Multi-time-scale water and energy storage and high-proportion green energy absorption method and system

Abstract

The invention discloses a multi-time-scale water and energy storage and high-proportion green energy absorption method and system. The method comprises the steps that a long-term scheduling layer, a medium-term scheduling layer and a short-term scheduling layer achieve cooperation through a data bus, and prediction data, reservoir hydrological data and power market signals are integrated in the long-term scheduling plan making stage; in the medium-term prediction correction and dynamic adjustment stage, the wind and light prediction module receives real-time meteorological data; in the short-term real-time control and frequency modulation response stage, a digital electro-hydraulic speed regulator in the scale controller adjusts the speed; the system comprises a physical layer, a data bus layer and a virtual layer. According to the method, the LSTM neural network and other advanced algorithms are adopted to predict and correct the wind and light output, the prediction precision is improved, more reliable data support is provided for scheduling decision making, the intelligent level and decision making quality of the system are continuously improved, the scheduling scheme better meets the actual operation requirement, and the flexibility and adaptability of a power grid are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of water storage and energy storage technology, and in particular to a multi-time scale water storage and energy storage and high-proportion green energy absorption method and system. Background Art

[0002] As the global energy mix accelerates its transition toward renewable energy, the large-scale integration of new energy sources, such as wind power and photovoltaics, into the grid has become an inevitable trend in power system development. However, these energy sources exhibit significant intermittency, volatility, and uncertainty. Their output levels are susceptible to random fluctuations due to meteorological conditions, leading to multiple challenges for power grids, including frequency fluctuations, voltage instability, and insufficient backup capacity. Especially when the penetration rate of new energy sources exceeds a certain threshold, traditional power grid dispatching systems struggle to effectively smooth power fluctuations, potentially triggering the risk of cascading failures and seriously threatening the safe and stable operation of the power system.

[0003] The traditional power dispatching system has exposed significant technical limitations when dealing with the high proportion of green energy connected to the grid. The existing dispatching model mostly adopts a single time scale planning, lacks a multi-level coordination mechanism for long, medium and short cycles, and is difficult to balance the contradiction between the random volatility of renewable energy output and the regulation inertia of conventional units. In the forecasting link, conventional methods have insufficient forecast accuracy for wind and solar output, and are unable to dynamically correct errors, resulting in a disconnect between the dispatching plan and the actual operating conditions; at the regulation level, conventional units such as thermal power and hydropower have a lagging response speed, making it difficult to meet the rapid frequency regulation requirements of seconds or even milliseconds, exacerbating the risk of grid frequency fluctuations. In addition, existing dispatching objectives are often limited to a single dimension of economy, and fail to fully consider carbon emission constraints and green electricity consumption responsibilities, making it difficult to achieve coordinated optimization of environmental and economic benefits.

[0004] Existing technologies still have significant gaps in multi-timescale collaborative optimization and intelligent regulation. Traditional dispatch models usually ignore emerging factors such as carbon trading costs and virtual inertia control, resulting in dispatch decisions that cannot adapt to the marketization and low-carbonization needs of new power systems. In terms of equipment coordination, the regulation potential of flexible resources such as pumped storage and energy storage has not been fully tapped, and their dynamic matching mechanism with new energy output is still imperfect. More importantly, existing technical solutions lack a closed-loop feedback mechanism across time scales, making it difficult to achieve effective linkage between long-term planning, mid-term corrections and short-term adjustments, which restricts the ability to accept a high proportion of green energy and the improvement of grid flexibility. Summary of the Invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a multi-time scale water storage and energy storage and high proportion green energy absorption method and system.

[0006] Technical solution: The multi-time-scale water storage and energy storage and high-proportion green energy absorption method described in the present invention realizes coordination among the long-term scheduling layer, the medium-term scheduling layer and the short-term scheduling layer through the data bus. The long-term scheduling plan stage formulates and integrates the forecast data, reservoir hydrological data and power market signals for the next 7 days; in the medium-term forecast correction and dynamic adjustment stage, every hour, the wind and solar prediction module receives real-time meteorological data, and uses the pre-trained LSTM neural network model to rollingly correct the wind and solar output forecast for the next hour, and dynamically adjusts the output plan of the hydropower unit; in the short-term real-time control and frequency response stage, when the grid frequency deviates or the wind and solar output suddenly changes, the short-term control layer starts immediately, and the digital electro-hydraulic speed regulator in the scale controller adjusts the rate; at the same time, the energy storage equipment responds synchronously, and the real-time adjustment amount is dynamically calculated according to the frequency deviation.

[0007] Furthermore, it is characterized in that the long-term scheduling layer generates reservoir energy storage plans and long-term wind and solar scheduling goals through algorithms, and calculates the reservoir water level range.

[0008] Furthermore, the mid-term scheduling layer adjustment instructions are sent to the hydropower station and energy storage control terminal in real time through the communication module, and the SOC target value of the energy storage equipment is recalculated to reserve sufficient capacity to cope with subsequent fluctuations.

[0009] Furthermore, when the short-term scheduling layer is started, the digital electro-hydraulic speed regulator in the scale controller adjusts the rate; the energy storage device responds synchronously, the real-time adjustment amount is dynamically calculated according to the frequency deviation, and the water-storage system collaborative instructions are executed through a low-latency communication link, while the actual output is fed back to the data controller.

[0010] Furthermore, the long-term layer scheduling optimization objective function is:

[0011]

[0012] Among them, λ1 represents the green electricity premium coefficient, ranging from 0.1 to 0.3 yuan / kWh, P 1，t represents the wind power and photovoltaic output during time period t, C 1，t represents the operating cost of the hydropower station, β represents the policy subsidy coefficient, T represents the total number of time periods in the scheduling cycle, and CO2 represents the carbon dioxide emitted.

[0013] Furthermore, the constraints of the long-term scheduling optimization objective function are: output regulation rate ≥ 4% / min, frequency regulation dead zone ≤ ± 0.02Hz, virtual inertia simulation constant H ≥ 5s,

[0014] C 1，t =k1·P 2，t +k2·Δh

[0015] Among them, k1 is the unit loss coefficient, k2 is the water level change cost, and Δh is the reservoir water level change.

[0016] Furthermore, the long-term dispatch plan formulation integrating forecast data, reservoir hydrological data and power market signals includes:

[0017] (1) The improved RRT algorithm is used to optimize the long-term layer, with the goal of determining the optimal combination of λ1, β, k1, and k2. Data-driven calibration is performed, and the green electricity premium coefficient is obtained by fitting historical green electricity transaction data with the marginal cost of hydropower stations. The formula is as follows:

[0018]

[0019] C is the marginal cost of the hydropower station;

[0020] (2) The reward function is introduced for reinforcement learning. The state space is the reservoir water level, energy storage SOC, and wind and solar power forecast initial force. The action space is the unit output adjustment amount and energy storage charge and discharge power. The reward function formula is as follows:

[0021]

[0022] Among them, r t is the bonus factor, CO21 is the amount of carbon dioxide emissions reduced relative to the baseline scenario;

[0023] (3) Population initialization: define the search space and generate the initial population;

[0024] (4) Node generation: Randomly generate new parameter combinations in the search space. The formula is as follows:

[0025] x new =x current +Δx·u

[0026] Among them, x current is the current parameter combination, Δx is the step size, and u is the random direction vector; a collision test is used to check whether the newly generated node meets all the constraints. If so, it is added to the tree and its adaptability value in the objective function is calculated as follows:

[0027]

[0028] (5) Algorithm improvement: To improve the search efficiency, the algorithm is improved by introducing an adaptive step size strategy, which adjusts the step size according to the difference between the current optimal solution and the current node:

[0029]

[0030] f(x best ) is the current optimal solution fitness value, xbest is the optimal solution, x current is the current node, d max is the maximum allowed step length;

[0031] In order to balance economic and environmental benefits, we introduce multi-objective optimization and use Pareto front analysis:

[0032] ParetoFront={(f1(x), f2(x))|x∈SearchSpace}

[0033] f1(x) is the economic benefit, f2(x) is the carbon emission reduction;

[0034] (6) Objective function evaluation: Calculate the fitness of each node in the objective function to ensure that all nodes meet the constraints, such as output regulation rate, frequency regulation dead zone, and virtual inertia constant;

[0035] (7) Update and termination: Select excellent nodes according to fitness to enter the next generation, make small mutations on the selected nodes, and converge if the fitness value change is less than the preset threshold, otherwise return to step (4);

[0036] (8) Result output: Output the node with the highest fitness as the optimal parameter combination, that is, the optimal combination of λ1, β, k1, and k2.

[0037] Furthermore, the use of the LSTM neural network to correct the wind and solar power output forecast and dynamically adjust the hydropower unit output plan includes:

[0038] The output range of the hydropower unit can be dynamically adjusted as follows:

[0039]

[0040] is the output of the hydropower unit, P normal Rated power of hydropower unit, The energy storage SOC target value is updated based on the wind and solar power output prediction error value. The formula is as follows:

[0041]

[0042] SOC is the target value, ΔP regulate is the dynamic adjustment power, η is the cycle efficiency, E is the total capacity SOC of energy storage current is the current value, Δt is the time interval;

[0043] The constraints are:

[0044]

[0045] Output adjustment and charge and discharge targets after rolling correction.

[0046] Furthermore, the control logic of the short-term real-time control and frequency modulation response stage is as follows:

[0047] The speed regulation of the hydropower unit is controlled at a rate of ≥4% rated power / minute. The dynamic equation is as follows:

[0048]

[0049] Where ΔP 2，t is the power adjustment of the hydropower unit, P 2，rated represents the rated power of the hydropower unit, τ is the integral time variable, t0 is the starting time of frequency deviation, K p is 0.8, K i is 0.2, Δf τ is the grid frequency deviation.

[0050] To support the rapid power of energy storage equipment, the charge and discharge instructions are as follows:

[0051]

[0052] ΔP t Provides fast power to energy storage devices and protects SOC t ≥95% prohibit charging, SOC t Discharge is prohibited when the value is less than 5%. PI control is used to adjust the frequency. The formula is as follows:

[0053]

[0054] Δ regulate，t To adjust power based on high and low frequency scenarios, when Δf>0.1Hz is a high frequency scenario, energy storage is called first, and when Δf<-0.1Hz is a low frequency scenario, water storage is coordinated and adjusted, and then the power adjustment amount of the hydropower unit and the energy storage charging and discharging instructions are output.

[0055] The multi-timescale water storage and energy storage and high-proportion green energy absorption system of the present invention includes:

[0056] The system has a three-layer architecture consisting of a physical layer, a data bus layer, and a virtual layer, and achieves dynamic mapping through a bidirectional data flow and feedback mechanism. The physical layer includes hydropower stations, energy storage equipment, wind farms, and photovoltaic power stations, deploying sensors to collect parameters such as temperature, power, and frequency in real time, and uploading them to the data bus via the OPC UA protocol. The data bus layer enables data interaction between the physical and virtual layers. The raw data is cleaned, compressed, and feature extracted through edge computing nodes. The virtual layer includes a three-layer optimization model and a simulation engine. The optimization results are sent to the physical layer for execution via the data bus, and closed-loop calibration is performed on execution deviations.

[0057] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0058] (1) Through coordinated regulation at multiple time scales, it can effectively deal with the intermittent and volatile problems caused by the high proportion of renewable energy grid connection; the rapid response of the short-term layer can quickly smooth out the grid frequency deviation and sudden changes in wind and solar power output, ensuring the stability of the grid frequency; the prediction correction and dynamic adjustment of the medium-term layer can plan in advance to reduce the impact of subsequent fluctuations; the formulation of reservoir energy storage plans in the long-term layer provides a long-term regulation basis for the system, thereby enhancing the grid's ability to cope with complex working conditions, reducing risks such as frequency fluctuations and voltage instability, and ensuring the safe and stable operation of the power system.

[0059] (2) By building a dynamic control system with three levels of linkage (long-term, medium-term, and short-term), the limitations of traditional single-time-scale planning can be broken, and multi-level collaborative optimization of long, medium, and short periods can be achieved. The various levels can exchange information in real time through the data bus, forming a complete control chain of "strategy-tactics-combat", making the dispatching decision more scientific, accurate, and flexible, and able to quickly respond to the dynamic changes of the power grid, thereby improving the system's adaptability and rapid recovery capabilities;

[0060] (3) Use advanced algorithms such as LSTM neural networks to predict and correct wind and solar power output, improve prediction accuracy, and provide more reliable data support for scheduling decisions. At the same time, use improved RRT algorithms, reinforcement learning and other optimization methods to optimize and solve the long-term scheduling parameters, continuously improve the system's intelligence level and decision-making quality, make the scheduling plan more in line with actual operation needs, and enhance the flexibility and adaptability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of the method of the present invention;

[0062] Figure 2 It is the algorithm flow chart of the present invention;

[0063] Figure 3 This is the distribution diagram of the digital twin system of the present invention. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0065] like Figure 1As shown, the multi-timescale water storage and high-proportion green energy absorption method described in this invention integrates forecast data for the next seven days (including wind speed, irradiance, and rainfall), reservoir hydrological data (reservoir capacity curve, inflow flow), and power market signals (peak and valley electricity prices, green energy certificate prices) into long-term scheduling plans (daily / weekly). An algorithm is used to generate reservoir storage plans and long-term wind and solar scheduling targets. The reservoir water level range is calculated to reserve storage capacity to adjust for wind and solar fluctuations, providing a basic constraint for subsequent scheduling.

[0066] During the hourly mid-term forecast correction and dynamic adjustment phase, the wind and solar power forecast module receives real-time meteorological data (such as satellite cloud images and ground-based wind tower data) every hour. Using a pre-trained LSTM neural network model, it rolls out the wind and solar power output forecast for the next hour, keeping the forecast error within 5%. Simultaneously, the SOC target value of the energy storage device is recalculated to reserve sufficient capacity to cope with subsequent fluctuations. Mid-term adjustment instructions are distributed in real time to the hydropower station and energy storage control terminal via the communication module, ensuring system flexibility.

[0067] During the short-term real-time control and frequency response phase (seconds / minutes), when the grid frequency deviates (e.g., Δf = +0.1Hz) or when there's a sudden change in wind and solar power output (e.g., a 50MW drop in wind power), the short-term control layer immediately activates. The digital electro-hydraulic speed regulator in the scale controller adjusts the speed, while the energy storage device responds synchronously. Real-time regulation is dynamically calculated based on the frequency deviation to ensure frequency stability. Coordinated commands between the water and storage systems are executed via a low-latency communication link (≤10ms), while actual output is fed back to the data controller.

[0068] The long-term, medium-term, and short-term scheduling layers collaborate via a data bus. For example, if the short-term layer detects increased losses from frequent hydropower unit starts and stops, it will provide feedback to the long-term layer, increasing the reserved reservoir regulation capacity in the next week's plan and reducing the number of unit operations. Furthermore, the LSTM forecasting model is retrained quarterly with new meteorological data to improve forecast accuracy (for example, the error rate has been reduced from 5% to 3%). The results of model iterations are automatically updated by the controller to ensure continuous system optimization.

[0069] like Figure 2 As shown, the long-term layer scheduling optimization objective function is:

[0070]

[0071] Among them, λ1 represents the green electricity premium coefficient, ranging from 0.1 to 0.3 yuan / kWh, P 1，t represents the wind power and photovoltaic output during time period t, C 1，t represents the operating cost of the hydropower station, β represents the policy subsidy coefficient, T represents the total number of time periods in the scheduling cycle, and CO2 represents the carbon dioxide emitted.

[0072] The constraints of the long-term scheduling optimization objective function are: output regulation rate ≥ 4% / min, frequency regulation dead zone ≤ ± 0.02Hz, virtual inertia simulation constant H ≥ 5s,

[0073] C 1,t =k1·P 2,t +k2·Δh

[0074] Among them, k1 is the unit loss coefficient, k2 is the water level change cost, and Δh is the reservoir water level change.

[0075] The long-term dispatch plan developed by integrating forecast data, reservoir hydrological data, and power market signals includes:

[0076] (1) The improved RRT algorithm is used to optimize the long-term layer, with the goal of determining the optimal combination of λ1, β, k1, and k2. Data-driven calibration is performed, and the green electricity premium coefficient is obtained by fitting historical green electricity transaction data with the marginal cost of hydropower stations. The formula is as follows:

[0077]

[0078] C is the marginal cost of the hydropower station;

[0079] (2) The reward function is introduced for reinforcement learning. The state space is the reservoir water level, energy storage SOC, and wind and solar power forecast initial force. The action space is the unit output adjustment amount and energy storage charge and discharge power. The reward function formula is as follows:

[0080]

[0081] Among them, r t is the bonus factor, CO21 is the amount of carbon dioxide emissions reduced relative to the baseline scenario;

[0082] (3) Population initialization: define the search space and generate the initial population;

[0083] (4) Node generation: Randomly generate new parameter combinations in the search space. The formula is as follows:

[0084] x new =x current +Δx·u

[0085] Among them, x current is the current parameter combination, Δx is the step size, and u is the random direction vector; a collision test is used to check whether the newly generated node meets all the constraints. If so, it is added to the tree and its adaptability value in the objective function is calculated as follows:

[0086]

[0087] (5) Algorithm improvement: To improve the search efficiency, the algorithm is improved by introducing an adaptive step size strategy, which adjusts the step size according to the difference between the current optimal solution and the current node:

[0088]

[0089] f(x best ) is the current optimal solution fitness value, x best is the optimal solution, x current is the current node, d max is the maximum allowed step length;

[0090] In order to balance economic and environmental benefits, we introduce multi-objective optimization and use Pareto front analysis:

[0091] ParetoFront={(f1(x), f2(x))|x∈SearchSpace}

[0092] f1(x) is the economic benefit, f2(x) is the carbon emission reduction;

[0093] (6) Objective function evaluation: Calculate the fitness of each node in the objective function to ensure that all nodes meet the constraints, such as output regulation rate, frequency regulation dead zone, and virtual inertia constant;

[0094] (7) Update and termination: Select excellent nodes according to fitness to enter the next generation, make small mutations on the selected nodes, and converge if the fitness value change is less than the preset threshold, otherwise return to step (4);

[0095] (8) Result output: Output the node with the highest fitness as the optimal parameter combination, that is, the optimal combination of λ1, β, k1, and k2.

[0096] The use of LSTM neural network to correct wind and solar power output forecasts and dynamically adjust hydropower unit output plans includes:

[0097] The output range of the hydropower unit can be dynamically adjusted as follows:

[0098]

[0099] is the output of the hydropower unit, P normal Rated power of hydropower unit, The energy storage SOC target value is updated based on the wind and solar power output prediction error value. The formula is as follows:

[0100]

[0101] SOC is the target value, ΔP regulate is the dynamic adjustment power, 77 is the cycle efficiency, and E is the total capacity SOC of energy storagecurrent is the current value, Δt is the time interval;

[0102] The constraints are:

[0103]

[0104] Output adjustment and charge and discharge targets after rolling correction.

[0105] The control logic of the short-term real-time control and frequency modulation response stage is as follows:

[0106] The speed regulation of the hydropower unit is controlled at a rate of ≥4% rated power / minute. The dynamic equation is as follows:

[0107]

[0108] Where ΔP 2，t is the power adjustment of the hydropower unit, P 2，rated represents the rated power of the hydropower unit, τ is the integral time variable, t0 is the starting time of frequency deviation, K p is 0.8, K i is 0.2, Δf τ is the grid frequency deviation.

[0109] To support the rapid power of energy storage equipment, the charge and discharge instructions are as follows:

[0110]

[0111] ΔP t Provides fast power to energy storage devices and protects SOC t ≥95% prohibit charging, SOC t Discharge is prohibited when the value is less than 5%. PI control is used to adjust the frequency. The formula is as follows:

[0112]

[0113] ΔP regulate，t To regulate power based on high- and low-frequency scenarios, when Δf > 0.1Hz, it's a high-frequency scenario, prioritizing energy storage. When Δf < -0.1Hz, it's a low-frequency scenario, with coordinated regulation of water and storage. The system then outputs power adjustments for the hydropower units and energy storage charge and discharge instructions. The short-term control layer achieves rapid grid frequency stabilization through minute-by-minute hydropower regulation and millisecond-by-millisecond energy storage response, while also forming a coarse-to-fine coordinated optimization strategy with the long-term scheduling layer.

[0114] The long-, medium-, and short-term layers form a multi-level coordinated mechanism of "strategy-tactics-combat." For example, when wind and solar power output suddenly fluctuates within a day, the short-term layer quickly responds to smooth the fluctuation. The medium-term layer then corrects subsequent forecasts and adjusts the unit mix. The long-term layer optimizes reservoir scheduling strategies in weekly plans to compensate for intraday regulation losses. This data bus forms a closed energy and information loop of "long-term energy storage-medium-term calibration-short-term regulation."

[0115] The multi-timescale water storage and high-proportion green energy absorption system described in this invention involves building a digital twin system. This system achieves multi-timescale coordinated optimization of hydropower storage and green energy through a three-dimensional fusion of physical entities, virtual models, and data interaction. The system comprises a three-layer architecture: a physical layer, a data bus layer, and a virtual layer. Dynamic mapping is achieved through bidirectional data flow and feedback mechanisms. The physical layer includes physical equipment such as hydropower stations, energy storage devices, wind farms, and photovoltaic power stations. Sensors are deployed to collect parameters such as temperature, power, and frequency in real time (with a sampling rate ≥ 1Hz) and upload them to the data bus via the OPC UA protocol. The data bus layer enables data exchange between the physical and virtual layers. Edge computing nodes clean, compress, and extract features (such as wind and solar output fluctuations and SOC change rates) from raw data. The virtual layer includes a three-layer optimization model and a simulation engine. Optimization results are transmitted to the physical layer via the data bus for execution (such as reservoir gate opening instructions and energy storage charge and discharge power instructions). Closed-loop calibration is also performed for execution deviations. Through the coordinated mechanism of "long-cycle energy storage-medium-cycle calibration-short-cycle regulation", the temporal and spatial contradiction between the volatility of wind and solar power output and the inertia of hydropower regulation is resolved, and multi-objective optimization of economy, reliability and environmental benefits is achieved.

Claims

1. A multi-timescale water storage and energy storage method with a high proportion of green energy absorption, characterized in that: The long-term scheduling layer, medium-term scheduling layer and short-term scheduling layer are coordinated through the data bus. The long-term scheduling plan stage integrates the forecast data, reservoir hydrological data and power market signals for the next 7 days; in the medium-term forecast correction and dynamic adjustment stage, every hour, the wind and solar forecast module receives real-time meteorological data, and uses the pre-trained LSTM neural network model to rollingly correct the wind and solar output forecast for the next hour, and dynamically adjust the hydropower unit output plan; in the short-term real-time control and frequency response stage, when the grid frequency deviates or the wind and solar output suddenly changes, the short-term control layer starts immediately, and the digital electro-hydraulic speed regulator in the scale controller adjusts the rate; at the same time, the energy storage equipment responds synchronously, and the real-time adjustment amount is dynamically calculated according to the frequency deviation.

2. The multi-timescale water storage and energy storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The long-term scheduling layer generates reservoir energy storage plans and long-term wind and solar scheduling targets through algorithms, and calculates the reservoir water level range.

3. The multi-timescale water storage and energy storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The mid-term scheduling layer adjustment instructions are sent to the hydropower station and energy storage control terminal in real time through the communication module. The SOC target value of the energy storage equipment is recalculated to reserve sufficient capacity to cope with subsequent fluctuations.

4. The multi-timescale water storage and energy storage and high-proportion green energy absorption method according to claim 1 is characterized in that: When the short-term scheduling layer is activated, the digital electro-hydraulic speed regulator in the scale controller adjusts the speed; the energy storage device responds synchronously, and the real-time adjustment amount is dynamically calculated based on the frequency deviation. The water-storage system collaborative instructions are executed through a low-latency communication link, and the actual output is fed back to the data controller.

5. The multi-timescale water storage and energy storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The long-term layer scheduling optimization objective function is: Among them, λ1 represents the green electricity premium coefficient, ranging from 0.1 to 0.3 yuan / kWh, P 1，t represents the wind power and photovoltaic power output during time period t, C 1，t represents the operating cost of the hydropower station, β represents the policy subsidy coefficient, T represents the total number of time periods in the scheduling cycle, and CO2 represents the carbon dioxide emitted.

6. The multi-timescale water storage and high-proportion green energy absorption method according to claim 5 is characterized in that: The constraints of the long-term scheduling optimization objective function are: output regulation rate ≥ 4% / min, frequency regulation dead zone ≤ ± 0.02Hz, virtual inertia simulation constant H ≥ 5s, C 1，t =k1·P 2,t +k2·Δh Among them, k1 is the unit loss coefficient, k2 is the water level change cost, and Δh is the reservoir water level change.

7. The multi-timescale water storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The long-term dispatch plan developed by integrating forecast data, reservoir hydrological data, and power market signals includes: (1) The improved RRT algorithm is used to optimize the long-term layer, with the goal of determining the optimal combination of λ1, β, k1, and k2. Data-driven calibration is performed, and the green electricity premium coefficient is obtained by fitting historical green electricity transaction data with the marginal cost of hydropower stations. The formula is as follows: C is the marginal cost of the hydropower station; (2) The reward function is introduced for reinforcement learning. The state space is the reservoir water level, energy storage SOC, and wind and solar power forecast initial force. The action space is the unit output adjustment amount and energy storage charge and discharge power. The reward function formula is as follows: Among them, r t is the bonus factor, CO21 is the amount of carbon dioxide emissions reduced relative to the baseline scenario; (3) Population initialization: define the search space and generate the initial population; (4) Node generation: Randomly generate new parameter combinations in the search space. The formula is as follows: x new =x current +Δx·u Among them, x current is the current parameter combination, Δx is the step size, and u is the random direction vector; a collision test is used to check whether the newly generated node meets all the constraints. If so, it is added to the tree and its adaptability value in the objective function is calculated as follows: (5) Algorithm improvement: To improve the search efficiency, the algorithm is improved by introducing an adaptive step size strategy, which adjusts the step size according to the difference between the current optimal solution and the current node: f(x best ) is the current optimal solution fitness value, x best is the optimal solution, x current is the current node, d max is the maximum allowed step length; In order to balance economic and environmental benefits, we introduce multi-objective optimization and use Pareto front analysis: ParetoFront={(f1(x), f2(x))|x∈SearchSpace} f1(x) is the economic benefit, f2(x) is the carbon emission reduction; (6) Objective function evaluation: Calculate the fitness of each node in the objective function to ensure that all nodes meet the constraints, such as output regulation rate, frequency regulation dead zone, and virtual inertia constant; (7) Update and termination: Select excellent nodes according to fitness to enter the next generation, make small mutations on the selected nodes, and converge if the fitness value change is less than the preset threshold, otherwise return to step (4); (8) Result output: Output the node with the highest fitness as the optimal parameter combination, that is, the optimal combination of λ1, β, k1, and k2.

8. The multi-timescale water storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The use of LSTM neural network to correct wind and solar power output forecasts and dynamically adjust hydropower unit output plans includes: The output range of the hydropower unit can be dynamically adjusted as follows: is the output of the hydropower unit, P normal Rated power of hydropower unit, The energy storage SOC target value is updated based on the wind and solar power output prediction error value. The formula is as follows: SOC is the target value, ΔP regulate is the dynamic adjustment power, η is the cycle efficiency, E is the total capacity SOC of energy storage current is the current value, Δt is the time interval; The constraints are: Output adjustment and charge and discharge targets after rolling correction.

9. The multi-timescale water storage and energy storage and high-proportion green energy absorption method according to claim 1 is characterized in that: The control logic of the short-term real-time control and frequency modulation response stage is as follows: The speed regulation of the hydropower unit is controlled at a rate of ≥4% rated power / minute. The dynamic equation is as follows: Where ΔP 2，t is the power adjustment of the hydropower unit, P 2，rated represents the rated power of the hydropower unit, τ is the integral time variable, t0 is the starting time of frequency deviation, K p is 0.8, K i is 0.2, Δf τ is the grid frequency deviation. To support the rapid power of energy storage equipment, the charge and discharge instructions are as follows: ΔP t Provides fast power to energy storage devices and protects SOC t ≥95% prohibit charging, SOC t Discharge is prohibited when the value is less than 5%. PI control is used to adjust the frequency. The formula is as follows: ΔP regulate，t To adjust power based on high and low frequency scenarios, when Δf>0.1Hz is a high frequency scenario, energy storage is called first, and when Δf<-0.1Hz is a low frequency scenario, water storage is coordinated and adjusted, and then the power adjustment amount of the hydropower unit and the energy storage charging and discharging instructions are output.

10. A multi-time scale water storage and energy storage system with a high proportion of green energy absorption, characterized by: include: The system has a three-layer architecture consisting of a physical layer, a data bus layer, and a virtual layer, and achieves dynamic mapping through a bidirectional data flow and feedback mechanism. The physical layer includes hydropower stations, energy storage equipment, wind farms, and photovoltaic power stations. Sensors are deployed to collect real-time parameters such as temperature, power, and frequency, and upload them to the data bus via the OPC UA protocol. The data bus layer enables data exchange between the physical and virtual layers. Clean, compress and extract features of raw data through edge computing nodes; The virtual layer contains three layers of optimization models and simulation engines. The optimization results are sent to the physical layer through the data bus for execution, and closed-loop calibration is performed on the execution deviation.

Images