Multi-energy complementary optimization scheduling method and terminal in extreme source load imbalance scene

Abstract

The invention discloses a multi-energy complementary optimization scheduling method and terminal in an extreme source-load imbalance scene, and the method comprises the steps: carrying out the dimension reduction of multi-source output features through principal component analysis, reducing the redundancy, constructing features which can be used for clustering and prediction, recognizing a core scene through time sequence clustering, and enabling a model to turn to scene scheduling. Respectively capturing a long-term trend and a short-term fluctuation through a multi-time scale prediction model, and generating high-precision prediction data; on this basis, an extreme source load unbalance index is calculated, and whether extreme conditions such as peak shaving gaps and power abandoning risks exist or not is recognized. Once an extreme scene is identified, prediction errors are embedded into a scheduling model in a robust constraint form by constructing worst-case uncertainty intervals of wind, light, hydropower and load, so that the scheduling feasibility is improved. And finally, constructing a ten-day-level medium-term and intra-day short-term two-layer scheduling model, and forming an overall optimal dual-time-scale scheduling scheme under the support of an improved genetic algorithm. In this way, the stability and the performability of the optimized scheduling scheme are improved.

Description

Technical Field

[0001] This invention relates to the technical field of power dispatching, and in particular to a multi-energy complementary optimization dispatching method and terminal under extreme source-load imbalance scenarios. Background Technology

[0002] Currently, the high proportion of renewable energy connected to the grid has exacerbated the contradictions in the operation of the power system, placing higher demands on the operation and regulation of hydropower stations and pumped storage stations: wind power exhibits the characteristics of "peak at night in winter and trough at midday in summer," while photovoltaic output is concentrated during the day. Both wind and photovoltaic (hereinafter collectively referred to as "wind and solar") are significantly affected by extreme weather, and are prone to sudden increases or decreases in output, requiring rapid responses from hydropower stations and pumped storage stations to smooth out output fluctuations; conventional hydropower stations rely on inflow runoff for output, and during the high-water season, they are prone to water wastage due to insufficient absorption capacity, while during the dry season, water volume is limited. The system's regulation leads to a decrease in peak-shaving capacity, and there is a lack of effective coordination mechanisms among hydropower stations with different regulation capabilities (seasonal, annual, weekly, and daily regulation), making it difficult to form a joint force to meet system demands. Electricity load exhibits a time-varying pattern of "summer midday peak and winter nighttime trough," which is mismatched with the characteristics of wind and solar power output. In extreme scenarios, hydropower stations and pumped storage stations need to undertake heavier peak-shaving tasks. If the regulation capacity is insufficient, it can easily lead to large-scale energy curtailment (wind, solar, and hydropower curtailment), a surge in peak-shaving pressure, and even system frequency fluctuations, threatening power security.

[0003] While existing research on multi-energy complementary dispatching of hydropower stations and pumped storage power stations has made some progress in source-load matching and improved operational economy, significant shortcomings remain. First, most studies focus on dispatching optimization under normal operating conditions, lacking targeted design for extreme source-load imbalance scenarios. They also lack systematic analysis of key issues such as sudden changes in wind and solar output caused by extreme weather and the accumulation of water inflow prediction errors, making it difficult to ensure the effective regulation of hydropower stations and pumped storage power stations under extreme scenarios. Second, most studies adopt a single-timescale dispatching model, failing to establish a long-, medium-, and short-term multi-timescale coordination mechanism, making it difficult to simultaneously consider both long-term operational economy and short-term real-time dispatching for hydropower stations and pumped storage power stations. Third, they do not fully consider the prediction errors of wind and solar output, power load, and hydropower station water inflow. Traditional deterministic dispatching schemes have weak anti-interference capabilities under extreme conditions, easily leading to a disconnect between hydropower station output plans, pumped storage power station charging and discharging plans, and actual system demands, resulting in execution deviations. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a multi-energy complementary optimization scheduling method and terminal under extreme source-load imbalance scenarios, which can accurately identify extreme scenarios, coordinate multiple time scales, and resist prediction uncertainties during the scheduling process.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-energy complementary optimization scheduling method for extreme source-load imbalance scenarios includes the following steps: S1. Calculate hydropower output data by collecting historical wind and solar power output data, load power data and hydropower inflow data, extract multi-dimensional output features of hydropower output data, and use principal component analysis to reduce the dimensionality of the output features; S2. Use temporal clustering algorithm to classify the dimensionality-reduced output features into scenarios, and obtain the core scenarios based on the classification results; S3. Based on the core scenario, establish a multi-time-scale prediction model, input historical wind and solar power output data, load power data and hydropower inflow data into the multi-time-scale prediction model for prediction, and obtain the prediction dataset. S4. Calculate the corresponding extreme source load imbalance index based on the predicted dataset, and determine whether an extreme source load imbalance scenario has occurred based on the calculation results. If so, proceed to step S5. S5. Based on the type of extreme source-load imbalance scenario, construct the worst-case uncertainty interval of wind and solar power output, hydropower inflow and load demand, and embed the uncertainty interval into power balance constraints, reserve constraints and grid transmission constraints to form a robust set of feasible constraints. S6. Based on the feasible constraint set, construct a ten-day mid-term scheduling optimization model, with the objectives of maximizing economic efficiency, maximizing hydropower peak-shaving contribution, and minimizing joint output fluctuation, to obtain the mid-term scheduling scheme and system operation boundary. S7. Construct an intraday short-term scheduling optimization model based on the system operation boundary, and add real-time network-side interaction constraints to the intraday short-term scheduling optimization model; S8. An improved genetic algorithm is used to solve the dual-time-scale model composed of the ten-day mid-term scheduling optimization model and the intraday short-term scheduling optimization model to obtain an optimized scheduling scheme.

[0006] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A multi-energy complementary optimization scheduling terminal for extreme source-load imbalance scenarios includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the multi-energy complementary optimization scheduling method for extreme source-load imbalance scenarios described above.

[0007] The beneficial effects of this invention are as follows: First, principal component analysis is used to reduce the dimensionality of multi-source power output features, reducing redundancy and constructing features that can be used for clustering and prediction. Then, time-series clustering is used to identify core scenarios, enabling the model to shift towards scenario-based scheduling. Next, a multi-timescale prediction model is used to capture long-term trends and short-term fluctuations, generating high-precision prediction data. Based on this, extreme source-load imbalance indices are calculated to identify extreme situations such as peak-shaving gaps and power curtailment risks. Once extreme scenarios are identified, the prediction error is embedded into the scheduling model as a robust constraint by constructing worst-case uncertainty intervals for wind, solar, hydropower, and load, improving scheduling feasibility. Finally, a two-layer scheduling model—a ten-day mid-term model and an intraday short-term model—is constructed, forming an overall optimal dual-timescale scheduling scheme with the support of an improved genetic algorithm.

[0008] This approach enables intelligent identification and robust response to extreme source-load imbalance scenarios, maintaining the stability and executability of dispatching schemes in complex environments such as severe wind and solar power fluctuations, uncertain hydropower inflows, and sudden load changes. Multi-timescale forecasting significantly improves the accuracy of load and output trend predictions; robust optimization ensures that even in the worst-case scenario, engineering constraints such as power balance, reserve capacity, and grid transmission are met; and a dual-timescale model simultaneously considers global dispatching efficiency and real-time operational requirements. Ultimately, this reduces energy curtailment, enhances peak-shaving capacity, and strengthens system operational safety under extreme conditions, transforming multi-energy complementary dispatching from experience-based control to a verifiable, predictable, and optimizable adaptive dispatching mechanism. Attached Figure Description

[0009] Figure 1 This is a flowchart of a multi-energy complementary optimization scheduling method under an extreme source-load imbalance scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a multi-energy complementary optimization scheduling terminal under an extreme source-load imbalance scenario according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the topology of a wind-solar-hydro-storage multi-energy complementary system according to an embodiment of the present invention; Figure 4 This is a flowchart of the extreme scenario determination steps in an embodiment of the present invention; Figure 5 This is a flowchart of the dual-time-scale robust optimization scheduling process according to an embodiment of the present invention.

[0010] Label Explanation: 1. A multi-energy complementary optimization scheduling terminal for extreme source-load imbalance scenarios; 2. Memory; 3. Processor. Detailed Implementation

[0011] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0012] Existing energy management methods involving hydropower and pumped storage power stations often focus only on simple power source-energy storage coordination, neglecting precise matching of power load variations over time and adequate consideration of grid interaction constraints. Furthermore, their scheduling algorithms lack adaptability to non-convex constraints, failing to meet the comprehensive requirements of high-proportion renewable energy systems for safe and stable operation, efficient resource utilization, and economic cost control. Therefore, there is an urgent need for an optimized scheduling method for hydropower and pumped storage power stations that can accurately identify extreme scenarios, coordinate multiple time scales, and withstand predictive uncertainties.

[0013] To at least solve the above problems, please refer to Figure 1 This invention provides a multi-energy complementary optimization scheduling method for extreme source-load imbalance scenarios, including the following steps: S1. Calculate hydropower output data by collecting historical wind and solar power output data, load power data and hydropower inflow data, extract multi-dimensional output features of hydropower output data, and use principal component analysis to reduce the dimensionality of the output features; S2. Use temporal clustering algorithm to classify the dimensionality-reduced output features into scenarios, and obtain the core scenarios based on the classification results; S3. Based on the core scenario, establish a multi-time-scale prediction model, input historical wind and solar power output data, load power data and hydropower inflow data into the multi-time-scale prediction model for prediction, and obtain the prediction dataset. S4. Calculate the corresponding extreme source load imbalance index based on the predicted dataset, and determine whether an extreme source load imbalance scenario has occurred based on the calculation results. If so, proceed to step S5. S5. Based on the type of extreme source-load imbalance scenario, construct the worst-case uncertainty interval of wind and solar power output, hydropower inflow and load demand, and embed the uncertainty interval into power balance constraints, reserve constraints and grid transmission constraints to form a robust set of feasible constraints. S6. Based on the feasible constraint set, construct a ten-day mid-term scheduling optimization model, with the objectives of maximizing economic efficiency, maximizing hydropower peak-shaving contribution, and minimizing joint output fluctuation, to obtain the mid-term scheduling scheme and system operation boundary. S7. Construct an intraday short-term scheduling optimization model based on the system operation boundary, and add real-time network-side interaction constraints to the intraday short-term scheduling optimization model; S8. An improved genetic algorithm is used to solve the dual-time-scale model composed of the ten-day mid-term scheduling optimization model and the intraday short-term scheduling optimization model to obtain an optimized scheduling scheme.

[0014] As described above, the beneficial effects of this invention are as follows: First, principal component analysis is used to reduce the dimensionality of multi-source power output features, reducing redundancy and constructing features that can be used for clustering and prediction. Then, time-series clustering is used to identify core scenarios, enabling the model to shift towards scenario-based scheduling. Next, a multi-timescale prediction model is used to capture long-term trends and short-term fluctuations, generating high-precision prediction data. Based on this, extreme source-load imbalance indices are calculated to identify extreme situations such as peak-shaving gaps and power curtailment risks. Once extreme scenarios are identified, the prediction error is embedded into the scheduling model in a robust constraint form by constructing worst-case uncertainty intervals for wind, solar, hydropower, and load, improving scheduling feasibility. Finally, a two-layer scheduling model, consisting of a ten-day mid-term model and an intraday short-term model, is constructed, forming an overall optimal dual-timescale scheduling scheme with the support of an improved genetic algorithm.

[0015] This approach enables intelligent identification and robust response to extreme source-load imbalance scenarios, maintaining the stability and executability of dispatching schemes in complex environments such as severe wind and solar power fluctuations, uncertain hydropower inflows, and sudden load changes. Multi-timescale forecasting significantly improves the accuracy of load and output trend predictions; robust optimization ensures that even in the worst-case scenario, engineering constraints such as power balance, reserve capacity, and grid transmission are met; and a dual-timescale model simultaneously considers global dispatching efficiency and real-time operational requirements. Ultimately, this reduces energy curtailment, enhances peak-shaving capacity, and strengthens system operational safety under extreme conditions, transforming multi-energy complementary dispatching from experience-based control to a verifiable, predictable, and optimizable adaptive dispatching mechanism.

[0016] Further, in step S4, calculating the corresponding extreme source-load imbalance index based on the predicted dataset includes: Based on the predicted dataset, we calculate the supply and demand imbalance index, instantaneous shock index, flexibility margin index, and coupling risk index. The entropy weight method is used to assign weight values ​​to each indicator and normalize the values ​​of each indicator. The overall risk value is obtained by accumulating the product of each indicator and its weight value; A threshold calibration model is constructed using gradient boosting trees based on meteorological factors, power grid operation factors, hydropower regulation performance factors, load structure factors, and pumped storage capacity factors. The dynamic thresholds of the normalized indicators are calibrated using the threshold calibration model to obtain the early warning thresholds corresponding to each indicator and the comprehensive risk threshold coupled with all indicators.

[0017] As described above, the entropy weight method is used to automatically calculate the index weights, ensuring that the weight allocation depends on the historical data's ability to distinguish extreme situations, thus improving objectivity. After normalization, the indicators are further used to calculate a comprehensive risk value, reflecting the overall risk level of the system. Simultaneously considering multiple factors such as meteorology, grid operation, hydropower regulation capacity, load structure, and pumped storage capacity, a gradient boosting tree is used to construct a threshold calibration model, enabling dynamic adjustment of the indicator warning thresholds and adapting the judgment criteria to seasonal, meteorological, and structural changes in the power system.

[0018] In this way, through multi-dimensional indicators and machine learning-driven threshold calibration, high-precision judgment of extreme source-load imbalance scenarios is achieved. This eliminates the system's reliance on static thresholds or human experience, allowing it to update risk limits in real time based on weather changes, equipment status, and power structure. This significantly improves the accuracy and recall rate of extreme scenario identification, reduces false positives and false negatives, and enables the dispatching system to identify high-risk periods in advance, providing a reliable basis for robust dispatching. In this embodiment, the threshold calibration model allows the system to adapt to different seasons (such as dry seasons and peak wind and solar seasons) and different operating conditions, improving the overall system's safety and adaptability.

[0019] Furthermore, in step S4, determining whether an extreme source-load imbalance scenario has occurred based on the calculation results includes: When the preset first number of normalized indicators are greater than the corresponding warning threshold, it is judged as a general extreme source-load imbalance scenario. When the preset second number of normalized indicators exceeds the corresponding warning threshold, it is judged as a severe extreme source-load imbalance scenario; When the number of normalized indicators exceeding the preset second number is greater than the corresponding warning threshold, it is determined to be an extremely severe source-load imbalance scenario.

[0020] As described above, the automated classification of extreme scenarios has been implemented, enabling the dispatch system to take robust measures of varying strengths based on risk level, such as increasing reserve capacity, enhancing external transmission redundancy, or strengthening pumped storage regulation. This hierarchical mechanism enhances the system's risk adaptability, allowing dispatch schemes to dynamically adjust robustness intervals and optimization objectives, thereby achieving a better balance between economy and safety. This mechanism significantly improves the dispatch system's response to multi-dimensional coupled extreme events, making power system operation under extreme conditions more controllable, reliable, and in line with engineering realities.

[0021] Further, step S5 includes: Based on the time information in the predicted dataset, the type of the extreme source-load imbalance scenario is determined to be either an extreme peak-shaving gap or an extreme power curtailment risk. If the scenario is determined to be an extreme peak-shaving gap, then the worst-case extreme data setting is performed based on the predicted dataset:

[0022]

[0023] In the formula, P W_P,worst ( t ( ) represents the worst power output data for weather during time period t. P L,worst ( t () represents the worst-case load data for time period t. P W_P,pre ( t ( ) represents the predicted wind and solar power output data for time period t. P L,pre ( t ( ) represents the load forecast data for time period t. Contribute to the uncertainties of landscape photography To meet the output demand of uncertain loads; Robust constraints for extreme peak shaving gap scenarios specifically include the following constraints: Power balance robust constraints:

[0024] In the formula, P W_P (t)(1 Δ P W_P (To contribute to the worst-case scenario of wind and solar power) P H_max (t) represents the maximum available hydropower output; P L (t)(1+ΔP L This represents the worst-case load demand. Robust constraints for hydropower backup:

[0025] In the formula, P H (t) represents the actual hydropower output during dispatch; α This is a reserve factor used to ensure that hydropower can retain output to cope with unexpected load increases; If the risk is assessed as extreme, then:

[0026]

[0027] In the formula, P W_P ( t This data represents the combined contribution of wind and solar power. Power balance robust constraints:

[0028] In the formula, P W_P (1+Δ P W_P (P) represents the worst-case scenario for the wind and solar power output. L (1-ΔP L This represents the worst-case scenario for the load. P G_max (t) represents the maximum power transmitted from the power grid; Power grid transmission constraints:

[0029] In the formula, P G ( t ) represents the power transmitted from the grid; β is the safety factor, used to avoid grid stability problems caused by overload of the transmission channel.

[0030] As described above, robust constraints are established according to different scenarios: for extreme peak-shaving gap scenarios, the lower limit of wind and solar power output and the upper limit of load are taken, plus hydropower reserve constraints; while for extreme curtailment risk scenarios, the upper limit of wind and solar power output and the lower limit of load are taken, plus grid transmission safety constraints. In this way, we can cope with the risks of different extreme scenarios and improve the reliability of subsequent dispatching.

[0031] Further, in step S6, a ten-day mid-term scheduling optimization model is constructed based on the aforementioned feasible constraints, with multiple objectives including maximizing economic efficiency, maximizing the contribution of hydropower peak shaving, and minimizing the fluctuation of combined output, including: The objective function formula for the ten-day mid-term scheduling optimization model is as follows:

[0032] In the formula, The smaller the value, the higher the economic benefit; The smaller the value, the greater the contribution of hydropower. The smaller the value, the lower the volatility; T 1 represents the scheduling cycle; C L ( t ( ) represents the electricity consumption benefit during time period t; C W_P ( t )and C H ( t The costs for wind and solar power and hydropower operation and maintenance during time period t are respectively. C Ab ( t ) represents the penalty cost for energy wastage during time period t; C G ( t() represents the power grid interaction benefits during time period t; P L ( t (t) represents the power system load demand during time period t; P W_P ( t )and P H ( t ) represents the combined wind and solar power output and hydropower output during time period t; μ for T Average combined output within 1.

[0033] As described above, by constructing a ten-day-level medium-term scheduling optimization model based on robust feasible constraints, and taking the maximization of economic efficiency, the maximization of hydropower peak-shaving contribution, and the minimization of joint output fluctuation as multiple objectives, it is possible to simultaneously optimize the cost structure, peak-shaving capacity, and operational stability of the power system on a medium- to long-term scale, thereby significantly improving the comprehensive operational performance of the multi-energy complementary system under extreme source-load imbalance conditions.

[0034] First, prioritizing maximum economic efficiency can effectively reduce the overall operation and maintenance costs and system electricity purchase and sale costs of multiple energy sources such as wind, solar, hydro, and pumped storage, while ensuring safety. This enables the system to operate at low cost over a longer timescale, avoiding energy waste and redundancy caused by traditional experience-based dispatching. Second, maximizing the peak-shaving contribution of hydropower ensures that this key flexible resource is rationally planned and prioritized for use in medium-term dispatching. This fully leverages the advantages of hydropower's rapid response and cross-day adjustment capabilities for peak shaving, effectively alleviating the pressure on system stability caused by fluctuations in wind and solar power output and enhancing the system's ability to withstand extreme peak-shaving gaps. Finally, minimizing combined output fluctuations can promote a smoother combined output curve among wind, solar, hydro, and pumped storage, enabling the system to maintain stable power output in the medium term, reducing large-scale adjustment actions, improving the overall system's safety margin and equipment lifespan, and providing a smoother and more user-friendly operating boundary for subsequent intraday short-term dispatching.

[0035] In summary, this multi-objective medium-term dispatch model achieves a balance between economy and security, and simultaneously improves peak-valley regulation capability and stability. It enables the power system to obtain a coordinated, cost-controllable, and stable optimized dispatch strategy even under extreme source-load imbalance scenarios, laying a core foundation for a robust dispatch system with dual time scales and significantly enhancing the system's reliability and flexibility.

[0036] Furthermore, the constraints of the ten-day mid-term scheduling optimization model are as follows: The total power output from grid interconnection, wind and solar power, and pumped storage power should equal the load power, expressed by the formula: P G ( t ) + PW_P ( t ) + P H ( t ) = P L ( t ) In the formula, P G ( t ) represents the power transmitted from the power grid during time period t; The output of wind power and photovoltaic power generation cannot exceed the upper and lower limits, which can be expressed by the formula: P W_P_min ( t )≤ P W_P ( t )≤ P W_P_max ( t ) In the formula, P W_P_max ( t () represents the maximum output of scenery during time period t. P W_P_min ( t () represents the minimum power output of the wind and solar power during time period t; Hydropower output needs to be calculated by combining regulation characteristics and water condition forecasts. The hydropower output must meet the upper and lower limits of power, as expressed by the formula: P H_min ( t )≤| P H ( t )|≤ P H_max ( t ) In the formula, P H_max ( t ) represents the maximum hydropower output during time period t. P H_min ( t ) represents the minimum hydropower output during time period t; Pumped storage power stations cannot exceed the upper and lower limits of their adjustable pumped storage capacity. The constraint formula is expressed as follows:

[0037] In the formula, E ps ( t (This refers to the pumped storage regulation capacity.) E ps,min ( t This is the minimum capacity for pumped storage regulation.E ps,max ( t () represents the maximum capacity for pumped storage regulation; Grid interaction cannot exceed the upper and lower limits of grid interaction power, as shown in the following formula: P G_min ( t )≤| P G ( t )|≤ P G_max ( t ) In the formula, P G_min ( t () represents the minimum power transmitted from the power grid.

[0038] As described above, by setting multi-dimensional operational constraints such as grid interaction, wind and solar power output, hydropower output, pumped storage capacity, and grid interaction power in the ten-day mid-term dispatch optimization model, it can be ensured that the generated mid-term dispatch scheme fully conforms to the engineering operation boundary and physical characteristics of the power system, thereby significantly improving the feasibility, stability, and security of the dispatch results.

[0039] First, power balance constraints ensure that the total output of wind, solar, hydro, pumped storage, and grid interaction remains consistent with load demand in each time period, avoiding false optimization results due to the mid-term dispatch model ignoring real-time supply and demand balance, thus ensuring the feasibility of the dispatch scheme. Second, upper and lower limit constraints on wind and solar power output, combined with unit capacity and meteorological conditions, prevent infeasible decisions that exceed unit capacity requirements during mid-term dispatch, improving the reliability of new energy output forecasts and the physical feasibility of dispatch schemes. Third, upper and lower limit constraints on hydropower output are linked to water situation forecasts, ensuring that hydropower operates within the limits of water level, water volume, and gate capacity, avoiding over-assumptions on reservoir regulation capacity, and effectively guaranteeing that hydropower can still play a major role in peak shaving under extreme source and load conditions.

[0040] Furthermore, the upper and lower limits of pumped storage power station capacity ensure that its charging and discharging behavior does not exceed the equipment's design energy capacity and power capability, avoiding excessive pumped storage regulation or distortion of inter-day regulation capability, thereby ensuring that pumped storage plays its true energy storage regulation value in medium-term dispatch. Finally, the upper and lower limits of grid interaction power constraints ensure that the power transmitted or received does not exceed the thermal stability limit and safety margin of the channel, fundamentally avoiding the risk of grid overload and improving the system's safe operation capability under large-scale renewable energy fluctuations.

[0041] In summary, these constraints collectively construct the engineering feasibility boundary of the medium-term dispatch model, ensuring that the optimized dispatch scheme is cost-effective and provides sufficient peak shaving while strictly conforming to the operating rules of the power system and equipment constraints. This significantly improves the reliability, feasibility, and safety controllability of the dispatch strategy, providing a robust and reliable operating boundary for subsequent intraday short-term dispatch.

[0042] Furthermore, the real-time network-side interaction constraints are as follows:

[0043] In the formula, P G_min ( t ) represents the minimum power transmitted from the power grid during time period t. P G_max ( t () represents the maximum power transmitted by the power grid during time period t. E G ( t ( ) represents the real-time interactive energy of the power grid at time t. E G_min ( t () represents the lower limit of real-time interactive energy of the power grid at time t. E G_max ( t ) represents the upper limit of real-time interactive energy of the power grid at time t.

[0044] As described above, by incorporating real-time grid-side interactive constraints into the intraday short-term dispatch optimization model, ensuring that the total system output does not exceed its safe operating limits at any given time, the optimized dispatch results strictly conform to the real-time carrying capacity and operating boundaries of the actual power grid. This constraint enables short-term dispatch to respond promptly to real-time disturbances such as rapid fluctuations in wind and solar power output, sudden load changes, and power flow variations, preventing infeasible solutions such as overload, exceeding limits, or violating power flow direction from occurring in the dispatch results. This significantly improves the engineering feasibility and operational reliability of the dispatch model. By limiting the total system output to operate within the real-time boundary range, the risks of line overload, voltage exceeding limits, and system instability can be effectively reduced, ensuring the safe and stable operation of the power grid even under highly uncertain scenarios such as high-proportion renewable energy grid connection and extreme source-load imbalance. Simultaneously, this constraint can also reduce equipment losses caused by frequent start-ups and large-scale adjustments, improve the precision and economy of dispatch, and achieve more efficient and safer real-time optimized control.

[0045] Furthermore, an optimized scheduling scheme is obtained, which then includes: The optimized scheduling scheme is verified by the indicators of energy curtailment rate and external transmission channel utilization rate. If the verification fails, the process returns to steps S6 and S7 to adjust the parameters and re-execute step S8.

[0046] The energy curtailment rate is calculated as follows:

[0047] In the formula, P W_P_max ( t () indicates the maximum solar power output during time period t. P H_max ( t () indicates the maximum hydropower output during time period t. P W_P ( t () represents the solar power output during time period t. P H ( t () represents the hydropower output during time period t. This represents the pumped storage output of the pumped storage power station during time period t. The calculation method for the utilization rate of the external delivery channel is as follows:

[0048] In the formula, P G ( t () represents the power transmitted from the power grid during time period t. P G_max ( t ) represents the maximum power transmitted from the power grid during time period t.

[0049] As described above, the scheduling results need to be validated, such as by measuring the utilization rate of the power grid's transmission channels and the energy curtailment rate, in order to quantify the beneficial effects based on the original control model.

[0050] Please refer to Figure 2 Another embodiment of the present invention provides a multi-energy complementary optimization scheduling terminal for extreme source-load imbalance scenarios, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the above-described multi-energy complementary optimization scheduling method for extreme source-load imbalance scenarios.

[0051] The multi-energy complementary optimization scheduling method and terminal for extreme source-load imbalance scenarios described above are applicable to accurately identifying extreme scenarios, coordinating multiple time scales, and resisting prediction uncertainties during the scheduling process. The following is a detailed description of the implementation methods: Please refer to Figure 1 One embodiment of the present invention is as follows: This embodiment constructs a robust optimization model with two time scales: a ten-day mid-term model and a daily short-term model, achieving a balance between global economic efficiency and local real-time performance. Improved algorithms and robust optimization strategies are introduced to enhance the scheduling scheme's resilience to prediction errors and extreme weather, ensuring the system's safe and stable operation under extreme scenarios. This is based on the logical unfolding of extreme scenario determination, dual-time-scale modeling, algorithm solution, and plan implementation. Please refer to the schematic diagram of the wind-solar-hydro-storage multi-energy complementary system topology in this embodiment. Figure 3 .

[0052] Specifically, the following steps are included: S1. Calculate hydropower output data by collecting historical wind and solar power output data, load power data and hydropower inflow data, extract multi-dimensional output features of hydropower output data, and use principal component analysis to reduce the dimensionality of the output features.

[0053] First, historical solar power output data, load data, and hydropower inflow data are collected. From these data, along with hydropower output data calculated using mathematical models, multi-dimensional features are extracted, including basic energy characteristics, peak and trough characteristics, slope and variation characteristics, and time-period distribution characteristics. After feature extraction, standardization is performed to eliminate dimensional differences, followed by dimensionality reduction using Principal Component Analysis (PCA). As a classic linear dimensionality reduction algorithm, PCA maps high-dimensional features to a low-dimensional principal component space through orthogonal transformation. The first principal component retains the maximum variance information, and subsequent principal components retain the second largest variance and are orthogonal to the preceding ones. In energy forecasting, multi-dimensional features are correlated and prone to redundancy. PCA calculates the covariance matrix, solves for eigenvalues ​​and eigenvectors, and selects the top k principal components with a cumulative variance contribution rate of 85%–95%, effectively eliminating redundant information, reducing dimensionality, and simultaneously reducing model computational costs and overfitting risks, providing efficient input features for scene recognition.

[0054] S2. Use temporal clustering algorithm to classify the output features after dimensionality reduction, and obtain the core scene based on the classification result.

[0055] Among them, the temporal clustering algorithm is the temporal K-means algorithm. This algorithm is used to identify core scenarios. Specifically, it constructs environmental variables by incorporating historical weather data such as wind speed and light intensity, thereby integrating curve patterns with weather factors to improve prediction accuracy. Unlike traditional K-means, which focuses on static distribution, temporal K-means introduces temporal distance metrics such as dynamic time warping distance (DTW) to accurately capture the temporal evolution patterns of data and avoid clustering bias that ignores temporal correlations. This algorithm can cluster massive amounts of energy data into typical core scenarios such as high wind and solar power output and extreme weather fluctuations. After incorporating environmental variables, the scenario features integrate meteorological influences, making them more consistent with actual operating conditions and providing representative samples for multi-timescale predictions.

[0056] S3. Based on the core scenario, establish a multi-time-scale prediction model, input historical wind and solar power output data, load power data and hydropower inflow data into the multi-time-scale prediction model for prediction, and obtain the prediction dataset.

[0057] In this embodiment, the multi-timescale prediction model consists of a long-term prediction model and a short-term prediction model. The long-term prediction model adopts the Transformer Neural Network (TN) prediction model. The Transformer prediction model is based on a self-attention mechanism, which improves the efficiency of long sequence processing through parallel computing, accurately captures long-distance temporal dependencies, and adapts to the needs of capturing long-period fluctuations. In the prediction process, the scene features, environmental variables, and other inputs are first embedded and encoded at their positions: embedding maps the features to low-dimensional dense vectors, and position encoding supplements the temporal information through sine-cosine functions to compensate for the lack of time-series perception due to the absence of a cyclic structure. The model is centered on a decoder and contains multiple layers of self-attention sublayers and feedforward neural network sublayers. The self-attention sublayers use a masked attention mechanism to prevent the leakage of future information. The sequence is generated element by element through autoregressive iteration. Based on the probability distribution of historical inputs and the output of generated segments, the optimal element is selected through sampling until the prediction is completed, effectively capturing long-period patterns such as seasonal changes and annual fluctuations in watershed inflows.

[0058] For short-term prediction models, the LSTM (Long Short-Term Memory) algorithm is considered. As an improved RNN, it addresses the vanishing and exploding gradient problems through a gating mechanism, adapting to the need for capturing high-frequency fluctuations. The core gating mechanism consists of an input gate, a forget gate, and an output gate, combined with Sigmoid and tanh activation functions to achieve selective information processing: the forget gate filters out redundant historical information, the input gate updates effective features to the cell state, and the output gate controls the information transmission ratio. Long-term dependencies are captured through collaborative updates of the "cell state-hidden state." The cell state stores key historical information, and the hidden state integrates information to output the single-step prediction result. In multi-step prediction, the predicted results are concatenated with historical sequences and iteratively input, accurately capturing short-term dynamic characteristics and response features such as minute-level surges and drops in wind power and intraday load fluctuations, meeting the requirements of timeliness and accuracy.

[0059] S4. Calculate the corresponding extreme source-load imbalance index based on the predicted dataset, and determine whether an extreme source-load imbalance scenario has occurred based on the calculation results. If so, proceed to step S5.

[0060] Please refer to Figure 4The core of this step is, firstly, to construct a multi-dimensional quantitative indicator system for extreme scenarios based on the seasonal output characteristics of wind and solar energy and the time-dependent load demand of the power system. This provides a precise basis for scenario identification in optimizing the scheduling of hydropower stations. Specifically, the seasonal output characteristics of wind and solar energy refer to the output patterns of wind and solar power generation influenced by seasonal changes in natural weather conditions; the time-dependent load demand refers to the demand patterns formed by changes in power load throughout the day and seasonally; and the regulation capacity of pumped storage power stations varies with weather conditions and seasonal floods and droughts. The dynamic matching relationship among these three factors is the core trigger for the formation of extreme scenarios; therefore, it is necessary to accurately define the boundaries of these scenarios through multi-dimensional quantitative indicators.

[0061] (1) Peak shaving gap rate γ gap ( t The peak load (T) is a core quantitative indicator that measures the gap between the power system's load demand and the maximum available power supply within time period t. It is used to determine extreme scenarios where the system's peak load capacity is insufficient. Its calculation formula is as follows:

[0062] The parameters in the formula are defined as follows: P L ( t ): Total load demand of the power system during time period t (unit: MW), covering the sum of various types of electricity loads such as industrial, commercial, and residential, reflecting the minimum scale of power demand of the system during this time period; P W_P_max ( t ): Maximum wind and solar power output during time period t (unit: MW) refers to the maximum active power that wind turbine generators and photovoltaic arrays can output under meteorological conditions (wind speed, solar intensity) during time period t, which is the maximum supply potential of renewable energy; P H_max ( t ): Maximum hydropower output during time period t (unit: MW) refers to the maximum active power that a hydropower station can output under full-load conditions based on current inflow, reservoir water level and other hydrological conditions during time period t. It is the core flexible resource for system peak regulation.

[0063] P ps,dis ( t ): Maximum discharge output (MW, including regulating capacity constraints) of pumped storage power station during time period t.

[0064] P G_max ( t ): The maximum power transmitted from the power grid during time period t (MW, subject to channel capacity limitations).

[0065] The physical meaning of this indicator is as follows: the numerator represents the difference between the load demand and the maximum supply capacity of wind, solar, and hydropower during time period t; the denominator is normalized by load demand to make the indicator comparable across time periods and load scales. γ gap ( t When the load exceeds a preset threshold by 10%, it is considered an extreme peak-shaving gap scenario. This scenario means that the system's maximum power supply capacity cannot meet the load demand, which may lead to risks such as power shortages and power rationing. The gap needs to be alleviated by scheduling measures such as increasing the output of hydropower stations and reducing non-rigid loads.

[0066] curtailment risk rate γ ab ( t ):

[0067] when γ ab ( t When the percentage is greater than 15%, it is classified as an "extreme power curtailment risk scenario"; In the formula: P W_P,max ( t ( ) represents the maximum potential output of the scenery during time period t. P H_max ( t This represents the maximum output of the hydropower plant. P ps,cha ( t ) represents the maximum charging capacity (MW) of the pumped storage power station during time period t. P Gn ( t Rated capacity (MW) of power grid transmission channels.

[0068] In addition, the threshold needs to be dynamically adjusted according to the season and time period. For example, in winter, the focus is on the extreme curtailment risk because there is a wind power peak and a load trough, so the curtailment risk rate threshold is adjusted. In summer, the focus is on the extreme peak shaving gap because there is a load peak and a sudden drop in photovoltaic power at noon, so the peak shaving gap rate is adjusted.

[0069] (2) While considering the indicators of the degree of imbalance between supply and demand, it is also necessary to consider the relevant indicators of the impact intensity in time series in order to characterize the dynamic change risk of energy data and quantify the instantaneous impact and cumulative effect of extreme scenarios.

[0070] sudden change rate of output R RE,shock,t :

[0071] In the formula: PW_P,rated For the total installed capacity of wind and solar power, when R RE,shock,t When the impact is greater than 15%, it is considered a sudden change in wind and light conditions. P W_P ( t () represents the actual contribution of the scenery during the t-period.

[0072] Load mutation rate R L,shock,t :

[0073] In the formula: P L,avg The average load for the day, when R L,shock,t When the load is greater than 10%, it is considered a sudden load change shock. P L ( t ) represents the load demand of the power system during time period t.

[0074] Extreme duration: T extreme It is the number of consecutive time periods in which a single imbalance indicator exceeds the warning threshold, expressed in hours, and must meet the following conditions. T extreme >3, in order to avoid misjudging instantaneous fluctuations as extreme scenarios.

[0075] (3) In addition, the system flexibility margin index should also be considered to characterize the system’s ability to operate below extreme scenarios. The lower the margin, the higher the risk after extreme scenarios occur.

[0076] Hydropower reserve margin M HYD,spare,t :

[0077] Hydropower reserve margin is a key indicator for hydropower stations during the dry season, reflecting their reserve capacity to cope with sudden load changes. In the formula, P HYD,plan,t Contribute to the hydropower project, P HYD,max,t For the maximum output of hydropower, when M HYD,spare,t When the peak load is less than 5%, the system lacks the flexibility to cope with the peak load gap.

[0078] Pumping regulation margin M PS,t :

[0079] Pumped storage dispatch margin reflects the remaining capacity of pumped storage power stations for peak shaving and valley filling. Where:P PS,rated This is the rated capacity of the pumped storage power station; P ps,dis ( t ) represents the maximum discharge output of the pumped storage power station during time period t; P ps,cha (t) represents the maximum charging capacity of the pumped storage power station during time period t. M PS,t When the level is less than 5%, the storage capacity is exhausted, and the imbalance problem cannot be alleviated.

[0080] Power grid transmission margin M Grid,t :

[0081] The power grid's transmission margin reflects the ability of transmission channels to absorb abandoned power. M Grid,t A value less than 5% indicates a blockage in the power transmission channel, exacerbating the risk of power curtailment. In the formula, P G ( t () represents the power transmitted from the power grid during time period t. P G_max ( t ) represents the maximum power transmitted from the power grid during time period t.

[0082] (4) Extreme scenarios are often not just one extreme situation, but are caused by the superposition of wind and solar fluctuations, load changes and grid constraints. In this case, it is necessary to establish coupling risk indicators to quantify the coupling risks.

[0083] Wind-solar-load coupling coefficient C RE-L,t :

[0084] The wind-solar-load coupling coefficient is used to measure the degree of superposition between a sudden drop in wind and solar load and a sudden increase in load. Where: R RE,shock,th and R L,shock,th These are the warning thresholds for sudden output change rate and sudden load change rate, respectively. C RE-L,t When the value is greater than 1.2, the coupling risk is significant.

[0085] Water-wind-solar coupling coefficient C HYD-RE,t :

[0086] The water-wind-solar coupling coefficient is used to measure the degree of overlap between insufficient water inflow and low wind and solar activity during the dry season. In the formula, Q HYD,tThe water flow rate during time period t. Q HYD,avg This is the historical average inflow rate for the same period. P RE,avg It contributes to the average output of the same period's scenery. When C HYD-RE,t When the value is greater than 1.5, the peak-shaving capacity decreases sharply.

[0087] Comprehensive flexible deficit coefficient C Flex,t :

[0088] The comprehensive flexibility deficit coefficient is used to quantify the coupled risk of simultaneous shortages of multiple flexible resources. Where: M HYD,spare,th , M PS,th , M Grid,th These are the threshold values ​​for the previously calculated hydropower reserve margin, pumped storage regulation margin, and grid transmission margin, respectively. When C Flex,t When the value is greater than 0.8, the system's overall flexibility is in deficit, and extreme risks increase dramatically.

[0089] After defining the indicators, weights need to be assigned to them using the entropy weighting method. Based on historical extreme scenario data, the information entropy of each indicator is calculated. The smaller the entropy value, the higher the indicator's distinguishability and the greater its weight, thus ensuring the objectivity of the weighting. Indicator normalization is then used to eliminate dimensional differences. An interval mapping method is employed to normalize the indicators to the [0,1] interval, preventing extreme value indicators from dominating the evaluation results. Indicators also need to be classified into positive and negative indicators. Positive indicators, such as the curtailment risk rate and coupling coefficient, indicate higher risk with larger values. Negative indicators, such as hydropower reserve margin and pumped storage regulation margin, indicate higher risk with smaller values.

[0090] Positive indicators:

[0091] Negative indicators: .

[0092] In the formula: x k,max and x k,min These are the historical maximum and minimum values ​​of the indicator, respectively.

[0093] After assigning and normalizing the weights, it is necessary to calibrate the dynamic thresholds of each indicator. The thresholds need to be dynamically adjusted considering multiple factors to achieve accurate adaptation. Considering five key influencing factors—meteorological factors, power grid operation factors, hydropower regulation performance factors, load structure factors, and pumped storage capacity factors—an XGBoost gradient boosting tree is used to construct a threshold calibration model. Historical extreme event data from the past five years are selected as training samples, and real extreme scenarios are labeled with their corresponding influencing factors and indicator thresholds. The model outputs warning thresholds and comprehensive risk thresholds for 12 indicators, taking the five influencing factors as input. Finally, 5-fold cross-validation is used to adjust parameters such as tree depth and learning rate to ensure the model's accuracy in predicting thresholds is ≥85%. Furthermore, the threshold adjustment must also meet the following rules: The meteorological department issued weather warnings: the warning threshold for peak shaving shortfall rate was lowered by 10%, and the warning threshold for power curtailment risk rate was raised by 15%. If the facility is under grid maintenance: the warning threshold for power transmission margin is lowered by 20%, and the warning threshold for power curtailment risk rate is lowered by 10%; If hydropower facilities are in the dry season: the early warning threshold for hydropower reserve margin is reduced by 30%, and the early warning threshold for coupling coefficient is reduced by 20%.

[0094] Based on the model's output, extreme scenarios need to be defined and categorized. If the normalized value of three or more indicators exceeds the dynamic threshold, an extreme scenario warning is triggered. When the system issues an extreme scenario warning, a comprehensive risk value calculation is required, using a weighted summation to determine the overall risk value. R total,t The formula is:

[0095] In the formula: w k For combined weights, Normalized index value; R total,t ∈[0,1], the larger the value, the higher the extreme risk.

[0096] The third step is to verify the coupling risk and duration. If the coupling risk index is greater than the threshold and the duration of the extreme scenario is not less than 3 hours, it is confirmed as a coupling extreme scenario.

[0097] Finally, based on R total,t Based on the verification results, extreme scenarios are divided into three levels, clearly defining the risk characteristics and response priorities for each level: Typical extreme value: 0.6 R total,t<0.75 indicates a single-dimensional imbalance, sufficient flexibility, and meets the requirements for single-indicator triggering, comprehensive risk value compliance, and the total number of coupled indicators exceeding one limit.

[0098] Severe extreme: 0.75 R total,t <0.9 indicates multi-dimensional imbalance and insufficient flexibility. The total number of single indicator triggers plus comprehensive risk value compliance and coupled indicators must be no less than two exceeding the standard.

[0099] Extreme weight: 0.9 R total,t The entire system is unbalanced, flexible resources are exhausted, and the system needs to meet the requirements of single indicator triggering plus comprehensive risk value compliance and no less than 3 coupled indicators exceeding the standard.

[0100] After determining the scenario, it is necessary to verify its effectiveness and iteratively correct it based on the determination results and the actual scenario, thus forming an operational closed loop. Recall and precision are calculated, and if both recall rates are less than 85%, the warning threshold for the coupled risk indicator is lowered; if precision is less than 85%, the overall risk threshold is raised. In addition, historical extreme scenario data is updated quarterly, and the threshold calibration model and weight matrix are retrained to adapt to dynamic changes in energy structure and power grid topology.

[0101] S5. Based on the type of extreme source-load imbalance scenario, construct the worst-case uncertainty interval for wind and solar power output, hydropower inflow and load demand, and embed the uncertainty interval into power balance constraints, reserve constraints and grid transmission constraints to form a robust set of feasible constraints.

[0102] This step requires the introduction of robust optimization processing to mitigate the impact of uncertainties in various output data on the stability of the scheduling results.

[0103] The robust optimization process adopted is as follows: First, based on historical power output data, worst-case deterministic data is generated to mitigate uncertainty. The wind and solar power output uncertainty (Δ) that has the most significant impact on scheduling is selected. P W_P ), uncertainty of hydropower inflow (Δ Q hydro ) and load uncertainty (Δ P L This process eliminates secondary interference factors and determines the value range based on historical prediction error statistics. Secondly, based on the sample detection and historical prediction error statistics in step S1, the fluctuation range of each parameter is determined and dynamically adjusted in conjunction with local seasonal characteristics.

[0104] After completing the worst-case data acquisition and uncertain parameter calculation, the above data and parameters are embedded into core constraints such as power balance and equipment processing limitations, thereby transforming them into worst-case deterministic constraints and ensuring the engineering feasibility of the scheduling scheme under extreme conditions.

[0105] If the scenario is determined to be an extreme peak-shaving gap, then the worst-case scenario data is set based on the obtained prediction data:

[0106]

[0107] In the formula, P W_P,worst ( t ( ) represents the worst power output data for weather during time period t. P L,worst ( t () represents the worst-case load data for time period t. P W_P,pre ( t ( ) represents the predicted wind and solar power output data for time period t. P L,pre ( t ( ) represents the load forecast data for time period t. Contribute to the uncertainties of landscape photography To meet the power output demand of uncertain loads.

[0108] Robust constraints for extreme peak shaving gap scenarios specifically include the following constraints: Power balance robust constraints:

[0109] In the formula: P W_P (t)(1 Δ P W_P (This represents the worst-case output of wind and solar power (taking the lower limit of the predicted value)). P H_max (t) represents the maximum available hydropower output (considering the lower limit of inflow); P L (t)(1+Δ P L ) represents the worst-case demand (taking the upper limit of the forecast value).

[0110] Robust constraints for hydropower backup: When water supply from hydropower is insufficient during the dry season, reserve capacity must be reserved to cope with sudden load changes. The formula is as follows:

[0111] In the formula, P H(t) represents the actual hydropower output during dispatch; This serves as a reserve factor to ensure that hydropower can retain output to cope with unexpected load increases.

[0112] If the scenario is judged to be an extreme risk of power curtailment, the core problem is power surplus. Robust constraints are needed to maximize the absorption of renewable energy and reduce curtailment, while simultaneously avoiding overload of pumped storage and grid transmission channels.

[0113]

[0114] The worst-case scenario of highest wind power output and lowest load needs to be considered to ensure that the total absorption capacity covers the wind and solar power output. The formula is:

[0115] In the formula, P W_P (1+Δ P W_P (P) represents the worst-case scenario for the wind and solar power output. L (1-ΔP L This represents the worst-case scenario for the load. P G_max (t) represents the maximum power transmitted from the power grid.

[0116] If the pumped storage capacity is insufficient, the pressure of power curtailment needs to be alleviated by transmitting power to other parts of the grid. It is essential to ensure that the transmitted power does not exceed the channel limit, as shown in the following formula:

[0117] In the formula: β is the safety factor, which is used to avoid power grid stability problems caused by overload of the external transmission channel.

[0118] S6. Based on the feasible constraint set, construct a ten-day mid-term scheduling optimization model, with the objectives of maximizing economic efficiency, maximizing hydropower peak-shaving contribution, and minimizing joint output fluctuation, to obtain the mid-term scheduling scheme and system operation boundary.

[0119] This step involves constructing an upper-level mid-term model, namely the mid-term optimization scheduling sub-model. This model is at the ten-day level, with a period T1 of 10 days, and robust optimization is introduced to handle prediction uncertainties. The objective function used in this step is a multi-objective function with "optimal economic efficiency, maximum contribution of hydropower for peak shaving, and minimum fluctuation of combined output", and the formula is as follows:

[0120] In the formula, The smaller the value, the higher the economic benefit; The smaller the value, the greater the contribution of hydropower. The smaller the value, the lower the volatility; T 1 represents the scheduling cycle; C L ( t ( ) represents the electricity consumption benefit during time period t; C W_P ( t )and C H ( t The costs for wind and solar power and hydropower operation and maintenance during time period t are respectively. C Ab ( t ) represents the penalty cost for energy wastage during time period t; C G ( t () represents the power grid interaction benefits during time period t; P L ( t (t) represents the power system load demand during time period t; P W_P ( t )and P H ( t ) represents the combined wind and solar power output and hydropower output during time period t; μ for T Average combined output within 1.

[0121] The constraints for optimizing the scheduling strategy in step S6 include the following: The total power output from grid interconnection, wind and solar power, and pumped storage power should equal the load power, expressed by the formula: P G ( t ) + P W_P ( t ) + P H ( t ) = P L ( t ) In the formula, P G ( t ) represents the power transmitted from the power grid during time period t.

[0122] The output of wind power and photovoltaic power generation cannot exceed the upper and lower limits, which can be expressed by the formula: P W_P_min ( t )≤ P W_P ( t )≤ P W_P_max (t ) In the formula, P W_P_max ( t () represents the maximum output of scenery during time period t. P W_P_min ( t ) represents the minimum power output of wind and solar energy during time period t.

[0123] Hydropower output needs to be calculated by combining regulation characteristics and water condition forecasts. The hydropower output must meet the upper and lower limits of power, as expressed by the formula: P H_min ( t )≤| P H ( t )|≤ P H_max ( t ) In the formula, P H_max ( t ) represents the maximum hydropower output during time period t. P H_min ( t ) represents the minimum hydropower output during time period t.

[0124] Pumped storage power stations cannot exceed the upper and lower limits of their adjustable pumped storage capacity. The constraint formula is expressed as follows:

[0125] In the formula, E ps ( t (This refers to the pumped storage regulation capacity.) E ps,min ( t This is the minimum capacity for pumped storage regulation. E ps,max ( t () represents the maximum capacity for pumped storage regulation.

[0126] Grid interaction cannot exceed the upper and lower limits of grid interaction power, as shown in the following formula: P G_min ( t )≤| P G ( t )|≤ P G_max ( t ) In the formula, P G_min ( t () represents the minimum power transmitted from the power grid.

[0127] S7. Construct a short-term intraday scheduling optimization model based on the system operation boundary, and add real-time network-side interaction constraints to the short-term intraday scheduling optimization model.

[0128] This step inherits the constraints already determined in the mid-term model, such as the upper and lower limits of system power and output for time period t. It also adds a real-time interactive constraint on the grid side, namely, that the total system output does not exceed its upper and lower limits. The formula is as follows:

[0129] In the formula, P G_min ( t ) represents the minimum power transmitted from the power grid during time period t. P G_max ( t () represents the maximum power transmitted by the power grid during time period t. E G ( t ( ) represents the real-time interactive energy of the power grid at time t. E G_min ( t () represents the lower limit of real-time interactive energy of the power grid at time t. E G_max ( t ) represents the upper limit of real-time interactive energy of the power grid at time t.

[0130] S8. An improved genetic algorithm is used to solve the dual-time-scale model composed of the ten-day mid-term scheduling optimization model and the intraday short-term scheduling optimization model to obtain an optimized scheduling scheme.

[0131] A dual-time-scale model combining a ten-day mid-term scheduling optimization model and an intraday short-term scheduling optimization model is used. For the robust optimization scheduling process based on this model, please refer to [reference needed]. Figure 5 .

[0132] This step employs the improved NSGAII-ASBX (Non-dominated Sorting Genetic Algorithm II with Adaptive Simulated Binary Crossover) algorithm, specifically an improved non-dominated sorting genetic algorithm II with adaptive simulated binary crossover, to solve the dual-timescale model. Optimization directions include improving the binary crossover operator to enhance adaptability to non-convex problems such as pumped storage capacity constraints and multi-scale hydropower regulation constraints, and introducing a population rotation iteration mechanism to improve convergence speed by over 15% in extreme scenarios. Additionally, a new "strong constraint determination for hydropower station output" is added to verify the output limitations of various types of hydropower stations during population updates, ensuring the engineering applicability of the solution.

[0133] S9. Verify the energy curtailment rate and external transmission channel utilization rate of the optimized scheduling scheme. If the verification fails, return to steps S6 and S7 to adjust the parameters and re-execute step S8.

[0134] This step verifies the effectiveness of the scheduling results and determines that the scheduling plan has advantages for extreme scenarios. It is necessary to compare the key indicators of dual-time-scale and single-time-scale (daily-level only) scheduling: In terms of economics, the dual-time-scale model reduces the overall cost by ≥0.60%, increases the contribution rate of hydropower peak shaving by ≥1.91%, and reduces the power output fluctuation rate by ≥14.90%.

[0135] In addition, it is necessary to examine other core performance indicators, such as the utilization rate of the power grid transmission channels and the energy curtailment rate. The beneficial effects need to be quantified based on the original control model. The specific formulas for these indicators are as follows:

[0136]

[0137] If any of the above indicators fail to meet the requirements, return to steps S6 and S7 to adjust the model weights or constraint parameters and solve the problem again.

[0138] The beneficial effects of this embodiment are reflected in the following aspects: First, multi-dimensional indicators are extracted from historical wind and solar power output, load, and hydropower inflow data. After standardization and dimensionless elimination, principal component analysis (PCA) is used to select principal components with a cumulative variance contribution rate of 85% to 95% to remove redundancy. Then, the time series K-means algorithm is introduced, meteorological variables are integrated, and the time series patterns of the data are captured by dynamic time regularization distance (DTW). The massive data is clustered into core scenarios to provide high-quality samples for subsequent processing.

[0139] A quantitative system of 12 indicators was designed. The basic indicators reflect supply and demand imbalances and time-series shocks, the flexibility margin indicators characterize the system's resilience, and the coupled risk indicators quantify the risks of multiple factors. Subsequently, entropy weighting and interval mapping were used for weighting and normalization, and a dynamic threshold model was established using XGBoost to classify extreme scenarios into three levels, clarifying the triggering conditions and response priorities for each level.

[0140] The prediction algorithm is selected accordingly. For long-term predictions (such as ten-day scheduling), Transformer is used, which uses self-attention mechanism to process long sequences and capture seasonal changes and annual fluctuations in watershed inflow. For short-term predictions, LSTM is used, which uses gating mechanism to solve gradient vanishing and combines cell-hidden state updates to capture minute-level changes in wind power and intraday load fluctuations, forming a complementary system of long and short cycles.

[0141] Robust constraints are established according to the scenario. For extreme peak-shaving gap scenarios, the lower limit of wind and solar power output and the upper limit of load are taken, plus hydropower reserve constraints. For extreme curtailment risk scenarios, the upper limit of wind and solar power output and the lower limit of load are taken, plus grid transmission safety constraints.

[0142] We construct a standardized closed-loop process of 9 steps, including data acquisition, feature processing, scene clustering, multi-scale prediction, extreme case determination, robust optimization, dual-scale modeling, algorithm solving, and plan implementation and verification. Each step clearly defines the inputs and outputs, technical methods, and quality standards. The plan implementation and verification step requires comparing the core indicators of dual-timescale and single-timescale scheduling. If the indicators do not meet the standards, we return to the model parameter adjustment step, forming an iterative mechanism of design, verification, and optimization.

[0143] In summary, PCA reduces the dimensionality of output features, decreasing the computational load of the prediction model. K-means is used to capture core scenarios, and a combined long-term and short-term prediction model is employed to obtain high-confidence prediction data. Precise quantitative judgment indicators and strict extreme scenario judgment rules are established. Accuracy feedback enables the system to adaptively adjust to accurately identify extreme source-load imbalances. Robust optimization techniques are combined to effectively address and predict various errors, ensuring stable system operation even under complex conditions such as sudden drops in wind and solar power and load abrupt changes, significantly reducing the energy curtailment rate by 10%-15%. Furthermore, a dual-timescale model successfully balances medium-term global economic optimization with short-term local real-time scheduling. Case studies demonstrate a significant improvement in system economy, specifically a 0.60% increase; simultaneously, the contribution rate of hydropower peak shaving increases by 1.91%, and the volatility of combined output decreases significantly by 14.90%, demonstrating overall performance significantly superior to traditional single-timescale scheduling modes.

[0144] According to another aspect of the invention, Figure 2 This is a schematic diagram illustrating a multi-energy complementary optimization scheduling terminal under an extreme source-load imbalance scenario according to an embodiment of the present invention. The terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the multi-energy complementary optimization scheduling method under an extreme source-load imbalance scenario as described above.

[0145] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for multi-energy complementary optimal scheduling in an extreme source-load imbalance scenario, characterized in that, The method comprises the steps of: S1, calculating the water and electricity output data by collecting historical wind and light output data, load power data and water and electricity inflow data, extracting the multi-dimensional output characteristics of the water and electricity output data, and using principal component analysis to reduce the dimension of the output characteristics; S2, using a time series clustering algorithm to classify the output characteristics after dimension reduction, and obtaining the core scene according to the classification result; S3, establishing a multi-time scale prediction model based on the core scene, inputting the historical wind and light output data, load power data and water and electricity inflow data into the multi-time scale prediction model for prediction, and obtaining a prediction data set; S4, calculating the corresponding extreme source and load imbalance index based on the prediction data set, judging whether an extreme source and load imbalance scene occurs according to the calculation result, and if so, executing step S5; S5, according to the type of the extreme source and load imbalance scene, constructing the worst-case uncertainty interval of wind and light output, water and electricity inflow and load demand, embedding the uncertainty interval into power balance constraints, reserve constraints and grid export constraints to form a feasible constraint set after robustness; S6, constructing a ten-level medium-term scheduling optimization model based on the feasible constraint set, taking the maximum economy, the maximum water energy peak shaving contribution and the minimum joint output fluctuation as multiple objectives, and obtaining a medium-term scheduling scheme and a system operation boundary; S7, constructing an intra-day short-term scheduling optimization model based on the system operation boundary, and adding real-time grid-side interaction constraints to the intra-day short-term scheduling optimization model; S8, solving the double-time scale model composed of the ten-level medium-term scheduling optimization model and the intra-day short-term scheduling optimization model by using an improved genetic algorithm to obtain an optimized scheduling scheme.

2. The multi-energy complementary optimal scheduling method under an extreme source-load imbalance scenario according to claim 1, characterized in that, In step S4, the calculation of the corresponding extreme source and load imbalance index based on the prediction data set comprises: calculating the imbalance index of supply and demand relationship, the impact intensity index, the flexible margin index and the coupling risk index based on the prediction data set; assigning weight values to each index by using an entropy weight method and normalizing the numerical values of each index; accumulating the products of each index and its weight value to obtain a comprehensive risk value; using a gradient boosting tree to construct a threshold calibration model based on meteorological factors, power grid operation factors, water and electricity regulation performance factors, load structure factors and pumped storage capacity factors, calibrating the dynamic threshold of the normalized index by using the threshold calibration model, and obtaining the corresponding early warning threshold of each index and the comprehensive risk threshold of all indexes coupling.

3. The method of claim 2, wherein, In step S4, the judgment of whether an extreme source and load imbalance scene occurs according to the calculation result comprises: when a preset first number of normalized indexes is greater than the corresponding early warning threshold, it is determined as a general extreme source and load imbalance scene; when a preset second number of normalized indexes is greater than the corresponding early warning threshold, it is determined as a serious extreme source and load imbalance scene; when more than a preset second number of normalized indexes is greater than the corresponding early warning threshold, it is determined as a special extreme source and load imbalance scene.

4. The method of claim 1, wherein, Step S5 comprises: determining the type of the extreme source and load imbalance scene as an extreme peak shaving gap or an extreme curtailment risk according to the time information in the prediction data set; If the scenario is judged to be an extreme peak regulation gap, then the worst extreme data set is set according to the prediction data set: wherein, P W_P,worst t is the worst wind and solar power output data for the t period, P L,worst t is the worst load data for the t period, P W_P,pre t is the wind and solar power forecast output data for the t period, P L,pre t is the load forecast data for the t period, is the wind and solar power uncertainty output, is the uncertainty load power output demand;​​​​ The extreme peak regulation gap scenario robust constraint specifically includes the following constraint conditions: The power balance robust constraint: In the formula, P W_P (t)(1 Δ P W_P (To contribute to the worst-case scenario of wind and solar power) P H_max (t) represents the maximum available hydropower output; P L (t)(1+ΔP L This represents the worst-case load demand. The hydroelectric backup robust constraint: In the formula, P H (t) is the actual dispatching output of water and electricity; α is the standby coefficient, which is used to ensure that the water and electricity reserved output should respond to the unexpected increase in load. If it is judged to be an extreme power abandonment risk: In the formula, P W_P t is the wind and solar combined output data;​ The power balance robust constraint: wherein P W_P (1 + ΔP P W_P ) is the worst case of wind and solar power output, P L (1 - ΔP L ) is the worst case of load, P G_max (t) is the maximum power exported to the grid; The power grid sending constraint: In the formula, P G t is the power sent out of the grid; β is a safety factor for avoiding grid stability problems caused by overloading of the sending channel.​ 5. The method of claim 1, wherein, In step S6, a ten-level medium-term scheduling optimization model is constructed on the basis of the feasible constraints, with the maximum economy, the maximum water energy peak regulation contribution and the minimum joint output fluctuation as the multi-objective, including: The objective function formula of the ten-level medium-term scheduling optimization model is as follows: In the formula, The smaller the value is, the higher the economic benefit is; The smaller the value is, the greater the water contribution is; The smaller the value is, the lower the volatility is; T 1 is a dispatching period; C L ( t ) is the electricity benefit in the t period; C W_P ( t ) and C H ( t ) are respectively the wind and light and water power operation and maintenance costs in the t period; C Ab ( t ) is the abandoned energy penalty cost in the t period; C G ( t ) is the grid interaction benefit in the t period; P L ( t ) is the power system load demand in the t period; P W_P ( t ) and P H ( t ) are respectively the wind and light combined output and water power output in the t period; The constraint conditions of the ten-level medium-term scheduling optimization model are as follows: is T 1 is the combined output mean value.

6. The multi-energy complementary optimal scheduling method in an extreme source-load imbalance scenario according to claim 5, characterized in that, The total power of the power grid interaction and the wind and light output and the pumped storage power station output should be equal to the load power, which is expressed by the formula as follows: The wind power generation and photovoltaic power generation output cannot exceed the upper and lower limits, which is expressed by the formula as follows: P G ( t ) + P W_P ( t ) + P H ( t ) = P L ( t ) In the formula, P G t t is the period of time for which power is exported to the grid.​ The water power output needs to be combined with the regulation characteristics and water situation prediction, so as to calculate the related water power output, which needs to meet the power upper and lower limit range, which is expressed by the formula as follows: P W_P_min ( t )≤ P W_P ( t )≤ P W_P_max ( t ) In the formula, P W_P_max ( t ) is the maximum output of wind and light at the t period, P W_P_min ( t ) is the minimum output of wind and light at the t period. The pumped storage power station cannot exceed the upper and lower limits of the pumped storage adjustable capacity, and the constraint formula is expressed as follows: P H_min ( t )≤| P H ( t )|≤ P H_max ( t ) In the formula, P H_max ( t ) is the maximum output of water and electricity at the t period, P H_min ( t ) is the minimum output of water and electricity at the t period. The power grid interaction cannot exceed the upper and lower limits of the power grid interaction power, and the formula is as follows: wherein E ps ( t ) is the pumped storage regulation capacity, E ps,min ( t ) is the pumped storage regulation minimum capacity, E ps,max ( t ) is the pumped storage regulation maximum capacity; The real-time grid side interaction constraint is as follows: P G_min ( t )≤| P G ( t )|≤ P G_max ( t ) In the formula, P G_min ( t () represents the minimum power transmitted from the power grid. P G_max (t) represents the maximum power transmitted from the power grid.

7. The method of claim 6, wherein, The optimization scheduling scheme is obtained, and then includes: In the formula, P G_min t is the minimum power sent to the grid in the t period, P G_max t is the maximum power sent to the grid in the t period, E G t is the real-time interactive energy of the grid at time t, E G_min t is the lower limit of the real-time interactive energy of the grid at time t, E G_max t is the upper limit of the real-time interactive energy of the grid at time t.​​​​​ 8. The method of claim 1, wherein, The optimization scheduling scheme is verified by the energy abandonment rate and the utilization rate of the sending channel, and if the verification is not up to standard, steps S6 and S7 are returned to adjust the parameters and execute step S8 again. The calculation method of the energy abandonment rate is as follows:

9. The method of claim 8, wherein, The calculation method of the sending channel utilization rate is as follows: wherein, P W_P_max ( t ) denotes the maximum wind and solar power at time t, P H_max ( t ) denotes the maximum hydro power at time t, P W_P ( t ) denotes the wind and solar power at time t, P H ( t ) denotes the hydro power at time t, denotes the pumped storage power at time t.

10. A multi-energy complementary optimization scheduling terminal under an extreme source and load imbalance scenario, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement each step of the multi-energy complementary optimization scheduling method under the extreme source and load imbalance scenario according to any one of claims 1 to 9. wherein P G ( t ) represents the grid export power at time t, P G_max ( t ) represents the maximum grid export power at time t. ​

Images