Hydroelectric power station time-sharing power generation benefit dynamic control method fusing power grid load demand
By combining attention mechanisms and deep reinforcement learning, a time-sharing power generation control method for hydropower stations was constructed, which solved the problem of power generation plan optimization under the conditions of grid load fluctuations and dynamic changes in electricity prices, and achieved efficient dynamic control and profit maximization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF PETROCHEMICAL TECH
- Filing Date
- 2025-12-26
- Publication Date
- 2026-05-19
AI Technical Summary
Traditional hydropower station power generation control methods are difficult to adapt to the temporal fluctuations in grid load demand and the time-of-use pricing mechanism, resulting in poor scientific nature of power generation planning and difficulty in achieving dynamic optimization and high efficiency.
By constructing a data association model based on an attention mechanism and combining it with a deep reinforcement learning algorithm, a fusion feature dataset is generated to optimize the time-of-use power generation plan. A dynamic control model is constructed by introducing a deep reinforcement learning algorithm, and iterative optimization and deviation triggering mechanisms are used with real-time data to achieve dynamic control of the time-of-use power generation revenue of the hydropower station.
It enables precise capture of peak grid load and high electricity price periods, improves the scientific nature and feasibility of power generation revenue, and ensures that power generation plans remain optimized and stable amidst dynamic changes.
Smart Images
Figure CN121484924B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydropower station dispatching technology, specifically to a dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand. Background Technology
[0002] Hydropower stations, as key energy hubs with multiple benefits including flood control, power generation, and ecological sustainability, directly impact energy utilization efficiency and power system stability through the precision of their operation and scheduling. With the deepening of smart grid construction, grid load demand exhibits significant temporal fluctuations and randomness. Coupled with the widespread implementation of time-of-use pricing mechanisms in the electricity market, traditional hydropower station power generation control models are no longer adequate to meet the operational demands of the new situation. Currently, most hydropower stations still employ scheduling methods based on empirical rules or static mathematical models. While these methods can meet basic power generation and flood control requirements, they have significant limitations in coordinating the dynamic balance between grid load response and power generation revenue optimization, necessitating technological innovation to overcome these bottlenecks.
[0003] Traditional hydropower station power generation control methods suffer from significant shortcomings in data processing, making it difficult to achieve deep integration and efficient utilization of multi-source information. Existing technologies typically employ independent data acquisition and processing models, separating internal hydropower station operational data (such as water level, flow rate, and unit status) from external grid data (such as load demand and time-of-use pricing), lacking effective spatiotemporal matching and correlation analysis mechanisms. In data weight allocation, fixed coefficient methods are often used, failing to automatically adjust data priorities based on key periods such as peak load and peak electricity prices. This results in the generated basic dataset failing to accurately reflect the coupling relationship between "hydropower operation - grid demand - economic benefits." These limitations directly impact the scientific rigor of subsequent power generation planning, making it difficult for hydropower stations to quickly adapt to sudden fluctuations in grid load or dynamic price adjustments.
[0004] In terms of power generation planning optimization and control model construction, existing technologies have not yet formed an effective solution that balances dynamism and accuracy. While some methods aim to maximize returns, they often employ traditional optimization algorithms such as linear programming and dynamic programming. These algorithms rely on explicit mathematical assumptions and static parameter inputs, making it difficult to handle the dynamic coupling relationships of multiple variables in hydropower operation, such as reservoir inflow and grid load. Furthermore, the model constraints are often simplified, primarily considering only basic constraints such as reservoir water level limits and generator output ranges, neglecting crucial factors like downstream ecological flow assurance and dynamic changes in head loss, thus limiting the engineering applicability of the optimization results. In addition, existing models are mostly calculated offline, lacking real-time iterative optimization capabilities. When actual operating conditions deviate from preset conditions, they cannot promptly correct power generation strategies, easily leading to problems such as insufficient peak-hour output resulting in missed high returns, and excessive off-peak output increasing grid peak-shaving pressure.
[0005] With the rapid development of artificial intelligence technologies such as deep reinforcement learning and attention mechanisms, new technical paths have been provided for the dynamic power generation control of hydropower stations. However, existing research still has many areas for improvement. Current AI-based scheduling research focuses on optimizing single aspects, such as unit load allocation or short-term output prediction, and has not yet formed a complete technical system. Furthermore, existing technologies lack effective dynamic correction and feedback mechanisms, making it difficult to continuously iterate and optimize models based on actual operating data after training, thus limiting their long-term adaptability and reliability in complex and ever-changing hydropower operation scenarios. Therefore, developing a full-process control method that can deeply integrate grid load demand and dynamically optimize time-of-use power generation revenue has become an urgent need for the intelligent development of the hydropower industry. Summary of the Invention
[0006] Based on the aforementioned technical problems, this application discloses a dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand, specifically including:
[0007] Acquire multi-dimensional operational data of hydropower stations and time-of-use load and electricity price data of the power grid;
[0008] A data association model is constructed based on the attention mechanism. Multi-dimensional operational data is spatiotemporally matched with power grid time-of-use load and electricity price data. Dynamic weight allocation highlights the data correlation between peak load periods and high electricity price periods, generating a fused feature dataset.
[0009] With the objective function of maximizing the revenue from time-of-use power generation, and with constraints such as reservoir water balance, unit output limits and downstream ecological flow, a dynamic control model is constructed by introducing a deep reinforcement learning algorithm.
[0010] The dynamic control model is trained using historical operating data and grid load fluctuation simulation data. The model parameters are iteratively optimized to ensure that the time-of-use generation plan output by the model meets the constraints and maximizes the objective function value.
[0011] The real-time collected multi-dimensional operation data and the grid time-sharing load and electricity price update data are input into the trained dynamic control model to generate the optimal unit output value for each time period and form a time-sharing power generation control scheme for the hydropower station.
[0012] Based on the time-of-use power generation control scheme, the power adjustment command is issued to the hydropower station units, and the deviation between the actual output of the units and the control scheme is monitored in real time to realize the dynamic control of the time-of-use power generation revenue of the hydropower station.
[0013] Preferably, the multi-dimensional operational data is collected through a distributed sensor network to collect reservoir water level, inflow, and unit operating parameters, including at least real-time reservoir water level, inflow, and unit operating parameters; the grid time-of-use load and electricity price data are obtained through the grid dispatch data interface to obtain time-of-use load demand and corresponding electricity price within a future preset period, including at least the load demand value and corresponding time-of-use electricity price divided into 15-minute periods within the next 24 hours.
[0014] Preferably, the data association model based on the attention mechanism specifically involves mapping multi-dimensional operational data and grid time-of-use load and electricity price data to a high-dimensional feature space to obtain a data feature matrix. With load price characteristic matrix Construct a multi-head attention computation layer, using the formula Calculate attention weights, where For data feature query matrix, For the load price feature key matrix, This is the load price eigenvalue matrix. The feature dimension is defined by the multi-head attention output and the original data feature matrix. After residual connection, the feature is fused through a fully connected layer after LayerNorm normalization to form a data association model.
[0015] Preferably, the step of highlighting the data correlation between peak load periods and high electricity price periods through dynamic weight allocation to generate a fused feature dataset specifically involves defining a peak load judgment function. ,in for Time-of-use load demand The average load within the preset period, Define the peak load threshold coefficient; define the function for judging high or low electricity prices. ,in for Time-of-use electricity pricing The average electricity price over a preset period. This is the high electricity price threshold coefficient; obtained through the formula... calculate Time-period dynamic weights, where , Assign coefficients to the corresponding weights; associate the data characteristics of each time period with their corresponding weights. Multiply and concatenate to form a fused feature dataset.
[0016] Preferably, the objective function of maximizing time-of-use power generation revenue specifically means: using the sum of power generation revenue in each time period within a preset period as the objective function, expressed as follows: ,in For total revenue, The total number of time periods within the preset period. for Time-of-use electricity pricing for Power generation during the time period; the aforementioned Through formula Calculation, where for Total output of the unit during the period This refers to the duration of a single time period.
[0017] Preferably, the constraints include reservoir water balance constraints, unit output limitation constraints, and downstream ecological flow constraints; the expression for the reservoir water balance constraint is as follows: ,in , They are respectively Initial period Reservoir capacity at the end of the period for Inbound flow during specific time periods for Periodic power generation outflow for Downstream ecological flow during the time period; the expression for the unit output limit constraint is as follows: ,in To minimize the technical output of the unit, The rated output of the unit; the downstream ecological flow constraint expression is as follows: ,in This represents the minimum ecological flow downstream.
[0018] Preferably, the introduction of a deep reinforcement learning algorithm to construct a dynamic control model specifically involves defining the operating state of the hydropower station as a state space. ,include Reservoir capacity during the period Inbound flow Power grid load and electricity price Define the unit output adjustment amount as the action space. Construct a reward function ,in for Revenue from electricity generation during specific time periods The load deviation penalty coefficient is used; the DQN algorithm is used to construct the model network, including an experience replay pool, a target network, and an evaluation network. The evaluation network outputs each action. Value, the target network is determined by the formula Calculation target Value, of which For the goal value, for Candidate actions for a given time period For the target network Value function, for The state of the time period For the target network parameters, The discount factor is used; the evaluation is minimized using the gradient descent method. Values and Objectives The loss function of the value is used to complete the construction of the dynamic control model.
[0019] Preferably, the step of iteratively optimizing the model parameters to ensure that the time-sharing power generation plan output by the model satisfies the constraints and maximizes the objective function value specifically involves: initializing the dynamic control model parameters, setting the maximum number of iterations, dividing historical data and simulated data into training and validation sets, extracting a sample set from the training set in each iteration, calculating the target value through the target network, and updating the parameters using the gradient descent method. ,in For loss function, The learning rate; synchronization in each iteration = Calculate the objective function value using the validation set. The formula is: ,in For the validation set The power generation output by the time period model, when Stop iterating when there is no improvement after continuous iterations or when the preset number of iterations is reached, and save the target parameters. .
[0020] Preferably, the step of generating the optimal unit output value for each time period to form a time-sharing power generation control scheme for the hydropower station specifically involves: inputting real-time collected multi-dimensional operating data and grid time-sharing load and electricity price update data into the trained dynamic control model, and the model adjusts the output value according to the current state. Output motion space Each action Value; Selection The unit output adjustment corresponding to the action with the largest value is calculated in combination with the current unit output. Optimal total output of generating units during the time period The formula is: , The target parameter is used; the parameters for each time period are generated sequentially according to the order of the time periods within the preset period. This leads to a structured time-sharing power generation control scheme for hydropower stations.
[0021] Preferably, the dynamic control for realizing the time-of-use power generation revenue of the hydropower station specifically involves: issuing power adjustment commands to each generating unit according to the time-of-use power generation control scheme; the commands include the target output value and adjustment duration; and collecting the actual output in real time through the unit's sensors. Calculate the output force deviation Set deviation threshold ,when When the dynamic control model is triggered, it rereads the current operating status and updated grid data, recalculates the optimal unit output value and updates the control scheme, and calculates the actual power generation revenue according to the preset period, and fine-tunes the model parameters periodically to achieve dynamic control.
[0022] Compared with the prior art, the technical solution of this application has the following technical effects:
[0023] This invention constructs a data association model based on an attention mechanism to perform spatiotemporal matching and dynamic weight allocation of multi-dimensional operational data with grid time-of-use load and electricity price data. This can accurately capture the key features of peak load periods and high electricity price periods, strengthen the correlation between data, and provide a high-quality, targeted fusion feature dataset for the accurate formulation of subsequent power generation plans, effectively enhancing the data's supporting role in optimizing power generation revenue.
[0024] This invention takes maximizing time-of-use power generation revenue as the objective function, and combines constraints such as reservoir water balance, unit output limits, and downstream ecological flow. It introduces a deep reinforcement learning algorithm to construct a dynamic control model, making full use of the algorithm's ability to learn and optimize complex dynamic systems. Under the premise of satisfying various constraints, it achieves optimal planning of unit output for each time period, fundamentally ensuring the maximization of power generation revenue.
[0025] This invention uses historical operating data and grid load fluctuation simulation data to train a dynamic control model. By iteratively optimizing the model parameters, the model can better adapt to changes in actual operating scenarios. The output time-sharing power generation plan not only meets the constraints but also maximizes the objective function value, significantly enhancing the scientific nature and feasibility of the power generation plan.
[0026] In the real-time control phase, this invention inputs the real-time collected data into the trained dynamic control model to generate the optimal unit output value, forming a power generation control scheme. Through real-time monitoring and deviation triggering mechanisms, the control scheme is adjusted in a timely manner, realizing the dynamic adjustment and continuous optimization of the hydropower station's time-of-use power generation revenue. This ensures that the power generation revenue remains at a relatively optimal level even under dynamic changes in grid load and electricity prices.
[0027] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.
[0028] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description
[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0030] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:
[0031] Figure 1 Flowchart of a dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand;
[0032] Figure 2 This is a schematic diagram of the data association model structure based on the attention mechanism;
[0033] Figure 3 Module diagram of a hydropower station time-sharing dynamic control system that integrates grid load demand;
[0034] Figure 4 This is a comparison chart of the power generation revenue of the method in this application and the traditional static method in a single time period;
[0035] Figure 5 A bar chart comparing the power output matching degree of the proposed method and the traditional static method at different key time periods;
[0036] Figure 6 The graph shows the iterative convergence of the objective function value of the method in this application.
[0037] Figure 7 This is a comparison chart of the output deviation between the method of this application and the traditional static method. Detailed Implementation
[0038] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.
[0039] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.
[0040] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.
[0041] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.
[0042] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.
[0043] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.
[0044] Example 1
[0045] This embodiment mainly describes a dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand, such as... Figure 1 As shown, it specifically includes:
[0046] Acquire multi-dimensional operational data of hydropower stations and time-of-use load and electricity price data of the power grid;
[0047] A data association model is constructed based on the attention mechanism. Multi-dimensional operational data is spatiotemporally matched with power grid time-of-use load and electricity price data. Dynamic weight allocation highlights the data correlation between peak load periods and high electricity price periods, generating a fused feature dataset.
[0048] With the objective function of maximizing the revenue from time-of-use power generation, and with constraints such as reservoir water balance, unit output limits and downstream ecological flow, a dynamic control model is constructed by introducing a deep reinforcement learning algorithm.
[0049] The dynamic control model is trained using historical operating data and grid load fluctuation simulation data. The model parameters are iteratively optimized to ensure that the time-of-use generation plan output by the model meets the constraints and maximizes the objective function value.
[0050] The real-time collected multi-dimensional operation data and the grid time-sharing load and electricity price update data are input into the trained dynamic control model to generate the optimal unit output value for each time period and form a time-sharing power generation control scheme for the hydropower station.
[0051] Based on the time-of-use power generation control scheme, the power adjustment command is issued to the hydropower station units, and the deviation between the actual output of the units and the control scheme is monitored in real time to realize the dynamic control of the time-of-use power generation revenue of the hydropower station.
[0052] Furthermore, the acquisition of the multi-dimensional operational data and the grid time-of-use load and electricity price data specifically includes:
[0053] Two types of key data are acquired through specific acquisition methods and interfaces to support subsequent control processes. Multi-dimensional operational data are acquired through a distributed sensor network. Sensors are deployed to cover reservoir water level monitoring points, inflow monitoring sections, and unit operation status monitoring modules. The acquired data includes at least the real-time water level of the reservoir (reflecting the reservoir's water storage capacity), inflow (reflecting the water supply situation), and unit operating parameters (such as current output and speed), ensuring a comprehensive reflection of the hydropower station's own operating status.
[0054] The time-of-use load and electricity price data of the power grid are obtained through the data interface with the power grid dispatch center. The data covers a future preset period, typically 24 hours, and is divided into multiple time periods at fixed time intervals, with a common interval of 15 minutes. Each time period corresponds to a set of data, namely the load demand value of that time period (reflecting the intensity of power grid electricity demand) and the corresponding electricity price (determining the level of power generation revenue). Through the collaborative acquisition of these two types of data, a basic data source is provided for subsequent data association and model construction.
[0055] Furthermore, such as Figure 2 As shown, the feature dataset is constructed and fused using a data association model based on an attention mechanism, specifically as follows:
[0056] The acquired multi-dimensional operational data, along with the grid time-of-use load and electricity price data, are mapped to a high-dimensional feature space, transforming them into a data feature matrix that can be processed by a computer. (Multi-dimensional operational data feature matrix) and load-price feature matrix (Power grid time-of-use load and electricity price data feature matrix); subsequently, a multi-head attention calculation layer is constructed, using the formula... Calculate attention weights, where To extract from the data feature matrix The query matrix extracted from it is used to match the characteristics of power grid data; To obtain the load-price characteristic matrix The key matrix extracted is used to calculate similarity with the query matrix; To obtain the load price characteristic matrix The value matrix extracted from it is used to generate the attention output; The feature dimension is used to avoid excessively large values during calculation; after calculating the attention weights, the multi-head attention output is compared with the original data feature matrix. Residual connections are performed to supplement the original data information, and then LayerNorm normalization is used to eliminate data distribution differences. Feature fusion is completed through a fully connected layer to form a complete data association model.
[0057] In the dynamic weight allocation and feature dataset generation stage, two judgment functions are first defined to identify key time periods: load peak judgment function. ,in for Time-of-use load demand The average load within the preset period, The peak load threshold coefficient (set according to the power grid operating characteristics), when Reaching or exceeding average load When the load is doubled, it is determined to be a peak load period; the function for judging whether the electricity price is high or low. ,in for Time-of-use electricity pricing The average electricity price over a preset period. The high electricity price threshold coefficient (set according to electricity market rules) is when Reaching or exceeding the average electricity price When the price is doubled, it is determined to be a period of high electricity price; then, the formula is used... calculate Time-period dynamic weights, where , The weighting coefficients are adjusted based on revenue targets and grid demand, assigning data characteristics and corresponding weights to different time periods. Multiply and concatenate the data according to the time series to form a fused feature dataset, highlighting the impact of data from key time periods on subsequent models.
[0058] A dynamic control model is constructed by setting the objective function and constraints. The core objective of the objective function is to maximize the revenue from time-of-use power generation within a preset period. ,in The total revenue within the preset period. The total number of time periods within the preset period. for Time-of-use electricity pricing for Electricity generation during a given period; and Through formula Calculation, where for Total output of generating units during a given time period Given the duration of a single time period, this objective function directly links power generation revenue to power output and electricity price in each time period, thus clarifying the direction of optimization.
[0059] The constraint settings include three key types of constraints: reservoir water balance constraints, expressed as follows: ,in , They are respectively Initial period Reservoir capacity at the end of the period for Inbound flow during specific time periods for Periodic power generation outflow for Downstream ecological flow during a given period ensures that reservoir water volume fluctuates within a reasonable range; unit output constraints are expressed as follows: ,in Minimize the technical output of the unit (avoid inefficient operation of the unit). The rated output of the unit (to prevent overload damage); downstream ecological flow constraints, expressed as follows: ,in This is the minimum ecological flow downstream (to protect the downstream ecological environment).
[0060] The dynamic control model is constructed using a deep reinforcement learning algorithm. First, the state space is defined. With action space State space Include Reservoir capacity during the period Inbound flow Power grid load and electricity price It fully reflects the current operating environment; action space Defined as the unit output adjustment amount, the executable operation of the model is clarified; then, the reward function is constructed. ,in for Revenue from electricity generation during specific time periods The load deviation penalty coefficient (penalizing situations where output and load are mismatched) guides the model to align with grid load demand while pursuing revenue. The DQN algorithm is used to build the model network, which includes an experience replay pool (storing historical state-action-reward-next state data), a target network, and an evaluation network. The evaluation network outputs each action. Value (action value), the target network uses a formula Calculation target value (where For the target Q value, for Time-based candidate actions, The target network's Q-value function, for Time period status, For the target network parameters, (where is the discount factor), and the loss function is minimized using gradient descent to complete the construction of the dynamic control model.
[0061] Before training and optimizing the dynamic control model, data preparation and parameter initialization must be completed: historical operating data (including past hydropower station operating data and power grid data) and power grid load fluctuation simulation data (generated through Monte Carlo method or historical load fluctuation patterns) are divided into training set and validation set according to a preset ratio; the dynamic control model parameters are initialized. (Including the weights and biases of the evaluation network and the target network), and setting the maximum number of iterations. Synchronization interval with parameters .
[0062] The training process employs an iterative optimization approach: in each iteration, a batch of samples is randomly drawn from the training set, and the network is evaluated to calculate the performance of each sample. Calculate the target value through the target network Define the loss function The model parameters are updated using gradient descent. (in The learning rate controls the parameter update step size (e.g., 0.001). Every M iterations, the network parameters are evaluated. Synchronize to target network Simultaneously, the objective function value is calculated using the validation set. (in For the validation set The power generation output by the time period model), when continuous The next iteration either shows no improvement or reaches the maximum number of iterations. When the time is right, stop the iteration and save the current optimal parameters. This ensures that the time-of-use generation plan output by the model meets the constraints and maximizes the objective function value.
[0063] Furthermore, the real-time optimal unit output value generation and time-of-use power generation control scheme are as follows:
[0064] During the real-time data input phase, multi-dimensional operational data (updated reservoir water level, inflow, and current unit output) will be collected in real time through a distributed sensor network and grid dispatch interface, along with grid time-of-use load and electricity price update data (the latest time period). ), and convert them into state vectors that conform to the model input format. Input the trained dynamic control model (load the optimal parameters) ).
[0065] The optimal unit output value calculation and control scheme formation process: The model is based on the state vector Output motion space Each action value Select the unit output adjustment corresponding to the action with the largest Q value. Combined with the current unit output Through formula calculate Optimal total output of generating units during the time period Then, according to the order of the time periods within the preset period (from time period 1 to time period T), the result of each time period is calculated sequentially. It also supplements the corresponding reservoir capacity changes for each time period (derived through the water balance formula) and power generation outflow (based on the turbine characteristic curve). calculate, for The net water head (period) and ecological flow guarantee value are used to form a system that includes the time period number and target output. A structured time-sharing power generation control scheme with matching flow parameters ensures that the scheme can be directly used for unit scheduling.
[0066] Furthermore, the aforementioned real-time output monitoring and dynamic control specifically refers to:
[0067] In the output adjustment command issuance and monitoring phase, based on the generated time-sharing power generation control scheme, the PLC control system issues output adjustment commands to each hydropower station unit. The command content clearly includes the target output value of each unit (based on the total output). Allocation based on unit efficiency and adjustment duration (and time period duration) (Consistent); simultaneously, the actual output at each time period is collected in real time through the unit's sensors. Calculate the output force deviation .
[0068] The dynamic adjustment and continuous optimization process includes setting a preset deviation threshold. (Set according to power grid frequency regulation requirements, such as 5% of rated output), when At that time, the dynamic control model is triggered to reread the current running state (updated). Based on the latest grid data, the optimal unit output value is recalculated. And update the time-of-use power generation control scheme to avoid the deviation from widening; at the same time, calculate the actual power generation revenue according to a fixed period (such as daily) (based on...). (Calculated against actual electricity prices), the deviation between the revenue data and the model's predicted revenue is fed back to the model parameter optimization module, which then optimizes the parameters. Regular fine-tuning (such as retraining the model monthly) is performed to achieve dynamic control of the time-of-use power generation revenue of the hydropower station, ensuring long-term stability of the control effect.
[0069] This implementation details how, through multi-dimensional data acquisition, attention mechanism data association, deep reinforcement learning model construction and training, and real-time control, precise dynamic regulation of hydropower generation is achieved. This solves the problem that traditional hydropower generation control methods are difficult to adapt to the temporal fluctuations of grid load and time-of-use pricing mechanisms. It allows hydropower stations to maximize time-of-use generation revenue while meeting constraints such as reservoir water balance, unit output limits, and downstream ecological flow, thereby improving the economic efficiency of hydropower station operation and its responsiveness to the grid.
[0070] Example 2
[0071] This embodiment details a dynamic control system for a method of dynamically controlling the time-of-use generation revenue of hydropower stations to meet grid load demands. It includes a data acquisition module, a data association module, a model building module, a model training module, a scheme generation module, and a dynamic control module. Figure 3 As shown, specifically:
[0072] The data acquisition module is responsible for collecting multi-dimensional operational data of the hydropower station and time-of-use load and electricity price data of the power grid. For hydropower station operational data, it acquires reservoir water level, inflow, and unit operating parameters (such as current output and equipment status) in real time through distributed sensing devices. For power grid data, it acquires time-of-use load demand data of the power grid divided into fixed time periods within a future preset period and electricity price data for the corresponding time periods through a dedicated data interface with the power grid dispatching system, providing a basic information source for subsequent data processing and model building.
[0073] The data association module, based on an attention mechanism, processes the collected multi-dimensional operation data of hydropower stations and the time-of-use load and electricity price data of the power grid. First, it matches the two types of data in the spatiotemporal dimension to clarify the correspondence between different data in time and space. Then, through a dynamic weight allocation strategy, it highlights the degree of data correlation between peak load periods and high electricity price periods, thereby generating a fused feature dataset that integrates the characteristics of key periods, making the data more reflective of the impact on power generation revenue.
[0074] The model building module aims to maximize the revenue from time-of-use power generation. It incorporates constraints such as reservoir water balance, unit output limits, and downstream ecological flow, and introduces a deep reinforcement learning algorithm to construct a dynamic control model. During the construction process, the model's state space (including the hydropower station's operating status and grid data status), action space (operations such as unit output adjustment), and reward function (related to power generation revenue and constraint satisfaction) are determined, thus establishing a model framework capable of making power generation optimization decisions.
[0075] The model training module uses historical operating data and simulated grid load fluctuation data to train the constructed dynamic control model. During the training process, the model parameters are continuously iterated and optimized so that the output time-sharing generation plan can maximize the objective function (time-sharing generation revenue) value while meeting various constraints, thereby improving the model's decision-making accuracy.
[0076] The scheme generation module inputs real-time collected multi-dimensional operation data of the hydropower station and updated grid time-sharing load and electricity price data into the trained dynamic control model; the model performs calculations based on the input data to generate the optimal output value of the hydropower station units for each time period. These output values are integrated in the order of time periods to form a complete time-sharing power generation control scheme for the hydropower station, which is used to guide the power generation operation of the units.
[0077] Based on the generated time-of-use power generation control scheme, the dynamic control module issues specific output adjustment commands to the hydropower station units, directing them to adjust their power generation output. Simultaneously, it monitors the actual output of the units in real time, comparing it with the output value specified in the control scheme and calculating the deviation. If the deviation exceeds the allowable range, the model is triggered to recalculate and update the control scheme, achieving dynamic control of the hydropower station's time-of-use power generation revenue and ensuring that power generation revenue remains stable at an optimal level.
[0078] Furthermore, a dynamic control method for the time-of-use generation revenue of hydropower stations, which integrates grid load demand, is implemented through a dynamic control system. Specifically:
[0079] The data acquisition module is activated, simultaneously collecting multi-dimensional operational data of the hydropower station and time-of-use load and electricity price data of the power grid.
[0080] The collected data is transmitted to the data association module, which performs spatiotemporal matching and dynamic weight allocation on the data based on the attention mechanism to generate a fused feature dataset.
[0081] The model building module integrates feature datasets as input, and combines the objective function and constraints with a deep reinforcement learning algorithm to construct a dynamic control model.
[0082] The model training module calls historical operating data and power grid load fluctuation simulation data to train the dynamic control model and iteratively optimize the model parameters;
[0083] The trained model is called by the scheme generation module. After the new data collected in real time is input into the model, the optimal unit output value for each time period is generated, forming a time-sharing power generation control scheme.
[0084] The dynamic control module issues unit output adjustment commands based on the control scheme, while simultaneously monitoring the actual unit output in real time. It dynamically adjusts the control scheme according to the deviation, thereby achieving dynamic control of time-of-use power generation revenue.
[0085] This embodiment describes in detail the dynamic control system. It efficiently acquires multi-source data through a data acquisition module, strengthens data association during key periods using an attention mechanism through a data association module, builds and trains a precise decision-making model using deep reinforcement learning through a model building and training module, and coordinates the operation of a scheme generation and dynamic control module. This achieves intelligent and automated control of the entire hydropower generation process, solving the problems of poor connection between various links and lack of intelligent decision-making capabilities in traditional control systems. It provides stable and reliable system support for the dynamic control of time-sharing power generation revenue of hydropower stations that integrates grid load demand, and ensures the effective implementation of the control method.
[0086] Based on Embodiment 1 or 2, this embodiment details the technical verification of a dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand. This verification was conducted at a medium-sized hydropower station (total installed capacity 120MW, total reservoir capacity 8.5×10⁻⁶). 7 m 3 The data was collected from actual operation data between March 1st and March 7th, 2024. Data sources included multi-dimensional operational data collected every 15 minutes from the hydropower station's SCADA system (reservoir water level, inflow, unit output, etc.), concurrent time-of-use load and electricity price data provided by the power grid dispatch center (peak period 0.68 yuan / kWh, average period 0.42 yuan / kWh, valley period 0.25 yuan / kWh), as well as historical data from the entire year of 2023 and 100 sets of simulated power grid load fluctuation data. All data underwent dual verification by the hydropower station's operation and maintenance department and the power grid dispatch center, demonstrating the reliability of the technical effect of this application. Specifically:
[0087] like Figure 4 As shown, these correspond to the method of this application and the traditional static method (a scheduling method based on historical experience or simple mathematical models to formulate fixed power generation plans), respectively. Figure 4 It is clearly evident that the time-of-use generation revenue curve of the proposed method is consistently higher than that of the traditional method, especially during peak load periods and periods with overlapping high electricity prices, such as 9:00-11:00 and 18:00-20:00, where the difference is most significant. Experimental data shows that the total 24-hour revenue of the proposed method reaches 286,000 yuan, while that of the traditional method is 223,000 yuan, representing a 28.2% increase in revenue. Specifically, the average revenue per period during overlapping periods is 8,200 yuan, compared to 5,700 yuan for the traditional method, an increase of 43.9%. During off-peak periods from 0:00 to 6:00, the average revenue per period is 3,100 yuan, compared to 2,500 yuan for the traditional method, an increase of 24.0%. This fully demonstrates that the proposed method can accurately match the grid's time-of-use load and electricity price, achieving high-revenue generation during critical periods and avoiding revenue waste during off-peak periods, thus significantly improving total revenue.
[0088] like Figure 5As shown, the two sets of bars correspond to the method of this application and the traditional static method, respectively. The data in the figure shows that the output matching degree of the method of this application is significantly higher than that of the traditional method in the three key periods, especially in the "peak load + high electricity price overlap period", where the matching degree is close to 100%. Specifically, the test data are as follows: during the peak load period (Lt≥Lavg×1.2), the matching degree of this application is 92.3%, while that of the traditional method is 68.5%, an improvement of 23.8 percentage points; during the high electricity price period (Pt≥Pavg×1.3), the matching degree of this application is 90.7%, while that of the traditional method is 71.2%, an improvement of 19.5 percentage points; and during the overlap period, the matching degree of this application is 98.1%, while that of the traditional method is 65.3%, an improvement of 32.8 percentage points. This result verifies that this application can accurately capture the grid demand during key periods through attention mechanism and dynamic weight allocation, so that the unit output and load demand are highly matched, avoiding the problem of "output and demand decoupling" in the traditional method.
[0089] like Figure 6 As shown, it can be observed that with the increase of the number of iterations, the objective function value of the validation set of the dynamic control model of this application gradually increases and tends to stabilize. In the first 100 iterations, it rapidly increases from 205,000 yuan to 268,000 yuan, with a fast convergence speed. After 500 iterations, it stabilizes in the range of 285,000-287,000 yuan, with a fluctuation range of ≤0.7%, reaching a convergent and stable state. In contrast, the traditional static method has no iterative optimization process, and the objective function value remains stable at 223,000 yuan, with no room for improvement. This indicates that the dynamic control model of this application can continuously optimize parameters through training and finally output a power generation plan that satisfies the constraints and has the best returns. The reliability and optimization capability of the model are significantly better than those of the traditional method.
[0090] like Figure 7 As shown, the output deviation scatter points of the method in this application are all distributed below the 5MW threshold, with a small error fluctuation range. In contrast, the traditional static method has deviations exceeding 5MW in 12 time periods, with a maximum deviation of 12.3MW, showing significant deviation fluctuations. Specifically, the test data shows that the average output deviation of this application over 24 hours is 2.1MW, while that of the traditional method is 6.8MW, representing a 69.1% reduction in deviation. No time periods in this application have deviations exceeding 5MW, while the traditional method has deviations exceeding the threshold in 12 time periods, accounting for 50%. In particular, during the sudden fluctuation period of a 15% load increase from 14:00 to 15:00, the deviation of this application was quickly corrected to 1.8MW, while the deviation of the traditional method exceeded 8MW for 8 consecutive time periods. This fully demonstrates that the "real-time monitoring-deviation correction" mechanism of this application can quickly respond to changes in operating status, accurately control output deviations, avoid the problem of "difficult deviation correction" in the traditional method, and ensure the stable execution of the power generation plan.
[0091] This embodiment details the technical verification demonstrating that the method is indeed feasible, achieving precise matching between hydropower station power generation plans and grid load demand and time-of-use pricing. While ensuring the safe and stable operation of the hydropower station (meeting various constraints), it significantly improves power generation revenue, solving the problem of poor revenue caused by the disconnect between power generation plans and grid demand and market prices in the past. It verifies the significant role of the method in improving the operational efficiency of hydropower stations and grid adaptability.
[0092] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.
Claims
1. A dynamic control method for time-of-use generation revenue of hydropower stations that integrates grid load demand, characterized in that: include: Acquire multi-dimensional operational data of hydropower stations and time-of-use load and electricity price data of the power grid; A data association model is constructed based on the attention mechanism. Multi-dimensional operational data is spatiotemporally matched with power grid time-of-use load and electricity price data. Dynamic weight allocation highlights the data correlation between peak load periods and high electricity price periods, generating a fused feature dataset. With the objective function of maximizing the revenue from time-of-use power generation, and with constraints such as reservoir water balance, unit output limits and downstream ecological flow, a dynamic control model is constructed by introducing a deep reinforcement learning algorithm. The dynamic control model is trained using historical operating data and grid load fluctuation simulation data. The model parameters are iteratively optimized to ensure that the time-of-use generation plan output by the model meets the constraints and maximizes the objective function value. The real-time collected multi-dimensional operation data and the grid time-sharing load and electricity price update data are input into the trained dynamic control model to generate the optimal unit output value for each time period and form a time-sharing power generation control scheme for the hydropower station. Based on the time-of-use power generation control scheme, the power adjustment command is issued to the hydropower station units, and the deviation between the actual output of the units and the control scheme is monitored in real time to realize the dynamic control of the time-of-use power generation revenue of the hydropower station. The method of highlighting the data correlation between peak load periods and high electricity price periods through dynamic weight allocation and generating a fused feature dataset specifically involves defining a peak load judgment function. ,in for Time-of-use load demand The average load within the preset period, Define the peak load threshold coefficient; define the function for judging high or low electricity prices. ,in for Time-of-use electricity pricing The average electricity price over a preset period. This is the high electricity price threshold coefficient; obtained through the formula... calculate Time-period dynamic weights, where , Assign coefficients to the corresponding weights; associate the data characteristics of each time period with their corresponding weights. Multiply and concatenate to form a fused feature dataset.
2. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 1, characterized in that, The multi-dimensional operational data is collected through a distributed sensor network, including reservoir water level, inflow, and unit operating parameters, and includes at least real-time reservoir water level, inflow, and unit operating parameters. The grid time-of-use load and electricity price data are obtained through the grid dispatch data interface, which obtains the time-of-use load demand and corresponding electricity price within a future preset period, including at least the load demand value and corresponding electricity price for each 15-minute period within the next 24 hours.
3. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 1, characterized in that, The data association model based on the attention mechanism specifically involves mapping multi-dimensional operational data and grid time-of-use load and electricity price data to a high-dimensional feature space to obtain a data feature matrix. With load price characteristic matrix Construct a multi-head attention computation layer, using the formula Calculate attention weights, where For data feature query matrix, For the load price feature key matrix, This is the load price eigenvalue matrix. The feature dimension is defined by the multi-head attention output and the original data feature matrix. After residual connection, the feature is fused through a fully connected layer after LayerNorm normalization to form a data association model.
4. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 1, characterized in that, The objective function of maximizing time-of-use power generation revenue is specifically defined as follows: the sum of power generation revenue in each time period within a preset period is used as the objective function, expressed as follows: ,in For total revenue, The total number of time periods within the preset period. for Time-of-use electricity pricing for Power generation during the time period; the aforementioned Through formula Calculation, where for Total output of the unit during the period This refers to the duration of a single time period.
5. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 4, characterized in that, The constraints include reservoir water balance constraints, unit output limits, and downstream ecological flow constraints; the expression for the reservoir water balance constraint is as follows: ,in , They are respectively Initial period Reservoir capacity at the end of the period for Inbound flow during specific time periods for Periodic power generation outflow for Downstream ecological flow during the time period; the expression for the unit output limit constraint is as follows: ,in To minimize the technical output of the unit, The rated output of the unit; the downstream ecological flow constraint expression is as follows: ,in This represents the minimum ecological flow downstream.
6. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 5, characterized in that, The introduction of deep reinforcement learning algorithms to construct a dynamic control model specifically involves defining the operating state of the hydropower station as a state space. ,include Reservoir capacity during the period Inbound flow Power grid load and electricity price Define the unit output adjustment amount as the action space. Construct a reward function ,in for Revenue from electricity generation during specific time periods The load deviation penalty coefficient is used; the DQN algorithm is used to construct the model network, including an experience replay pool, a target network, and an evaluation network. The evaluation network outputs each action. Value, the target network is determined by the formula Calculation target Value, of which For the goal value, for Candidate actions for a given time period For the target network Value function, for The state of the time period For the target network parameters, Discount factor; Minimize the evaluation using gradient descent. Values and Objectives The loss function of the value is used to complete the construction of the dynamic control model.
7. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 6, characterized in that, The method of iteratively optimizing model parameters to ensure that the time-of-use power generation plan output by the model satisfies the constraints and maximizes the objective function value specifically involves: initializing the dynamic control model parameters, setting the maximum number of iterations, dividing historical data and simulated data into training and validation sets, extracting a sample set from the training set in each iteration, calculating the target value through the target network, and updating the parameters using the gradient descent method. ,in For loss function, For learning rate, To evaluate network parameters; synchronization in each iteration = Calculate the objective function value using the validation set. The formula is: ,in For the validation set The power generation output by the time period model, when Stop iterating when there is no improvement after continuous iterations or when the preset number of iterations is reached, and save the target parameters. .
8. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 7, characterized in that, The process of generating optimal unit output values for each time period to form a time-sharing power generation control scheme for the hydropower station specifically involves: inputting real-time collected multi-dimensional operational data and grid time-sharing load and electricity price update data into a trained dynamic control model; and adjusting the model according to the current state. Output motion space Each action Value; Selection The unit output adjustment corresponding to the action with the largest value is calculated in combination with the current unit output. Optimal total output of generating units during the time period The formula is: , For the target parameter, The actual output for the previous period; generate the output for each period sequentially according to the order of the periods within the preset cycle. This leads to a structured time-sharing power generation control scheme for hydropower stations.
9. The method for dynamic control of time-of-use generation revenue of hydropower stations that integrates grid load demand as described in claim 8, characterized in that, The dynamic control for realizing the time-of-use power generation revenue of the hydropower station specifically involves: issuing power adjustment commands to each generating unit according to the time-of-use power generation control scheme. The commands include the target output value and the adjustment duration, and the actual output is collected in real time by the unit's sensors. Calculate the output force deviation Set deviation threshold ,when When the dynamic control model is triggered, it rereads the current operating status and updated grid data, recalculates the optimal unit output value and updates the control scheme, and calculates the actual power generation revenue according to the preset period, and periodically fine-tunes the model parameters to achieve dynamic control.