted result of the forebay water level correction and the prediction uncertainty are constructed into a prediction mapping sequence, which is then time-aligned with the head sequence. The head difference is calculated, and a start-up / shutdown threshold sequence is generated. When the predicted water level still meets the head scheduling constraints after considering the uncertainty, it is marked as a positive start-up / shutdown trigger condition; when the constraints are not met, it is marked as a reverse trigger condition. This provides an automated reference for unit start-up / shutdown decisions, improving the foresight and safety of the start-up / shutdown strategy.

[0017] In another embodiment, assuming the scheduled head sequence is [28.5m, 28.6m, 28.7m], the predicted head corresponding to the water level is [28.8m, 28.9m, 29.0m], the prediction uncertainty is ±0.05m, and the calculated head difference is greater than 0.2m, a positive start-up trigger condition is generated for 3 consecutive minutes; when the head difference drops below −0.1m, the corresponding time period is marked as a shutdown trigger condition, thus realizing refined start-up and shutdown trigger division.

[0018] Step S4: Determine the unit coordination priority based on the start-stop trigger conditions; perform load protection based on the unit coordination priority data and record the unit operation results; transmit the unit operation results to the forebay water level dynamic prediction model and iterate to output a small hydropower start-stop control report.

[0019] In one embodiment, available generating units are selected based on start-up and shutdown trigger conditions, and the computer group coordinates priorities by combining the current load rate, number of start-ups and shutdowns, and runtime. Load protection simulations are executed according to priority, adjusting the guide vane opening of the generating units step by step, while simultaneously recording water flow velocity and turbine operating status, and the computer group's operational results. These results are fed back to the forebay water level dynamic prediction model, iteratively updating the model parameters to generate a small hydropower start-up and shutdown control report. This achieves closed-loop optimization of prediction and control, improving unit operating safety and load scheduling efficiency.

[0020] In another embodiment, assuming the power station has three generating units with current load rates of 65%, 48%, and 72%, the coordination priority is calculated as 2→1→3. In the load protection simulation, the guide vane opening of unit 2 is adjusted from 42% to 50%, corresponding to a water flow velocity increase from 3.1 m / s to 3.6 m / s, and an 8.5% increase in the radial hydrodynamic force of the runner. After feedback updates, the water level error in the prediction model for the next cycle decreases from 0.06 m to 0.03 m, generating a complete start-up and shutdown control report, providing quantitative basis for the safe and stable operation of small hydropower.

[0021] It should be noted that you should refer to [link / reference]. Figure 2 The process begins with S1, which collects hydrological data from small hydropower stations and performs outlier detection and filtering to obtain basic hydrological data. Next, S2 acquires historical hydrological records and combines them with the basic hydrological data to construct a dynamic prediction model for the forebay water level, generating prediction results. These predictions are then corrected using the basic hydrological data to obtain the corrected prediction results and prediction uncertainty for the forebay water level. Following this, S3 acquires small hydropower station head scheduling data and, combined with the corrected prediction results and prediction uncertainty for the forebay water level, establishes start-stop trigger conditions. Then, S4 determines the unit coordination priority based on the start-stop trigger conditions, performs load protection based on this, records the unit operation results, and simultaneously feeds the unit operation results back to the dynamic prediction model for the forebay water level for iteration. Finally, the process ends after outputting a small hydropower station start-stop control report.

[0022] Preferably, step S2, which involves constructing a dynamic prediction model for the forebay water level based on historical hydrological records and basic hydrological data, and generating the forebay water level prediction results, includes: A dynamic prediction model for forebay water level is constructed based on historical hydrological records and basic hydrological data. The dynamic prediction model for forebay water level includes an input normalization layer, a hydrological feature extraction layer, and a water level time series prediction layer. In one embodiment, a dynamic prediction model for forebay water level is constructed based on historical hydrological records and real-time basic hydrological data. The model includes an input normalization layer, a hydrological feature extraction layer, and a water level time-series prediction layer. The input normalization layer standardizes historical water level, flow rate, and rainfall data to ensure a mean of 0 and a variance of 1, guaranteeing that data of different dimensions can be modeled uniformly. The hydrological feature extraction layer uses the ReLU activation function and multi-scale convolution to perform feature mapping on the standardized hydrological data, generating multi-scale activation feature maps of time series and local water level changes. The water level time-series prediction layer predicts pooling features through linear regression or one-dimensional convolution operations, outputting the predicted forebay water level for several future time periods.

[0023] In another embodiment, it is assumed that three consecutive years of historical hydrological data are acquired, with a daily sampling interval of 1 minute, totaling approximately 1,576,800 data points. The basic hydrological data includes real-time water level, flow rate, and rainfall. The input normalization layer standardizes the water level data with a mean of approximately 312.5m to a mean of 0 and a variance of 1. The hydrological feature extraction layer generates multi-scale activation feature maps with feature dimensions of [512×32] and [256×64], which are then pooled to generate 128-dimensional pooled data. The water level time series prediction layer outputs a predicted water level sequence for the next 10 minutes as [312.48m, 312.50m, 312.52m, ...], providing a quantifiable basis for start / stop triggering conditions and load scheduling.

[0024] The input normalization layer is used to standardize historical hydrological records and basic hydrological data to obtain normalized hydrological data; the hydrological feature extraction layer is used to map activation functions onto the normalized hydrological data to obtain multi-scale activation feature maps; pooling data is generated based on the multi-scale activation feature maps. In one embodiment, the input normalization layer standardizes historical hydrological records and basic hydrological data, mapping data from different sources and units to the [-1,1] interval. The hydrological feature extraction layer uses multi-layer convolution combined with the ReLU activation function to process the standardized data, extracting time-series variation patterns and local water level fluctuation features to form a multi-scale activation feature map. Subsequently, the feature map is pooled to reduce the spatial dimension to a fixed length, generating a pooled data vector to represent the dynamic water level changes within different time windows, providing input features for the water level time-series prediction layer.

[0025] In another embodiment, it is assumed that after standardization, the historical data has a water level sequence range of [-0.6, 0.7] and a flow rate sequence range of [-1.2, 1.3]. Multi-scale convolution generates two activation feature maps with sizes of [512×32] and [256×64], respectively. After pooling, a 128-dimensional vector is obtained, each representing the water level and flow rate changes over the past 30 minutes, used to capture rapid fluctuations and slow trends. This pooled data reflects both the overall hydrological situation and preserves local abrupt changes, providing accurate input for short-term forecasting.

[0026] It should be noted that you should refer to [link / reference]. Figure 3 The paper presents the structural process of the forebay water level dynamic prediction model: First, the normalization layer standardizes historical hydrological records and basic hydrological data (such as water level, flow, and rainfall) (e.g., Z-score, mapping to [-1,1]) to obtain normalized hydrological data; then, the hydrological feature extraction layer generates multi-scale activation feature maps (such as 32×32 / 16×16 / 8×8 or [512×32] / [256×64] dimensions) through the ReLU activation function mapping module (combined with multi-scale convolution), and then obtains pooled data (such as a 128-dimensional vector) through the 2×2 max pooling layer; finally, the water level time series prediction layer performs linear regression and other operations on the pooled data to output the forebay water level prediction results.

[0027] The water level time series prediction layer is used to perform linear regression calculations on pooled data to generate forebay water level prediction results.

[0028] In one embodiment, the water level time-series prediction layer takes pooled data as input and predicts future water level trends through linear regression or one-dimensional convolution. This layer outputs a sequence of predicted forebay water levels, which may include water level values ​​at several prediction times. Combined with the error distribution obtained during model training, it provides a prediction confidence interval. The prediction results can be used to generate start-up and shutdown triggering conditions and unit scheduling schemes, while also providing initial reference values ​​for subsequent water level parameter corrections, thus achieving a closed-loop real-time water level prediction system.

[0029] In another embodiment, assuming that after pooled data is input into the prediction layer, the linear regression outputs a sequence of predicted water levels for the next 10 minutes as [312.48m, 312.50m, 312.51m, 312.53m, 312.55m, 312.54m, 312.56m, 312.57m, 312.58m, 312.59m], with a mean prediction error of 0.06m and a standard deviation of 0.03m. Combining the prediction residuals, a confidence interval of ±0.05m is calculated, providing a quantitative basis for determining start-up and shutdown trigger conditions and controlling unit operation. This also supports real-time iterative updates, improving the accuracy of small hydropower water level prediction and the reliability of unit start-up and shutdown decisions.

[0030] Preferably, in step S2, the water level prediction results of the forebay are corrected using basic hydrological data, and the corrected water level prediction results and prediction uncertainties of the forebay are output, including: Extract real-time water level data from basic hydrological data; calculate water level residual sequence based on real-time water level data and forebay water level prediction results; determine prediction fusion weights based on water level residual sequence; In one embodiment, the real-time water level sequence is first extracted from the basic hydrological data. Combined with the prediction results generated by the forebay water level dynamic prediction model, the water level residual sequence for each time point is calculated; that is, the residual equals the actual water level minus the predicted water level. Based on the residual sequence, a weighted moving average or adaptive Kalman filter method is used to analyze the short-term and long-term error distributions, determining the fusion weight of each prediction model in the current time window. A higher fusion weight indicates higher prediction reliability under current conditions, while a lower weight indicates a larger prediction deviation.

[0031] In another embodiment, assuming the real-time water level sequence is sampled every minute for the past 60 minutes, totaling 60 points, the prediction model outputs a forebay water level sequence of [312.50, 312.51, 312.52, ...] m, and the actual measured water level sequence is [312.48, 312.52, 312.53, ...] m, then the residual sequence is [-0.02, 0.01, 0.01, ...] m. The prediction fusion weight sequence is calculated using the residual adaptive moving average as [0.65, 0.70, 0.68, ...], where the weights are reduced for time points with large short-term fluctuations and increased for time points with stable long-term trends. This weight sequence can be directly used to dynamically adjust the output fusion strategy of the prediction model, ensuring that the corrected prediction result is closer to the real-time water level.

[0032] The dynamic prediction model for the forebay water level is adjusted according to the prediction fusion weight, and the prediction result of the forebay water level correction is output; the prediction uncertainty is calculated based on the prediction result of the forebay water level correction.

[0033] In one embodiment, the prediction results output by the dynamic prediction model of the forebay water level are corrected according to the prediction fusion weights. Specific methods include weighted averaging, residual compensation, or recursive iterative adjustment, fusing prediction results under different weights to generate a corrected prediction sequence for the forebay water level. Simultaneously, the prediction uncertainty, i.e., the confidence interval or standard deviation range at each prediction time, is calculated using the historical residual distribution and the current residual fluctuation amplitude, reflecting the reliability and potential error of the corrected prediction results. The corrected prediction results and uncertainty sequence can be used for unit start-up and shutdown determination, scheduling priority calculation, and protective load decision-making, realizing real-time closed-loop water level control for small hydropower plants.

[0034] In another embodiment, assuming the forebay water level prediction sequence after fusion correction is [312.49, 312.50, 312.52, 312.53, 312.54]m, the prediction uncertainty sequence is calculated as [±0.04, ±0.03, ±0.03, ±0.05, ±0.04]m by combining historical residuals and current fluctuations. By comparing the corrected prediction value with the actual observed water level, it can be found that the average deviation is about 0.02m and the standard deviation is 0.03m, indicating that the corrected prediction can effectively reduce short-term errors. At the same time, the uncertainty sequence can be used to dynamically determine the safety boundary of unit start-up and shutdown, providing a quantitative basis for dispatch control.

[0035] Preferably, step S3 includes: Acquire small hydropower head scheduling data; extract the head sequence from the small hydropower head scheduling data; integrate the forebay water level correction prediction results and prediction uncertainty into a prediction mapping sequence according to the time dimension; In one embodiment, firstly, small hydropower head scheduling data is acquired, including real-time upper reservoir water level, downstream water level, and unit operating status, and the head sequence is extracted, i.e., the head difference or effective head difference corresponding to each scheduling moment. Subsequently, the forebay water level correction prediction results and the prediction uncertainty sequence are integrated along the time dimension to generate a prediction mapping sequence, with each moment corresponding to the corrected water level, confidence interval, and expected head change range. During integration, linear interpolation or weighted averaging methods can be used to ensure time alignment and data continuity, while retaining uncertainty information for subsequent risk assessment.

[0036] In another embodiment, assuming the obtained head scheduling sequence is [4.85, 4.87, 4.88, 4.90, 4.92]m, the forebay level correction prediction sequence is [312.49, 312.50, 312.52, 312.53, 312.54]m, and the prediction uncertainty sequence is [±0.04, ±0.03, ±0.03, ±0.05, ±0.04]m, the generated prediction mapping sequence contains the corrected water level and confidence interval information for each time point, such as (312.49, ±0.04, head 4.85m). This sequence can be used to further generate unit start-up and shutdown decision inputs, ensuring that the scheduling plan maximizes unit utilization efficiency while guaranteeing safety margins.

[0037] A start / stop threshold sequence is generated based on the predicted mapping sequence and the head sequence; start / stop triggering conditions are established based on the start / stop threshold sequence.

[0038] In one embodiment, based on the predicted mapping sequence and the actual head sequence, a unit start-up and shutdown strategy threshold is set, including an upper head start-up threshold and a lower head stop-shutdown threshold. Specifically, at each time point, a safety margin (such as prediction uncertainty multiplied by a coefficient) is subtracted or added to the predicted head to obtain a start-up and shutdown threshold sequence. Then, start-up and shutdown triggering conditions are established based on this sequence: when the actual head reaches or exceeds the start-up threshold, the unit is triggered to start; when the actual head is below the stop-shutdown threshold, the unit is triggered to shut down. This method enables joint control of head and forebay water level correction predictions, reducing frequent start-ups and shutdowns caused by water level fluctuations and improving the operational stability of small hydropower plants.

[0039] In another embodiment, assuming the start-stop threshold sequence is a start threshold of [4.90, 4.91, 4.92, 4.94, 4.95]m and a stop threshold of [4.80, 4.81, 4.82, 4.84, 4.85]m, the actual head at each time point in the corresponding predicted mapping sequence is [4.85, 4.87, 4.88, 4.90, 4.92]m. Based on the triggering conditions, in frames 1 and 2, if the head is below the start threshold, no action is taken; in frames 3 and 4, if the head reaches the start threshold, the unit is triggered to start; and in frame 5, if the head is still above the start threshold, operation continues. This setting, by simulating head at multiple times and predictive uncertainty, can verify the rationality and stability of the start-stop conditions under different fluctuation scenarios, providing a quantitative basis for small hydropower dispatching strategies.

[0040] Preferably, generating the start / stop threshold sequence based on the predicted mapping sequence and the head sequence includes: Extract prediction mapping segments based on the prediction mapping sequence; extract head matching segments based on the head sequence; calculate the head difference based on the prediction mapping segments and the head matching segments; In one embodiment, firstly, based on the predicted mapping sequence, the entire time series is divided into several predicted mapping segments according to the water level change trend. Each segment contains the corrected predicted water level and corresponding uncertainty information, and the start and end times of the segment are marked. Then, based on the actual water head sequence, the water head change trend is matched and segmented to ensure that each matched segment corresponds to a predicted mapping segment in time. For each pair of time-corresponding segments, the water head difference is calculated, which is the actual water head sequence minus the corrected water level value in the predicted mapping sequence, resulting in a water head deviation sequence. This water head difference sequence can quantify the water head fluctuation amplitude, providing basic data for subsequent start / stop trigger segment selection, while retaining the timestamps and uncertainty information of the segments for dynamic threshold calculation.

[0041] In another embodiment, assume that the corrected water levels for the predicted mapping sequence at five consecutive time points are [312.48, 312.50, 312.51, 312.53, 312.54] m, corresponding to a head sequence of [4.87, 4.88, 4.90, 4.92, 4.93] m. After dividing the predicted mapping segment and the head matching segment according to the trend, the calculated head difference sequence is [0.39, 0.38, 0.39, 0.39, 0.39] m, representing the deviation of the actual head from the predicted water level at each time point.

[0042] Select forward and reverse trigger segments based on the head difference; generate a start / stop threshold sequence based on the forward and reverse trigger segments.

[0043] In one embodiment, based on the head difference sequence, forward and reverse triggering conditions are determined for each segment: when the head difference is greater than a preset forward triggering threshold, the segment is designated as a forward triggering segment; when the head difference is less than a preset reverse triggering threshold, it is designated as a reverse triggering segment. Subsequently, the time series of the forward and reverse triggering segments are integrated with the corresponding predictive mapping sequences to generate the final start-up and shutdown threshold sequence. The start-up and shutdown threshold sequence contains the unit start-up threshold and stop-up threshold at each time point, which can be directly used by the scheduling system to trigger unit start-up and shutdown operations, thereby realizing a scheduling strategy that combines dynamic head control with predictive correction.

[0044] In another embodiment, assuming the head difference sequence is [0.39, 0.38, 0.39, 0.39, 0.39]m, the forward trigger threshold is set to 0.38m, and the reverse trigger threshold is set to 0.36m. Then, the head difference in frames 1, 3, 4, and 5 is greater than or equal to the forward threshold and is classified as a forward trigger segment; the difference in frame 2 is slightly higher than the reverse threshold and can be considered as a transition segment that is not triggered temporarily. This generates a start-stop threshold sequence, for example, a start threshold of [4.87, 4.90, 4.91, 4.93, 4.94]m and a stop threshold of [4.77, 4.80, 4.81, 4.83, 4.84]m. This can be used to verify the rationality of the unit's start-stop in each segment during simulated operation, providing data support for the dynamic scheduling of small hydropower.

[0045] Preferably, the selection of forward and reverse trigger segments based on the head difference includes: Set forward head threshold and reverse head threshold, where the forward head threshold is greater than zero and the reverse head threshold is less than zero; In one embodiment, forward head thresholds and reverse head thresholds are set based on the operating experience of small hydropower units and head prediction results. The forward head threshold is set to a value greater than zero to determine unit start-up conditions; the reverse head threshold is set to a value less than zero to determine unit shutdown conditions. The thresholds can be determined by combining historical head fluctuation amplitudes, prediction uncertainty, and start-up / shutdown safety margins, and can be dynamically updated according to a time series to adapt to head change trends. Once these thresholds are set, they provide a criterion for the subsequent division of forward and reverse trigger segments.

[0046] In another embodiment, assuming the forward head threshold is set to 0.35m and the reverse head threshold is set to -0.25m, and the corresponding head difference sequence at 10 consecutive time points is [0.32, 0.36, 0.38, 0.33, 0.40, -0.20, -0.28, -0.26, 0.34, 0.37]m. This setting ensures that the forward threshold covers the rising head band and the reverse threshold covers the falling head band, providing a quantitative standard for triggering segment identification.

[0047] The moment when the head difference transitions from below the positive head threshold to above the positive head threshold is taken as the positive triggering start point; starting from the positive triggering start point, the subsequent head difference is continuously tracked until the head difference falls back below the positive head threshold, and this segment is marked as a positive triggering segment; In one embodiment, when the head difference sequence transitions from below the positive head threshold to above the positive head threshold, this moment is determined as the positive trigger start point. From this start point, subsequent head differences are continuously tracked until the head difference falls back below the positive head threshold, and this time period is marked as a positive trigger segment. The entire process can be performed separately for each unit or each head prediction sequence to generate a complete list of positive trigger segments for unit startup scheduling or dynamic head control.

[0048] In another embodiment, based on the assumed head difference sequence [0.32, 0.36, 0.38, 0.33, 0.40, –0.20, –0.28, –0.26, 0.34, 0.37]m and a positive threshold of 0.35m, the second moment when the head difference transitions from 0.32m to 0.36m is marked as the positive triggering start point; subsequently, continuous tracking occurs, and the head difference falls back to below 0.33m at the fourth moment, determining the second to fourth moments as a positive triggering segment; at the fifth moment, the head difference of 0.40m exceeds the threshold again, marking another positive triggering segment, and so on, generating a sequence of positive triggering segments.

[0049] Identify the moment when the head difference transitions from above the reverse head threshold to below the reverse head threshold as the starting point for reverse triggering; starting from the reverse triggering point, continuously track subsequent head differences until the head difference rises back above the reverse head threshold, and mark this segment as a reverse trigger segment.

[0050] In one embodiment, when the head difference sequence transitions from above the reverse head threshold to below the reverse head threshold, this moment is determined as the reverse trigger start point. Starting from this point, subsequent head differences are continuously tracked until the head difference rises back above the reverse head threshold, and this time period is marked as the reverse trigger segment.

[0051] In another embodiment, based on the assumed head difference sequence [0.32, 0.36, 0.38, 0.33, 0.40, –0.20, –0.28, –0.26, 0.34, 0.37]m and the reverse threshold –0.25m, the head difference at time 7 (–0.28m) is below the reverse threshold, marking it as the start of the reverse trigger. Subsequently, the head difference rises back to –0.26m at time 9, still below the threshold, and rises to 0.34m at time 10, above the threshold. Timees 7 to 10 are thus determined to be the reverse trigger segment. This yields a complete reverse trigger segment sequence, which can be used together with the forward trigger segment to construct the unit start-up and shutdown scheduling strategy.

[0052] Preferably, step S4 includes: Available generating units are selected based on start-up and shutdown trigger conditions, and the load allocation of available generating units is calculated; the coordination priority of generating units is determined based on the load allocation. In one embodiment, the system first acquires start-up and shutdown trigger conditions such as the current reservoir forebay water level, upstream inflow, and real-time operating status of each generating unit. Based on preset start-up and shutdown trigger rules, a list of generating units that can be put into operation is filtered, and the available load allocation for each unit is calculated, taking into account the unit's rated power, historical operating efficiency, and instantaneous head conditions. Subsequently, the unit coordination priority is determined based on the load allocation: units with higher loads or higher efficiency are prioritized to ensure balanced scheduling and avoid inefficient operation or overload of units.

[0053] In another embodiment, assume the small hydropower station has 5 available generating units with rated capacities of [12MW, 10MW, 8MW, 6MW, 4MW]. Current water level triggering conditions allow units 2, 3, and 4 to be activated, while units 1 and 5 are temporarily deactivated. Based on head and historical efficiency, the load allocation is calculated to be [0MW, 8.5MW, 6.8MW, 5.0MW, 0MW]. Therefore, the coordination priority is determined as Unit 2 > Unit 3 > Unit 4. This allocation ensures the units operate within a safe power range and provides a clear sequence for subsequent load protection and start-up / shutdown control.

[0054] Load protection is implemented based on unit coordination priority data, and unit operation results are recorded. The unit operation results are transmitted to the forebay water level dynamic prediction model and iterated to output a small hydropower start-up and shutdown control report.

[0055] In one embodiment, based on unit coordination priorities, the system performs load protection control on each unit, including limiting overload, adjusting the rate of change of output, and monitoring key parameters such as rotational speed and turbine vibration. After load protection is completed, the operating results of each unit are recorded, including instantaneous load, rotational speed, start-up and shutdown status, and head changes. These operating results are transmitted in real time to the forebay water level dynamic prediction model, which iteratively calculates and predicts future water level changes and unit operating conditions, generating a small hydropower start-up and shutdown control report. The report includes the start-up and shutdown times of each unit, load change curves, and predicted water level ranges, providing quantitative basis for scheduling decisions and safe operation.

[0056] In another embodiment, assume that the unit load (unit: MW) and start / stop status are as follows over 10 consecutive time steps: Unit 2 load: [8.5, 8.3, 8.6, 8.4, 8.2, 8.5, 8.3, 8.6, 8.1, 8.4]; Unit 3 load: [6.8, 6.7, 6.9, 6.8, 6.6, 6.9, 6.7, 6.8, 6.5, 6.7]; Unit 4 load: [5.0, 5.1, 4.9, 5.0, 5.2, 5.0, 5.1, 4.8, 5.0, 5.1]; assume the maximum allowable load offset is ±0.5MW (consistent threshold). The calculated load offset sequence (Unit 2) is: [0.2, 0.3, 0.1, 0.4, 0.5, 0.2, 0.3, 0.1, 0.5, 0.4]. Time steps less than or equal to τ (frames 1, 2, 3, 6, 7, and 8, a total of 6 frames) are identified as "load continuity zones"; time steps greater than τ (frames 4, 5, 9, and 10, a total of 4 frames) are identified as "load discontinuity zones".

[0057] In the iterative prediction, the predicted range of the forebay water level is [105.0, 105.5] m. The unit start-up and shutdown strategies are adjusted according to the load changes in the continuous / discontinuous zones. For example, if the predicted water level of Unit 4 drops to the lower safety limit in the 5th frame, the start-up and shutdown are temporarily suspended; Units 2 and 3 continue to generate electricity to ensure the overall load stability of the system. Finally, a small hydropower start-up and shutdown control report is generated, including the start-up and shutdown time of each unit, load change curve, predicted water level, and safety status, providing a quantitative basis for dispatching.

[0058] Of particular importance is the transmission of unit operation results to the forebay water level dynamic prediction model, followed by iteration, to output a small hydropower start-up and shutdown control report, including: Operational deviation data is extracted based on merged operation records; hydrological correlation deviation sequence is determined based on operational deviation data and basic hydrological data; the hydrological correlation deviation sequence is transmitted to the forebay water level dynamic prediction model to generate iterative correction parameters; In one embodiment, the system first extracts operational deviation data based on merged operation records (including real-time unit flow, start-up and shutdown status, water level data, etc.), and matches the deviation data with basic hydrological data (such as inflow, rainfall, runoff, etc.) to obtain a hydrologically correlated deviation sequence. Subsequently, this deviation sequence is input into a forebay water level dynamic prediction model (e.g., based on LSTM or an autoregressive model). Through iterative prediction and error feedback, a correction parameter sequence is generated to adjust the water level prediction results for future time steps. This iterative correction parameter reflects the impact of real-time hydrological conditions and unit operational deviations on the forebay water level, providing a quantitative basis for start-up and shutdown control.

[0059] In another embodiment, assume that the operational deviations (in m³ / s) for 10 consecutive time segments extracted from the merged operational records are: [1.2,–0.8,0.5,1.0,–1.1,0.9,–0.6,1.3,–0.7,0.8], and the corresponding inflow sequence of the basic hydrological data is: [35,37,36,34,38,36,35,37,36,34]. The deviation sequence obtained through hydrological correlation calculation is: [0.9,–0.6,0.4,0.8,–0.9,0.7,–0.5,1.0,–0.6,0.6]. The sequence was input into the iterative prediction model, and after three iterations, the corrected parameter sequence was obtained: [0.92,–0.58,0.41,0.79,–0.88,0.69,–0.48,1.02,–0.61,0.63], which was used to guide the dynamic adjustment of the forebay water level.

[0060] The forebay water level correction prediction results are continuously updated based on iterative correction parameters, and the forebay water level correction prediction results are used to generate a small hydropower start-up and shutdown control report.

[0061] In one embodiment, the system uses iteratively corrected parameters to update the predicted forebay water level in real time. The prediction results, combined with unit operating constraints (such as minimum head, start-up and shutdown time limits, and downstream flow requirements), generate a small hydropower start-up and shutdown control report. The report includes predicted water levels for several future time segments, unit start-up and shutdown recommendations, and water level safety assessment indicators, providing a reference for scheduling decisions. This process ensures the continuity and safety of forebay water level control and reduces operational losses caused by frequent unit start-ups and shutdowns.

[0062] In another embodiment, assuming the current water level in the forebay is 320.5m, the corrected water levels (in meters) for the next 10 time segments predicted based on iterative correction parameters are: [320.7, 321.0, 320.8, 321.2, 320.9, 321.1, 321.0, 321.3, 320.9, 321.2]. According to preset start-up and shutdown rules: when the water level is higher than 321.0m, it is recommended to start Unit 1; when it is lower than 320.7m, it is recommended to shut down Unit 1. This generates a small hydropower start-up and shutdown control report, indicating that Unit 1 should be started in segments 2, 4, 6, 8, and 10, while remaining in either running or shut-down state in the remaining time segments. This report also includes water level deviation analysis and safety warning information, facilitating real-time decision-making by dispatchers.

[0063] Preferably, the process of filtering available generating units based on start-stop triggering conditions and calculating the load allocation for available generating units includes: Select available generating units based on start-stop trigger conditions; record the unit load level of available generating units; determine the unit load trend based on the unit load level; calculate the load surge based on the unit load trend; determine the load allocation of available generating units based on the load surge.

[0064] In one embodiment, the system first acquires the current reservoir forebay water level, upstream inflow, and start-up / shutdown trigger conditions such as the operating status of each generating unit, and then filters out a list of available generating units based on these conditions. Subsequently, it records the real-time load factor of each available generating unit, i.e., the ratio of current output power to rated power. By analyzing the load factor's trend over time, the system calculates the load surge for each generating unit to assess its ability to withstand instantaneous load fluctuations. Finally, based on the load surge, combined with the unit's rated power, efficiency, and start-up / shutdown constraints, the system calculates the load allocation for each available generating unit to ensure overall system load balance and that the units operate within a safe power range.

[0065] In another embodiment, assume a small hydropower station has five generating units with rated capacities of 12MW, 10MW, 8MW, 6MW, and 4MW. Based on the current start-stop trigger conditions, units 1, 2, and 3 are determined to be available. Their load factors are recorded as 0.75, 0.60, and 0.50, corresponding to current output powers of 9.0MW, 6.0MW, and 4.0MW. By analyzing the load changes over three consecutive minutes, the load trends are calculated to be +0.02, -0.01, and +0.03 MW / min, resulting in load surges of 0.5, 0.3, and 0.4MW. Combining the rated capacities of the units and their surge withstand capabilities, the final load allocation is determined to be 9.2MW, 5.8MW, and 4.5MW. This allocation scheme can ensure that available units meet the system load while reducing instantaneous power fluctuations and optimizing unit operational stability.

[0066] Preferably, load protection is performed based on unit coordination priority data, and the unit operation results are recorded, including: Load protection simulation is performed based on unit coordination priority data, and load protection data is recorded synchronously; the unit guide vane opening is calculated based on the load protection data; the water flow velocity is calculated based on the unit guide vane opening; and the radial hydrodynamic force of the runner is evaluated based on the water flow velocity. In one embodiment, the system first performs load protection simulations on each available unit based on pre-calculated unit coordination priority data. During the simulation, load protection data for each unit is recorded synchronously, including real-time unit power output, guide vane opening changes, and unit flow fluctuations. Subsequently, based on the load protection data, the guide vane opening curves for each unit are calculated to reflect the unit's response capability to instantaneous load changes. Furthermore, by analyzing the relationship between guide vane opening and inlet flow rate, the system calculates the water velocity distribution of the unit runner, thereby assessing the hydrodynamic forces acting on the runner in the radial direction, providing fundamental data for subsequent runner structure analysis and bearing load assessment.

[0067] In another embodiment, it is assumed that the hydropower station has three generating units, with coordination priorities of Unit 1 > Unit 2 > Unit 3. During the load protection simulation, the changes in the guide vane opening of the generating units over time were recorded as follows: Unit 1 opening 45°→50°→48°, Unit 2 opening 40°→42°→41°, and Unit 3 opening 35°→37°→36°. Based on the relationship between the guide vane opening and the influent flow rate, the radial flow velocities of the runner were calculated as follows: Unit 1: 7.2 m / s, 7.5 m / s, 7.3 m / s; Unit 2: 6.8 m / s, 7.0 m / s, 6.9 m / s; Unit 3: 6.2 m / s, 6.4 m / s, 6.3 m / s. This hydrodynamic assessment provides a quantitative basis for the radial load distribution analysis of the runner and prepares input data for the next step of bearing load calculation.

[0068] Of particular importance is the assessment of the radial hydrodynamics of the impeller based on the water flow velocity; Determine the water flow pressure on the surface of a single impeller based on the water flow velocity; obtain the force-bearing area of ​​a single impeller blade; calculate the hydrodynamic load based on the force-bearing area of ​​the single impeller blade and the water flow pressure on the surface of the single impeller; In one embodiment, the system first uses flow velocity sensors or CFD simulation data to determine the local water flow pressure distribution on the surface of a single impeller at different flow velocities. Then, the force-bearing area of ​​each impeller blade is obtained, and the hydrodynamic load on each blade is calculated by multiplying the local water flow pressure by the corresponding force-bearing area. In practical operation, the blade surface can be divided into several small grid cells (e.g., 5mm × 5mm), the local hydrodynamic load is calculated for each grid cell, and then the loads of all grid cells are integrated or summed to obtain the total load on the single blade, which can be used for subsequent radial decomposition and overall impeller hydrodynamic evaluation.

[0069] In another embodiment, assuming a single impeller blade is 0.8m long and 0.15m wide, it is divided into 160 grid cells. Through flow measurement or CFD simulation, the local flow pressure (in Pa) sequence for each grid cell is obtained as: [1200, 1250, 1180, 1220, ..., 1210], with a force-bearing area of ​​0.0075m². The hydrodynamic load on each grid cell is then calculated as pressure multiplied by area. For example, the load on the first grid cell is approximately 1200 × 0.0075 ≈ 9 N, the load on the second grid cell is approximately 1250 × 0.0075 ≈ 9.375 N, and so on. Summing the loads from all grid cells yields a total hydrodynamic load of approximately 1450 N for the blade. This process can be repeated for different flow velocities or impeller positions to form a complete hydrodynamic load distribution curve.

[0070] Radial decomposition is performed based on hydrodynamic loads, and the radial hydrodynamic loads are recorded. The radial hydrodynamics of the runner are evaluated by multiplying the radial hydrodynamic loads by the preset number of blades.

[0071] In one embodiment, the system performs radial decomposition based on the hydrodynamic load of a single blade, according to the direction of force on the blade, to obtain radial components. Multiplying the radial hydrodynamic load of a single blade by the total number of blades on the runner allows for the assessment of the magnitude and distribution of the radial hydrodynamic force across the entire runner, thus providing a basis for bearing design, structural optimization, and start-stop control. Radial decomposition can be performed using an orthogonal decomposition method, projecting each load vector onto the radial direction and summing the results.

[0072] In another embodiment, the radial hydrodynamic load on a single blade is assumed to be [7.2, 7.5, 6.8, 7.1, ..., 7.0] N after decomposition, with a total of 160 mesh elements. The total radial hydrodynamic force on a single blade is approximately 1200 N. If there are 20 blades on the runner, the overall radial hydrodynamic force of the runner is approximately 1200 × 20 = 24,000 N. Further, a radial load distribution table or graph can be generated to show the hydrodynamic variations at different blades and mesh positions, in order to analyze the radial force balance and potential local stress concentration of the runner.

[0073] The bearing load sequence is generated based on the radial hydrodynamics of the runner; the load fluctuation value is statistically analyzed based on the generated bearing load sequence; and the load fluctuation value is used as the unit operation result.

[0074] In one embodiment, a bearing load sequence is generated based on the runner radial hydrodynamic data, reflecting the changes in radial force and torque experienced by each unit at different time periods. Subsequently, statistical analysis is performed on the bearing load sequence to calculate load fluctuation values, including the maximum load, minimum load, and average load fluctuation amplitude. These load fluctuation values ​​are recorded as unit operating results and used for unit structural health monitoring, maintenance decisions, and subsequent small hydropower start-up and shutdown control iterations.

[0075] In another embodiment, assume the bearing load sequence (in kN) generated from the radial hydrodynamic input of the runner is as follows: Unit 1: [12.5, 13.0, 12.8, 13.2, 12.9]; Unit 2: [10.8, 11.0, 10.9, 11.2, 11.0]; Unit 3: [9.5, 9.8, 9.6, 9.9, 9.7]. Calculations show that the load fluctuation range for Unit 1 is 0.7 kN, for Unit 2 it is 0.4 kN, and for Unit 3 it is 0.4 kN. These load fluctuation values ​​are recorded as unit operating results for subsequent generation of small hydropower start-up and shutdown control reports and scheduling optimization.

[0076] Preferably, load protection simulation is performed based on unit coordination priority data, and load protection data is recorded synchronously, including: Priority segmentation sequences are extracted based on unit coordination priority data; load protection simulation is performed based on priority segmentation sequences to determine real-time load status; In one embodiment, the system first extracts the priority segment sequence of each unit within the scheduling cycle based on the unit coordination priority data. Each segment corresponds to the unit's start-up and shutdown sequence and control priority level. Using this priority segment sequence, load protection simulation is performed, including simulating the power response, guide vane opening changes, and flow distribution of each unit under different load fluctuations. Through simulation, the load status of each unit can be calculated in real time to reflect the unit's stability under instantaneous load impacts. The simulation results are used to generate the real-time load status sequence of each unit, providing basic data for subsequent comparison and protection segment selection.

[0077] In another embodiment, assuming the hydropower station has three generating units, with priority segmentation sequences as follows: Unit 1: [High, Medium, High], Unit 2: [Medium, Low, Medium], and Unit 3: [Low, Low, Medium]. In the load protection simulation, the real-time load status (in MW) is obtained as follows: Unit 1: [45, 50, 48], Unit 2: [38, 36, 39], and Unit 3: [30, 32, 31]. This simulation can intuitively reflect the response capabilities of each unit under different load segments, providing a quantitative basis for selecting protection segments.

[0078] The real-time load status is compared segment by segment with the preset load protection target to obtain the load protection comparison result; based on the load protection comparison result, triggerable protection segments are selected; and load protection data is recorded according to the triggerable protection segments.

[0079] In one embodiment, the system compares the real-time load status with preset load protection targets (such as maximum allowable load fluctuation, start-stop safety thresholds, etc.) segment by segment to calculate load deviations and exceedances. Based on the comparison results, protectable segments that meet the triggering conditions are selected, with each segment corresponding to a time period and unit combination. Subsequently, load protection data, including unit power, guide vane opening, flow rate, and protection trigger identifier, is recorded according to the triggerable protection segments for subsequent scheduling optimization and safe operation assessment.

[0080] In another embodiment, assuming the load deviation (in MW) is compared over five consecutive scheduling time segments as follows: Unit 1: [0.8, 1.2, 0.5, 1.0, 1.5], Unit 2: [1.1, 0.9, 1.3, 0.7, 1.0], Unit 3: [0.6, 0.8, 0.7, 1.0, 0.9]. Based on a preset protection threshold of 1.0 MW, the triggerable protection segments are segments 2 and 5 of Unit 1, segments 1 and 3 of Unit 2, and segment 4 of Unit 3. The system records the load protection data of these segments as load protection trigger information for subsequent optimization of unit start-up and shutdown control strategies and load stability analysis.

[0081] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0082] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units, characterized in that, Includes the following steps: Step S1: Collect hydrological data from small hydropower stations, perform outlier detection and filtering, and output basic hydrological data; Step S2: Obtain historical hydrological records; construct a dynamic prediction model for the forebay water level based on historical hydrological records and basic hydrological data, and generate the forebay water level prediction results; use basic hydrological data to correct the water level parameters of the forebay water level prediction results, and output the corrected forebay water level prediction results and prediction uncertainty. Step S3: Obtain small hydropower head scheduling data; establish start-stop trigger conditions based on the forebay water level correction prediction results, prediction uncertainty, and small hydropower head scheduling data; Step S4: Determine the unit coordination priority based on the start-stop trigger conditions; perform load protection based on the unit coordination priority data and record the unit operation results; transmit the unit operation results to the forebay water level dynamic prediction model and iterate to output a small hydropower start-stop control report.

2. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 1, characterized in that, Step S2 involves constructing a dynamic prediction model for the forebay water level based on historical hydrological records and basic hydrological data, and generating the forebay water level prediction results, including: A dynamic prediction model for forebay water level is constructed based on historical hydrological records and basic hydrological data. The dynamic prediction model for forebay water level includes an input normalization layer, a hydrological feature extraction layer, and a water level time series prediction layer. The input normalization layer is used to standardize historical hydrological records and basic hydrological data to obtain normalized hydrological data; the hydrological feature extraction layer is used to map activation functions onto the normalized hydrological data to obtain multi-scale activation feature maps; pooling data is generated based on the multi-scale activation feature maps. The water level time series prediction layer is used to perform linear regression calculations on pooled data to generate forebay water level prediction results.

3. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 1, characterized in that, In step S2, the water level prediction results of the forebay are corrected using basic hydrological data, and the corrected water level prediction results of the forebay and the prediction uncertainty are output, including: Extract real-time water level data from basic hydrological data; calculate water level residual sequence based on real-time water level data and forebay water level prediction results; determine prediction fusion weights based on water level residual sequence; The dynamic prediction model for the forebay water level is adjusted according to the prediction fusion weight, and the prediction result of the forebay water level correction is output; the prediction uncertainty is calculated based on the prediction result of the forebay water level correction.

4. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 1, characterized in that, Step S3 includes: Acquire small hydropower head scheduling data; extract the head sequence from the small hydropower head scheduling data; integrate the forebay water level correction prediction results and prediction uncertainty into a prediction mapping sequence according to the time dimension; A start / stop threshold sequence is generated based on the predicted mapping sequence and the head sequence; start / stop triggering conditions are established based on the start / stop threshold sequence.

5. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 4, characterized in that, The start / stop threshold sequence is generated based on the predicted mapping sequence and the head sequence, including: Extract prediction mapping segments based on the prediction mapping sequence; extract head matching segments based on the head sequence; calculate the head difference based on the prediction mapping segments and the head matching segments; Select forward and reverse trigger segments based on the head difference; generate a start / stop threshold sequence based on the forward and reverse trigger segments.

6. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 5, characterized in that, The selection of forward and reverse trigger segments based on head difference includes: Set forward head threshold and reverse head threshold, where the forward head threshold is greater than zero and the reverse head threshold is less than zero; The moment when the head difference transitions from below the positive head threshold to above the positive head threshold is taken as the positive triggering start point; starting from the positive triggering start point, the subsequent head difference is continuously tracked until the head difference falls back below the positive head threshold, and this segment is marked as a positive triggering segment; Identify the moment when the head difference transitions from above the reverse head threshold to below the reverse head threshold as the starting point for reverse triggering; starting from the reverse triggering point, continuously track subsequent head differences until the head difference rises back above the reverse head threshold, and mark this segment as a reverse trigger segment.

7. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 1, characterized in that, Step S4 includes: Available generating units are selected based on start-up and shutdown trigger conditions, and the load allocation of available generating units is calculated; the coordination priority of generating units is determined based on the load allocation. Load protection is implemented based on unit coordination priority data, and unit operation results are recorded. The unit operation results are transmitted to the forebay water level dynamic prediction model and iterated to output a small hydropower start-up and shutdown control report.

8. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 7, characterized in that, Available generating units are selected based on start-stop trigger conditions, and the load allocation of available generating units is calculated, including: Select available generating units based on start-stop trigger conditions; record the unit load level of available generating units; determine the unit load trend based on the unit load level; calculate the load surge based on the unit load trend; determine the load allocation of available generating units based on the load surge.

9. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 7, characterized in that, Load protection is performed based on unit coordination priority data, and the unit operation results are recorded, including: Load protection simulation is performed based on unit coordination priority data, and load protection data is recorded synchronously; the unit guide vane opening is calculated based on the load protection data; the water flow velocity is calculated based on the unit guide vane opening; and the radial hydrodynamic force of the runner is evaluated based on the water flow velocity. The bearing load sequence is generated based on the radial hydrodynamics of the runner; the load fluctuation value is statistically analyzed based on the generated bearing load sequence; and the load fluctuation value is used as the unit operation result.

10. The method for real-time prediction of hydrological data and start-up / shutdown control of small hydropower units according to claim 9, characterized in that, Load protection simulation is performed based on unit coordination priority data, and load protection data is recorded synchronously, including: Priority segmentation sequences are extracted based on unit coordination priority data; load protection simulation is performed based on priority segmentation sequences to determine real-time load status; The real-time load status is compared segment by segment with the preset load protection target to obtain the load protection comparison result; based on the load protection comparison result, triggerable protection segments are selected; and load protection data is recorded according to the triggerable protection segments.

Images