Pumped storage power station carbon evaluation and optimization method and system

Abstract

The invention discloses a pumped storage power station carbon evaluation and optimization method and system, and relates to the technical field of energy storage equipment control, and the method comprises the steps: arranging and collecting the operation data of a sensor from a pumped storage power station and the surrounding environment thereof, and carrying out the preprocessing of the operation data, and obtaining formatted data; and based on the formatted data, constructing a deep reinforcement learning model in cooperation with a deep deterministic strategy gradient algorithm, and re-inputting the operation data of the sensor into the deep reinforcement learning model to obtain current and future carbon emission evaluation results. Through the novel optimization method combining deep reinforcement learning and the sensor network, the problems that a traditional pumped storage power station optimization method neglects the influence of an external environment, lacks perspectiveness and is insufficient in dynamic adaptability are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage device control, and particularly to a carbon assessment and optimization method and system for a pumped storage power station. Background Art

[0002] With the increasing global demand for renewable energy, pumped storage power stations, as an important part of energy storage technology, play an increasingly prominent role. A pumped storage power station stores excess electrical energy as potential energy in the upper reservoir during low electricity demand periods and then converts it back into electrical energy for release during peak electricity consumption periods, thereby effectively regulating the power grid load.

[0003] Existing optimization technologies for pumped storage power stations have improved the operation efficiency of power stations to a certain extent, but these methods generally have some limitations. Firstly, traditional methods often focus on the direct control of internal equipment in the power station and ignore the impact of external environmental factors (such as temperature, humidity, etc.) on the operation of the power station, resulting in incomplete optimization schemes. Secondly, existing evaluation models mostly adopt static analysis means and cannot dynamically adapt to changing operating conditions, which limits the prediction accuracy and response speed. In addition, current optimization strategies are usually based on historical data analysis, lacking foresight and unable to effectively predict future carbon emission trends. Finally, when implementing optimization strategies, existing technologies rarely consider the uncertainties and complexities in specific operations, which may lead to poor optimization effects or even countereffects. Summary of the Invention

[0004] Aiming at the defects existing in the prior art, the present invention provides a carbon assessment and optimization method and system for a pumped storage power station, which is a new optimization method combining deep reinforcement learning and a sensor network, capable of real-time monitoring and adjusting the operation parameters of the power station, effectively reducing carbon emissions, and improving the overall operation efficiency, and solving the problems that existing optimization methods for pumped storage power stations ignore the impact of the external environment and predict future carbon emission trends and adjust optimization strategies in real time.

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a carbon assessment and optimization method for a pumped storage power station, including:

[0007] Collecting the operation data of sensors arranged in and around the pumped storage power station and preprocessing the operation data to obtain formatted data;

[0008] Based on the formatted data, constructing a deep reinforcement learning model in cooperation with the deep deterministic policy gradient algorithm;

[0009] Re-input the operation data of the sensor into the deep reinforcement learning model to obtain the current and future carbon emission assessment results;

[0010] Formulate specific optimization strategies based on the carbon emission assessment results to obtain an optimization plan for reducing carbon emissions and implement it;

[0011] Monitor the implementation effect of the optimization plan and feedback its monitoring data to the deep reinforcement learning model.

[0012] As a preferred embodiment of the carbon assessment and optimization method for the pumped storage power station of the present invention, wherein:

[0013] The steps of obtaining formatted data by arranging and collecting the operation data of sensors from the pumped storage power station and its surrounding environment and preprocessing the operation data are as follows:

[0014] Install multiple sensors in the pumped storage power station, and each sensor is equipped with a LoRaWAN wireless communication module and transmits operation data to the central server;

[0015] Use a filtering algorithm to remove abnormal data points from the operation data and synchronize the timestamps of all operation data using GPS clocks;

[0016] Convert the operation data with timestamps into a unified JSON format to obtain unified format operation data;

[0017] Deep clean the unified format operation data, fill in missing values and correct biases to obtain formatted data.

[0018] As a preferred embodiment of the carbon assessment and optimization method for the pumped storage power station of the present invention, wherein:

[0019] The steps of constructing a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm are as follows:

[0020] Extract the pressure reading P and the temperature reading Temp from the formatted data as components of the state vector S current ;

[0021] Adopt the Actor-Critic structure to calculate the value Q(S current , Act) of the state-action pair, and the formula is as follows:

[0022] Q(S current , Act) = w1·P + w2·Temp + w3·Flow + w4·Act - w5·e -λ·τ (1)

[0023] Where: w1 is the weight coefficient for adjusting the pressure P, w2 is the weight coefficient for adjusting the temperature Temp, w3 is the weight coefficient for adjusting the water flow rate Flow, w4 is the weight coefficient for adjusting the action Act, w5 is the influence weight coefficient for adjusting the time decay term, λ is the weight coefficient for controlling future rewards, and τ refers to the current moment;

[0024] After calculating the value Q(S current , Act) of the state-action pair, initialize the weights of the Actor network and the Critic network, and construct a deep reinforcement learning model in cooperation with the deep deterministic policy gradient algorithm.

[0025] As a preferred solution of the carbon assessment and optimization method for the pumped storage power station described in the present invention, where:

[0026] The step of re-inputting the operation data of the sensor into the deep reinforcement learning model to obtain the current and future carbon emission assessment results is as follows:

[0027] Extract the pressure reading P, the temperature reading Temp, and the water flow rate Flow from the operation data of the sensor and combine them into a state vector S current = [P, Temp, Flow];

[0028] Use the state vector S current Through the Actor network, generate the optimal action Act, and substitute it into formula (1) to obtain Q(S current , Act) again;

[0029] By substituting the re-obtained Q(S current , Act) into the random forest algorithm, obtain the current and future carbon emission assessment results.

[0030] As a preferred solution of the carbon assessment and optimization method for the pumped storage power station described in the present invention, where:

[0031] The step of formulating a specific optimization strategy according to the carbon emission assessment result, obtaining an optimization plan for reducing carbon emissions and implementing it is as follows:

[0032] Based on the obtained current and future carbon emission assessment results, compare the carbon emission values in the time period and take the maximum value;

[0033] Identify the operations or conditions with high carbon emissions according to the time period of the maximum carbon emission value, and analyze and extract the pressure reading P, the temperature reading Temp, and the water flow rate Flow;

[0034] Correspondingly modify the operation values according to the obtained pressure reading P, temperature reading Temp, and water flow rate Flow to obtain an optimization plan for reducing carbon emissions;

[0035] Modify the equipment settings according to the optimized solution for reducing carbon emissions and implement it.

[0036] As a preferred solution of the carbon assessment and optimization method for the pumped - storage power station of the present invention, wherein:

[0037] By monitoring the implementation effect of the optimized solution and feeding its monitoring data back to the deep reinforcement learning model, the specific steps are as follows:

[0038] Obtain the pressure reading P, temperature reading Temp, and water flow rate Flow after the implementation of the optimized solution from the sensor;

[0039] And recombine the pressure reading P, temperature reading Temp, and water flow rate Flow after implementation into S current =[P, Temp, Flow];

[0040] For the recombined S current =[P, Temp, Flow], generate a new optimal action Act through the Actor network, and input the new optimal action Act and S current =[P, Temp, Flow] into the deep reinforcement learning model to calculate the value Q(S current , Act);

[0041] According to the new data and the calculated value Q(S current , Act) of the state - action pair, use the gradient descent optimization algorithm to update the weights of the Actor and Critic networks;

[0042] Regularly repeat the above process to continuously optimize the parameters of the deep reinforcement learning model and the optimization strategy, ensuring that the system is always in the best operating state.

[0043] In a second aspect, the present invention provides a carbon assessment and optimization system for a pumped - storage power station, including:

[0044] A data collection and pre - processing module, which obtains the operation data of sensors by arranging and collecting them from the pumped - storage power station and its surrounding environment, and pre - processes the operation data to obtain formatted data;

[0045] A deep reinforcement learning model construction module, which constructs a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm;

[0046] A carbon emission assessment module, which re - inputs the operation data of the sensors into the deep reinforcement learning model to obtain the current and future carbon emission assessment results;

[0047] An optimization strategy formulation and implementation module formulates specific optimization strategies according to the carbon emission assessment results, obtains an optimization plan for reducing carbon emissions and implements it;

[0048] A monitoring and feedback module monitors the implementation effect of the optimization plan and feeds its monitoring data back to the deep reinforcement learning model.

[0049] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the carbon assessment and optimization method for a pumped-storage power station as described in the first aspect of the present invention is implemented.

[0050] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the carbon assessment and optimization method for a pumped-storage power station as described in the first aspect of the present invention is implemented.

[0051] The beneficial effects of the present invention are as follows: Through the combination of deep reinforcement learning and a sensor network, the real-time monitoring and dynamic response to external environmental factors (such as temperature, humidity, etc.) are realized, and the comprehensiveness and accuracy of the optimization plan are improved. The use of dynamic analysis to replace traditional static means significantly improves the prediction accuracy and response speed, and enhances the forward-looking and adaptability of the system. In particular, the present invention fully considers the uncertainties and complexities in operation, effectively avoiding the problem of poor optimization effects. In summary, the present invention not only greatly improves the operation efficiency of the power station, but also effectively reduces carbon emissions, supports the efficient utilization of clean energy, and has significant technological progress significance and broad application prospects. Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0053] Figure 1 It is a flowchart of the carbon assessment and optimization method for a pumped-storage power station in Embodiment 1. Detailed Embodiments

[0054] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0055] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways than those specifically described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0056] Secondly, as used herein, an "embodiment" or "embodiments" refer to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an isolated or alternative embodiment that excludes other embodiments.

[0057] Embodiment 1, referring to Figure 1 , is the first embodiment of the present invention. This embodiment provides a method for carbon assessment and optimization of a pumped-storage power station, including the following steps:

[0058] S1. By arranging and collecting the operation data of sensors from the pumped-storage power station and its surrounding environment, and preprocessing the operation data, formatted data is obtained.

[0059] Specifically, a plurality of sensors are installed in the pumped-storage power station, and each sensor is equipped with a LoRaWAN wireless communication module and transmits operation data to a central server.

[0060] It should be noted that according to the requirements of the power station operation parameters, sensors capable of measuring key indicators such as pressure, temperature, and water flow rate are selected, and the sensors are reasonably arranged inside and in the surrounding environment of the pumped-storage power station to ensure that all important areas are covered. Each sensor is equipped with a LoRaWAN wireless communication module to achieve data transmission with the central server. Parameters such as the working frequency and transmission power of the LoRaWAN module are set to ensure stable and reliable long-distance data transmission. By adopting LoRaWAN technology, comprehensive monitoring of the power station environment is achieved, and the real-time and reliability of data transmission are ensured. This is crucial for dynamically adjusting the operation state of the power station. Especially in a power station environment with complex terrain and wide distribution, LoRaWAN provides a low-power, long-distance data transmission solution, reducing the wiring cost and complexity and improving the data acquisition efficiency.

[0061] Specifically, a filtering algorithm is used to remove abnormal data points from the operation data, and the timestamps of all operation data are synchronized using a GPS clock.

[0062] It should be noted that algorithms such as Kalman filtering or median filtering are used to remove outliers from the original data to ensure data accuracy. The timestamps of all sensor data are synchronized using GPS clocks to ensure data consistency among sensors. The preprocessed data is preliminarily sorted and prepared for conversion to a unified format. Precise data cleaning and time synchronization significantly improve the basic quality of subsequent analysis, reduce decision-making errors caused by data errors, and enhance the reliability and accuracy of the overall system. For example, when constructing a deep reinforcement learning model, high-quality data input can significantly improve the learning effect and prediction accuracy of the model.

[0063] Specifically, the running data with timestamps is converted into a unified JSON format.

[0064] It should be noted that a JSON format template suitable for storing and exchanging power plant operation data is designed, including field names, types, and hierarchical relationships. The preprocessed data is mapped and converted according to the defined JSON template to generate a data file in a unified format. The converted data is stored in a central server and regularly backed up to prevent data loss. The unified data format simplifies the data management and sharing process, enhances system compatibility, enables seamless integration of data from different sources, and improves work efficiency. Especially in the process of multi-source data integration, the unified data format can reduce the complexity of data conversion, speed up the processing speed, and thus accelerate the response speed of the entire optimization process.

[0065] Specifically, the running data in the unified format is deeply cleaned to fill in missing values and correct biases to obtain formatted data.

[0066] It should be noted that the positions of missing values in the dataset are identified through data analysis tools. According to the trends of adjacent time periods or other relevant variables, methods such as linear interpolation or mean filling are used to fill in the missing values. Historical data is compared with current data, and statistical methods (such as standard deviation) are used to detect and correct possible biases. The deep cleaning process further improves the quality of the data, ensures that the data input into the model is as accurate as possible, and thus improves the effectiveness of the optimization strategy formulated based on this data. For example, in the adjustment of power plant operation parameters, the optimization strategy formulated based on high-quality data will be more accurate and effective, ultimately achieving the goal of reducing carbon emissions. In addition, high-quality data is crucial for improving the performance of deep reinforcement learning models because it directly affects the prediction ability and reliability of the models.

[0067] S2. Based on the formatted data, a deep reinforcement learning model is constructed in cooperation with the deep deterministic policy gradient algorithm.

[0068] Specifically, the pressure reading P and temperature reading Temp are extracted from the formatted data as the state vector S currentComponents.

[0069] It should be noted that pressure readings and temperature readings are selected from the preprocessed formatted data as part of the state vector. These parameters are key factors affecting the operating efficiency and carbon emissions of power plants. The selected data points (such as pressure readings and temperature readings) are standardized to ensure that data of different magnitudes can be uniformly input into the model. Standardization can be achieved by subtracting the mean and dividing by the standard deviation. The standardized data points are combined into a unified state vector for subsequent input into the deep reinforcement learning model. By selecting pressure readings and temperature readings as the core part of the state vector, the present invention directly models the most critical factors in power plant operation. This selection not only simplifies the complexity of the model but also improves the effectiveness of the model. Existing technologies often ignore the impact of external environmental factors on power plant operation, resulting in incomplete optimization solutions. The present invention realizes real-time monitoring and dynamic response to the external environment by introducing these key parameters, thereby formulating a more accurate optimization plan.

[0070] Specifically, the Actor-Critic structure is used to calculate the value Q(S current , Act) of the state-action pair, and the formula is derived as follows:

[0071] Q(S current , Act) = w1·P + w2·Temp + w3·Flow + w4·Act - w5·e -λ·τ ,

[0072] Where: w1 is the weight coefficient for adjusting the pressure P, w2 is the weight coefficient for adjusting the temperature Temp, w3 is the weight coefficient for adjusting the water flow rate Flow, w4 is the weight coefficient for adjusting the action Act, w5 is the weight coefficient for adjusting the influence of the time decay term, λ is the weight coefficient for controlling future rewards, and τ represents the current moment.

[0073] It should be noted that after calculating the value of the state-action pair according to the above formula, the weights of the Actor and Critic networks are initialized. This step ensures that the model can start learning from the initial state and gradually optimize its prediction ability. The model is initially trained using historical data, and the weight coefficients are adjusted to adapt to the actual application scenario. Using the Actor-Critic structure, the present invention can provide the optimal action suggestion while evaluating the current state. This method overcomes the problems of slow response speed and low prediction accuracy of traditional static analysis methods. Traditional optimization methods are usually based on historical data analysis and lack foresight. By dynamically calculating the value of the state-action pair, the present invention significantly improves the foresight and adaptive ability of the system, making the optimization strategy more flexible and efficient.

[0074] Specifically, after calculating the value Q(S current , Act) of the state-action pair, initialize the weights of the Actor and Critic networks, and construct a deep reinforcement learning model in conjunction with the Deep Deterministic Policy Gradient algorithm.

[0075] It should be noted that the DDPG algorithm is used to train the preliminarily constructed model. DDPG is a reinforcement learning method that combines the Actor-Critic architecture and deterministic policy gradients, and is particularly suitable for dealing with problems in continuous action spaces.

[0076] Tuning parameters: Continuously adjust the model parameters during the training process, including weight coefficients and learning rates, etc., to improve the accuracy and stability of the model. Verify the effect of the model through simulation experiments or actual test data to ensure that it can accurately predict the operating status and carbon emissions of the power station. The application of the DDPG algorithm enables the present invention to maintain high-efficiency decision-making capabilities in the face of complex power station operating environments. This algorithm can not only handle high-dimensional state spaces but also effectively address the challenges of continuous action spaces. When implementing optimization strategies in the prior art, the uncertainties and complexities in specific operations are rarely considered, which may lead to poor optimization effects or even counterproductive results. By introducing the DDPG algorithm, the present invention fully considers the uncertainties and complexities in operations, effectively avoids these problems, and improves the optimization effect.

[0077] S3. Re-input the operating data of the sensor into the deep reinforcement learning model to obtain the current and future carbon emission assessment results.

[0078] Specifically, extract the pressure reading P, temperature reading Temp, and water flow rate Flow from the operating data of the sensor and combine them into a state vector S current = [P, Temp, Flow].

[0079] It should be noted that pressure readings, temperature readings, and water flow rates are selected from the preprocessed formatted data as the core components of the state vector. These parameters are key factors affecting the operating efficiency and carbon emissions of the power station. The selected data points are standardized to ensure that data of different magnitudes can be uniformly input into the model. Standardization can be achieved by subtracting the mean and dividing by the standard deviation to generate the state vector: the standardized pressure readings, temperature readings, and water flow rates are combined into a unified state vector. This state vector will be used as the input to the deep reinforcement learning model. By selecting pressure readings, temperature readings, and water flow rates as the core parts of the state vector, the present invention directly models the most critical factors in the operation of the power station. This selection not only simplifies the complexity of the model but also improves the effectiveness of the model. Existing technologies often ignore the impact of external environmental factors on the operation of the power station, resulting in incomplete optimization solutions. By introducing these key parameters, the present invention realizes real-time monitoring and dynamic response to the external environment, thereby formulating a more accurate optimization solution.

[0080] Specifically, the state vector S is used current Through the Actor network, the optimal action Act is generated and substituted into the above formula to obtain Q(S current , Act).

[0081] It should be noted that the generated state vector is input into the Actor network, which is responsible for generating the optimal action suggestions. According to the input state vector, the Actor network outputs a continuous action value. This action value represents the best operation to be taken in the current state (such as adjusting pressure, temperature, or water flow rate). The generated action value is substituted into the previously defined formula to recalculate the value of the state-action pair:

[0082] Q(S current , Act) = w1·P + w2·Temp + w3·Flow + w4·Act - w5·e -λ·τ ,

[0083] Among them, w1 is the weight coefficient for adjusting the pressure P, w2 is the weight coefficient for adjusting the temperature Temp, w3 is the weight coefficient for adjusting the water flow rate Flow, w4 is the weight coefficient for adjusting the action Act, w5 is the influence weight coefficient for adjusting the time decay term, λ is the weight coefficient for controlling future rewards, τ represents the current moment, and the optimal action is generated using the Actor network. This method overcomes the problems of slow response speed and low prediction accuracy of traditional static analysis methods. By dynamically calculating the value of the state-action pair, the forward-looking and adaptive capabilities of the system are significantly improved. Traditional optimization methods usually rely on historical data analysis and lack forward-looking. The present invention significantly improves the forward-looking and adaptive capabilities of the system by dynamically calculating the value of the state-action pair, making the optimization strategy more flexible and efficient.

[0084] Specifically, by substituting Q(S current , Act) into the random forest algorithm, the current and future carbon emission assessment results are obtained.

[0085] It should be noted that the calculated value of the state-action pair Q(S current , Act) is used as one of the input features and input into the pre-trained random forest model. The random forest model outputs the current and future carbon emission assessment results based on the input features. These results can help decision-makers understand the impact of current operations on carbon emissions and formulate optimization strategies accordingly. By combining the deep reinforcement learning model and the random forest algorithm, the present invention can not only dynamically adjust the power station operation state but also accurately predict the carbon emission trend. This method fully utilizes the advantages of the two models and provides a more comprehensive optimization solution. Existing technologies rarely consider the uncertainties and complexities in specific operations when implementing optimization strategies, which may lead to poor optimization effects or even counter-effects. The present invention effectively avoids these problems and improves the optimization effect by introducing the random forest algorithm. In addition, the random forest algorithm has advantages in dealing with high-dimensional data and non-linear relationships and can provide more accurate carbon emission predictions, thus supporting the formulation of more effective optimization strategies.

[0086] S4. Formulate specific optimization strategies based on the carbon emission assessment results, obtain an optimization plan for reducing carbon emissions, and implement it.

[0087] Specifically, compare the carbon emission values in the time period through the obtained current and future carbon emission assessment results and take the maximum value.

[0088] It should be noted that the current and future carbon emission assessment results are obtained from the random forest algorithm. These results are usually presented in the form of a time series, dividing the entire assessment period into multiple fixed time periods (such as every hour, day, or week) for piecewise analysis. Within each time period, the total carbon emissions are calculated. This can be achieved by accumulating all the carbon emission data points within that time period. By comparing the carbon emission values in each time period, the maximum value and its corresponding time period are found. Through a detailed analysis of the carbon emissions in each time period, high-carbon emission periods can be accurately located, providing a clear target for subsequent optimization strategies. Existing technologies often rely on historical data analysis and lack detailed analysis of specific time periods. The present invention can more accurately identify high-carbon emission periods by comparing the carbon emissions in different time periods, thereby formulating more targeted optimization strategies.

[0089] Specifically, based on the time period corresponding to the maximum carbon emission value, identify the operations or conditions of high carbon emissions, and analyze and extract the pressure reading P, temperature reading Temp, and water flow rate Flow.

[0090] It should be noted that according to the time period of the maximum carbon emission value determined in the previous step, further analyze the operation data within this time period, and extract key parameters such as pressure reading, temperature reading, and water flow rate from the data of this time period. These parameters are the main factors affecting the power plant operation efficiency and carbon emissions. Combining the operation logs and other relevant information of the power plant, analyze the specific operations or conditions that lead to high carbon emissions, such as excessive equipment load, cooling system failure, etc. By detailed analysis of the specific operations and conditions during the high-carbon emission period, the root causes of high carbon emissions can be accurately identified. This fine-grained analysis helps to formulate more targeted optimization measures. Existing technologies usually rely on global data analysis and are difficult to accurately identify the causes of local high carbon emissions. The present invention can more accurately discover the root causes of problems by focusing on the specific operations and conditions during the high-carbon emission period, thereby formulating more effective optimization strategies.

[0091] Specifically, modify the operation values corresponding to the obtained pressure reading P, temperature reading Temp, and water flow rate Flow to obtain an optimized solution for reducing carbon emissions.

[0092] It should be noted that by clarifying the carbon emission reduction target to be achieved, such as reducing carbon emissions by 10%, specific adjustment suggestions are put forward based on the extracted key parameters (pressure readings, temperature readings, and water flow rates). For example, appropriately reducing the pressure, increasing the cooling water flow rate, etc. Use simulation tools or models to simulate and verify the proposed adjustment plan to ensure that it can effectively reduce carbon emissions and maintain the normal operation of the power station. According to the simulation results, a detailed optimization plan is generated, including specific adjustment parameters and implementation steps. By proposing specific optimization plans through adjustments based on key parameters, this method is not only targeted but also ensures the effectiveness and feasibility of the optimization measures. Existing technologies often lack specific optimization measures, resulting in poor actual application effects. Through detailed analysis and simulation verification, the present invention can generate practical and feasible optimization plans, significantly improving the optimization effect.

[0093] Specifically, modify the equipment settings according to the optimization plan for carbon emission reduction and implement it.

[0094] It should be noted that according to the generated optimization plan, a detailed implementation plan is formulated, including specific steps for equipment adjustment, personnel arrangements, and a schedule. Gradually adjust the settings of relevant equipment according to the plan, such as adjusting the pressure valve, adjusting the cooling system, etc. During the implementation process, continuously monitor the operation status of the power station and the carbon emission situation, promptly discover and handle possible problems, evaluate the actual effect of the optimization plan based on the monitoring data, and make further adjustments and improvements as needed. Through the systematic implementation plan and real-time monitoring, ensure that the optimization measures can be successfully implemented and achieve the expected effect. This closed-loop control mechanism can continuously optimize and improve the optimization plan to ensure long-term effects. Existing technologies rarely consider the uncertainties and complexities in specific operations when implementing optimization strategies, which may lead to poor optimization effects or even countereffects. Through real-time monitoring and feedback mechanisms, the present invention can effectively address these uncertainties and ensure the successful implementation of the optimization plan.

[0095] S5. Monitor the implementation effect of the optimization plan and feed its monitoring data back to the deep reinforcement learning model.

[0096] Specifically, obtain the pressure reading P, temperature reading Temp, and water flow rate Flow after the implementation of the optimization plan from the sensors.

[0097] It should be noted that after the implementation of the optimization scheme, key parameters such as pressure readings, temperature readings, and water flow rates are collected in real time through sensors installed in the pumped-storage power station and its surrounding environment. The collected data is transmitted to the central server using the LoRaWAN wireless communication module to ensure the timeliness and accuracy of the data. The raw data transmitted is preliminarily processed, including removing outliers and synchronizing timestamps, to ensure the basic quality of subsequent analysis. By collecting and transmitting the operation data in real time after the implementation of the optimization scheme, the present invention can timely understand the actual operation status of the power station and provide a basis for subsequent feedback and adjustment. The prior art often lacks real-time monitoring of the implementation effect of the optimization scheme, resulting in the inability to timely discover and correct potential problems. The present invention ensures the effectiveness of the optimization scheme through real-time data collection and can make adjustments in a timely manner.

[0098] Specifically, the pressure reading P, temperature reading Temp, and water flow rate Flow after implementation are recombined into S current = [P, Temp, Flow].

[0099] It should be noted that the pressure reading, temperature reading, and water flow rate are selected from the preliminarily processed data as the core components of the state vector, and the selected data points are standardized to ensure that data of different magnitudes can be uniformly input into the model. Standardization can be achieved by subtracting the mean and dividing by the standard deviation. The standardized pressure reading, temperature reading, and water flow rate are combined into a unified state vector for subsequent input into the deep reinforcement learning model. By reconstructing the state vector, the present invention can accurately reflect the actual operation status after the implementation of the optimization scheme and provide accurate data support for subsequent feedback and adjustment. The prior art often ignores the impact of external environmental factors on the operation of the power station, resulting in an incomplete optimization scheme. The present invention realizes real-time monitoring and dynamic response to the external environment by introducing these key parameters, thereby formulating a more accurate optimization scheme.

[0100] Specifically, the recombined S current = [P, Temp, Flow] value passes through the Actor network to generate the optimal action Act, and the new optimal action Act and S current = [P, Temp, Flow] are input into the deep reinforcement learning model to calculate the value Q(S current , Act).

[0101] It should be noted that the generated state vector is input into the Actor network, which is responsible for generating optimal action suggestions. According to the input state vector, the Actor network outputs a continuous action value. This action value represents the best operation to be taken in the current state (such as adjusting pressure, temperature, or water flow rate). Substitute the generated action value into the previously defined formula to recalculate the value Q(S current , Act). By using the Actor network to generate optimal actions and combining with the calculation of the value of the state-action pair, the present invention can dynamically adjust the operation state of the power station and enhance the adaptive ability of the system. Traditional methods usually optimize based on fixed rules and lack flexibility. By dynamically calculating the value of the state-action pair, the present invention significantly improves the foresight and adaptive ability of the system, making the optimization strategy more flexible and efficient.

[0102] Specifically, according to the new data and calculation results, use the gradient descent optimization algorithm to update the weights of the Actor and Critic networks.

[0103] It should be noted that according to the new data and calculation results, calculate the error between the predicted values and the actual values of the Actor and Critic networks, and the weights to minimize the prediction error. Repeat the above process to continuously update the network weights until the predetermined accuracy requirement or the number of training times is reached. By using the gradient descent optimization algorithm to update the network weights, the present invention can continuously improve the prediction ability and stability of the model, ensuring its high efficiency in complex environments. Existing technologies often have difficulty maintaining the stability and accuracy of the model when facing complex and changeable operating environments. The present invention effectively addresses this challenge by continuously optimizing the network weights and enhances the robustness of the model.

[0104] Specifically, repeat the above process regularly to continuously optimize the parameters of the deep reinforcement learning model and the optimization strategy to ensure that the system is always in the best operating state.

[0105] It should be noted that determine the time period for regularly repeating the above process, such as evaluating and adjusting once per hour, per day, or per week. Through an automated script or program, regularly and automatically execute steps such as data collection, state vector generation, optimal action calculation, and network weight update. In each period, continuously optimize the parameters of the deep reinforcement learning model and the optimization strategy to ensure that the system is always in the best operating state. By regularly repeating the above process, the present invention can achieve self-optimization and continuous improvement of the system, ensuring its long-term efficient operation. Existing technologies often lack a continuous optimization mechanism, resulting in the difficulty of maintaining the optimization effect in the long term. The present invention ensures the long-term stability and high efficiency of the system through regular repetition and continuous optimization, significantly improving the overall operation efficiency of the power station.

[0106] This embodiment also provides a carbon assessment and optimization system for a pumped - storage power station, including:

[0107] A data collection and pre - processing module, which obtains formatted data by arranging and collecting the operation data of sensors from the pumped - storage power station and its surrounding environment, and pre - processing the operation data.

[0108] A deep reinforcement learning model construction module, which constructs a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm.

[0109] A carbon emission assessment module, which re - inputs the operation data of the sensors into the deep reinforcement learning model to obtain the current and future carbon emission assessment results.

[0110] An optimization strategy formulation and implementation module, which formulates specific optimization strategies according to the carbon emission assessment results, obtains an optimization plan for reducing carbon emissions and implements it.

[0111] A monitoring and feedback module, which monitors the implementation effect of the optimization plan and feeds its monitoring data back to the deep reinforcement learning model.

[0112] This embodiment also provides a computer device, which is applicable to the case of the carbon assessment and optimization method for a pumped - storage power station, including: a memory and a processor; the memory is used to store computer - executable instructions, and the processor is used to execute the computer - executable instructions to implement the carbon assessment and optimization method for a pumped - storage power station as proposed in the above - mentioned embodiment.

[0113] This computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non - volatile storage medium and an internal memory. The non - volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non - volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, NFC (Near - Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad set on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0114] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the carbon assessment and optimization method for pumped storage power stations as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM for short), electrically erasable programmable read-only memory (EEPROM for short), erasable programmable read-only memory (EPROM for short), programmable read-only memory (PROM for short), read-only memory (ROM for short), magnetic memory, flash memory, magnetic disk or optical disc.

[0115] In summary, through the combination of deep reinforcement learning and sensor network, the present invention realizes the real-time monitoring and dynamic response to external environmental factors (such as temperature, humidity, etc.), improving the comprehensiveness and accuracy of the optimization scheme. By adopting dynamic analysis to replace traditional static means, the prediction accuracy and response speed are significantly improved, enhancing the forward-looking and adaptability of the system. In particular, the present invention fully considers the uncertainties and complexities in operation, effectively avoiding the problem of poor optimization effect. In summary, the present invention not only greatly improves the operation efficiency of the power station, but also effectively reduces carbon emissions, supports the efficient utilization of clean energy, and has significant technological progress significance and broad application prospects.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A carbon assessment and optimization method for a pumped storage power station, characterized in that, including: collecting the operation data of sensors arranged in the pumped - storage power station and its surrounding environment, and pre - processing the operation data to obtain formatted data; constructing a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm; re - inputting the operation data of the sensors into the deep reinforcement learning model to obtain the current and future carbon emission assessment results; formulating specific optimization strategies according to the carbon emission assessment results, obtaining an optimization plan for reducing carbon emissions and implementing it; monitoring the implementation effect of the optimization plan and feeding back its monitoring data to the deep reinforcement learning model.

2. The carbon assessment and optimization method for a pumped storage power station according to claim 1, wherein: The steps of collecting the operation data of sensors arranged in the pumped - storage power station and its surrounding environment, and pre - processing the operation data to obtain formatted data are as follows: Installing multiple sensors in the pumped - storage power station, each sensor is equipped with a LoRaWAN wireless communication module and transmits operation data to the central server; Using a filtering algorithm to remove abnormal data points from the operation data, and synchronizing the timestamps of all operation data using GPS clock; Converting the operation data with timestamps into a unified JSON format to obtain operation data in a unified format; Performing deep cleaning on the operation data in the unified format, filling in missing values and correcting biases to obtain formatted data.

3. The carbon assessment and optimization method for a pumped storage power station according to claim 2, characterized in that: The steps of constructing a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm are as follows: Extract the pressure reading P and the temperature reading Temp from the formatted data as components of the state vector S current ; The Actor-Critic structure is used to calculate the value Q(S current , Act), and the formula is derived as follows: Q(S current , Act) = w1·P + w2·Temp + w3·Flow + w4·Act - w5·e -λ·τ (1) where: w1 is the weight coefficient for adjusting pressure P, w2 is the weight coefficient for adjusting temperature Temp, w3 is the weight coefficient for adjusting water flow rate Flow, w4 is the weight coefficient for adjusting action Act, w5 is the weight coefficient for adjusting the influence of the time decay term, λ is the weight coefficient for controlling future rewards, and τ refers to the current moment; After calculating the value Q(S current , Act) of the state-action pair, initialize the weights of the Actor network and the Critic network, and construct a deep reinforcement learning model in conjunction with the Deep Deterministic Policy Gradient algorithm.

4. The carbon assessment and optimization method for a pumped storage power station according to claim 3, characterized in that: The steps of re - inputting the operation data of the sensors into the deep reinforcement learning model to obtain the current and future carbon emission assessment results are as follows: Extract the pressure reading P, the temperature reading Temp, and the water flow rate Flow from the operating data of the sensor and combine them into a state vector S current = [P, Temp, Flow]; Use the state vector S current Through the Actor network, generate the optimal action Act and substitute it into formula (1) to obtain Q(S current , Act); By substituting the newly obtained Q(S current , Act) into the random forest algorithm, the current and future carbon emission assessment results are obtained.

5. A carbon assessment and optimization method for a pumped storage power station according to claim 4, characterized in that: The steps of formulating specific optimization strategies according to the carbon emission assessment results, obtaining an optimization plan for reducing carbon emissions and implementing it are as follows: Comparing the carbon emission values in the time period through the obtained current and future carbon emission assessment results and taking the maximum value; Identifying high - carbon - emission operations or conditions according to the time period with the maximum carbon emission value, and analyzing and extracting the pressure reading P, temperature reading Temp and water flow rate Flow; Correspondingly modifying the operation values through the obtained pressure reading P, temperature reading Temp and water flow rate Flow to obtain an optimization plan for reducing carbon emissions; Modifying the equipment settings according to the optimization plan for reducing carbon emissions and implementing it.

6. The carbon assessment and optimization method for a pumped storage power station according to claim 5, characterized in that: The steps of monitoring the implementation effect of the optimization plan and feeding back its monitoring data to the deep reinforcement learning model are as follows: Obtaining the pressure reading P, temperature reading Temp and water flow rate Flow after the implementation of the optimization plan from the sensors; And recompose the pressure reading P, temperature reading Temp, and water flow rate Flow after implementation into S current = [P, Temp, Flow]; The recombined S current =[P,Temp,Flow] values pass through the Actor network to generate a new optimal action Act, and the new optimal action Act and S current =[P,Temp,Flow] are input into the deep reinforcement learning model to calculate the value Q(S current , Act); According to the new data and the calculated value Q(S current , Act) of the state-action pair, the weights of the Actor and Critic networks are updated using the gradient descent method to optimize the algorithm; Regularly repeating the above process to continuously optimize the parameters of the deep reinforcement learning model and the optimization strategy to ensure that the system is always in the best operating state.

7. A carbon assessment and optimization system for a pumped-storage power station, based on the method for carbon assessment and optimization of a pumped-storage power station according to any one of claims 1 to 6, characterized in that, including: Data collection and preprocessing module, which obtains formatted data by arranging and collecting the operation data of sensors from a pumped-storage power station and its surrounding environment, and preprocessing the operation data; Deep reinforcement learning model construction module, which constructs a deep reinforcement learning model based on the formatted data in cooperation with the deep deterministic policy gradient algorithm; Carbon emission assessment module, which re-enters the operation data of the sensors into the deep reinforcement learning model to obtain the current and future carbon emission assessment results; Optimization strategy formulation and implementation module, which formulates specific optimization strategies according to the carbon emission assessment results to obtain an optimization plan for reducing carbon emissions and implements it; Monitoring and feedback module, which monitors the implementation effect of the optimization plan and feeds its monitoring data back to the deep reinforcement learning model.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the carbon assessment and optimization method for a pumped-storage power station according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the carbon assessment and optimization method for a pumped-storage power station according to any one of claims 1 to 6.

Images