Automatic control method for hydropower station workshop drainage system based on operation condition self-adaption

Abstract

According to the automatic control method for the drainage system of the hydropower station plant based on self-adaption of the operation working conditions, the operation modes of the drainage pump in the flood season / non-flood season are automatically switched, and various conditions needing to be manually confirmed and valve operation are automatically achieved in the process; the real-time water level, the water level rising rate, the equipment health degree score, the pump operation hours, the weather forecast and the historical record of the water collecting well are obtained, the self-adaptive threshold value is obtained based on a deep reinforcement learning algorithm, the drainage pump is started and stopped to respond to rainfall or equipment state changes in real time, and safe drainage, energy consumption cost and equipment abrasion are balanced.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower station plant drainage, and in particular to an automatic control method for a hydropower station plant drainage system based on self-adaptation to operating conditions. Background Art

[0002] Large hydropower plants are usually located near rivers or underground, and are vulnerable to river water infiltration, groundwater intrusion, and rainwater. The leakage drainage system can promptly drain water from the plant and dam floors, the ditches of the corridors on each floor, and the water seepage from various parts of the plant, as well as drain water from the unit top cover and main shaft seal, to prevent the plant from being flooded. The maintenance drainage system can drain water from the penstock, volute, and tailwater pipe during unit maintenance, as well as leakage from the tailwater pipe disc valve during normal operation. The drainage system of large hydropower plants is of vital importance, and flooding of the plant is the primary content of the five key risk controls in hydropower stations. Therefore, it is of great significance to develop an automatic control method for the hydropower plant drainage system based on adaptive operating conditions to ensure the safety of equipment and personnel in the plant.

[0003] At present, the drainage systems of large hydropower plants are all controlled by PLC programs using fixed thresholds. When the water level in the collection well rises to the "working pump start water level", the working pump is automatically started; when it continues to rise to the "standby pump start water level", the standby pump is automatically started; when the water level drops to the "pump stop water level", the working pump is automatically stopped.

[0004] During the flood season / non-flood season, the operating personnel adjust the drainage pump control mode by opening or closing the drainage pump inlet and outlet valves and adjusting the control mode switch handle.

[0005] Disadvantages of existing technology: (1) The drainage system of a large hydropower plant is controlled by a PLC program using a fixed threshold. The fixed threshold is usually determined based on experience or preliminary debugging, and it is difficult to accurately match the complex production process and diverse control requirements.

[0006] (2) The drainage methods during the flood season and non-flood season are quite different. The hydropower plant operators need to adjust the drainage pump control mode by opening or closing the drainage pump inlet and outlet valves, adjusting the control mode switch handle, etc., which increases the operator's operation workload and increases the error rate, posing a safety risk.

[0007] (3) When the unit is under maintenance, it is necessary to open the unit's volute disc valve and tailwater disc valve to drain the accumulated water in the penstock, volute, and tailwater pipe to the maintenance sump, and then use the deep well pump to drain the water to the downstream tailwater. If the disc valve opening and pump start-up are not balanced, it is not conducive to safe and economical operation.

[0008] (4) The drainage system operates independently from the factory fire protection and dehumidification systems and cannot be coordinated during emergency response. Summary of the Invention

[0009] The purpose of the present invention is to provide an automatic control method for the drainage system of a hydropower station plant based on self-adaptation of operating conditions. This method automatically switches the operation mode of the drainage pump during the flood season / non-flood season, and various conditions and valve operations that require manual confirmation are automatically implemented in the process. In addition, by obtaining the real-time water level of the collection well, the water level rise rate, the equipment health score, the pump operation hours, the weather forecast, and the historical records, based on the deep reinforcement learning algorithm, an adaptive threshold is obtained, so that the drainage pump can start and stop in real time in response to rainfall or equipment status changes, balancing safe drainage, energy consumption costs and equipment wear.

[0010] In order to achieve the above technical features, the purpose of the present invention is to achieve the following: a method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation, comprising the following steps: Step 1: Check whether the automatic control conditions of the drainage system are met. If so, the operation process proceeds to the next step; if not, an alarm is issued and the operation process exits; Step 2: Determine whether the unit is in the maintenance and drainage period. If so, the LCU program automatically opens the inlet and outlet water valves of all maintenance pumps remotely. If not, a command is issued to open the inlet and outlet water valves of some maintenance pumps to keep the water level in the water collection well consistent with the number of pumps in operation. Step 3: Collect data: Real-time collection of multi-source real-time data related to water level changes in water collection wells; Step 4, normalization processing: Normalize the data collected in step 3; Step 5: Dynamically adjust the pumping threshold based on the DQN algorithm; Step 6: Determine whether the fire protection system is activated; Step 7: Determine whether the humidity is greater than or equal to 60%.

[0011] Preferably, the automatic control conditions of the drainage system in step 1 specifically include: there is no total fault in the drainage system; the status of each valve is normal; the control mode of the drainage joint control cabinet is "remote"; and the control mode of each deep well pump is "remote".

[0012] Preferably, in the specific judgment process of the total fault of the drainage system, when one of the following conditions is not met, the total fault light of the drainage system will light up: total fault of X# pump; water level sensor fault; AC220V power supply disappears; DC220V power supply disappears; 1-way DC24V power supply disappears; 2-way DC24V power supply disappears.

[0013] Preferably, in the specific judgment process of the X# pump total fault, when one of the following conditions is not met, the X# pump total fault light will light up: X# pump power supply disappears; X# pump soft start failure; X# pump circuit breaker trips; X# pump starts failure; X# pump lubrication water is interrupted.

[0014] Preferably, the multi-source real-time data collected in step 3 includes: Real-time water level in the water collection well, water level rise rate, equipment health, operating hours of each pump, weather forecast, and historical records; The equipment health includes vibration data and insulation detection status; The weather forecast includes the current rainfall intensity and the rainfall forecast amount; The historical records include a pump start threshold and the current pump start and stop status.

[0015] Preferably, the specific process in step 4 is: Normalize the multi-source real-time data and normalize the sample amplitude to [0,1] as follows: ; Where, is the normalized value, is the eigenvalue, is the minimum value, is the maximum value.

[0016] Preferably, the specific process in step 5 is: During the drainage training process, the quadruple of sampled data is obtained ( , , , ), and then update the network parameters; the current state The action corresponding to the maximum Q value is selected by the estimation network: ; This action is then fed into the target network to calculate the estimated Q-value of the policy update reward: ; Where, is the reward value, is the discount factor, denotes the parameters of the estimated network, represents the parameters of the target network, Execute actions for algorithmic decisions, To estimate the maximum Q-value action of the network output, To use the target network Estimate the maximum Q value of the next state, Current status The action corresponding to the maximum Q value is selected by the estimation network, The estimated Q-value for calculating the policy update reward.

[0017] Preferably, the estimated Q value of the policy update reward is used to obtain the value function iteration process of the DQN algorithm: ; in, is the learning rate; The loss function is for: .

[0018] Preferably, the process of dynamically adjusting the pump start-up based on the DQN algorithm is as follows: 1) First enter Status of the water collection well at all times , output The Q value of all actions in the state; 2) Based on probability Randomly select the drainage pump action with probability Use Current State The maximum Q value action output by the prediction network ; 3) After the drainage pump is started, the water level in the water collection well drops and the state of the water collection well changes accordingly. , the four-tuple obtained by starting the drainage pump and interacting with the environment ( , , , ) as network training data.

[0019] 4) When training the network, ( , ) input into the estimation network and calculate the current Q value; Enter the target network and select the action corresponding to the maximum Q value ; Then the action and the next state Input the target network and calculate the target Q value ; Then calculate the loss function , by minimizing the above loss function, the parameters of the policy network are updated, and this cycle is repeated until the drainage is completed.

[0020] Preferably, the specific process in step 6 is: When the fire protection system is started, a large amount of water will be released instantly. This water needs to be quickly discharged into the leakage collection well through the drainage ditch. If the drainage capacity is insufficient, it may cause: flooding in the factory and submergence of electrical equipment. Based on this, when the fire water is sprayed, the leakage drainage pump is started together.

[0021] Preferably, the specific process in step 7 is: When water accumulates in the cable trench, the humidity will soar, which will cause the insulation of components to deteriorate or short circuit. Based on this, the dehumidification system is jointly started when the humidity is high.

[0022] The present invention has the following beneficial effects: 1. The present invention automatically switches the operation mode of the drainage pump during flood season / non-flood season by adding an internal logic judgment function. Various conditions and valve operations that require manual confirmation are automatically implemented in the process, reducing the number of equipment operation items for operators and the probability of misoperation.

[0023] 2. The present invention obtains adaptive thresholds based on the DQN algorithm, which solves the problems of threshold rigidity and multi-target conflict faced by traditional methods in dynamic environments.

[0024] 3. The present invention autonomously learns the optimal threshold value and responds to rainfall and equipment status changes in real time, while optimizing safety, energy consumption, and equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention will be further described below with reference to the accompanying drawings and examples.

[0026] Figure 1 This is a flow chart of the control of a drainage pump for maintenance in a hydropower station building based on self-adaptation of operating conditions according to the present invention.

[0027] Figure 2 This is a control flow chart of a hydropower plant building leakage drainage pump based on operating condition adaptation according to the present invention.

[0028] Figure 3 This is a diagram of the conditions for automatic control of the drainage system of the present invention.

[0029] Figure 4 This is the general fault judgment condition diagram of the drainage system of the present invention.

[0030] Figure 5 This is the general fault judgment condition diagram of the X# pump of the present invention.

[0031] Figure 6 This is a logic diagram for dynamically adjusting the pump start threshold based on the DQN algorithm of the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0033] Example 1: Step 1: Check whether the automatic control conditions of the drainage system are met, such as Figure 3-Figure 5As shown: There is no general fault in the drainage system; the status of each valve is normal; the control mode of the drainage joint control cabinet is "remote"; the control mode of each deep well pump is "remote". When one of the following conditions is not met, the drainage system general fault light is on: X# pump general fault; water level sensor fault; AC220V power supply disappears; DC220V power supply disappears; 1-way DC24V power supply disappears; 2-way DC24V power supply disappears. When one of the following conditions is not met, the X# pump general fault light is on: X# pump power supply disappears; X# pump soft start fault; X# pump circuit breaker trips; X# pump startup failure; X# pump lubrication water interruption. If satisfied, the operation process will proceed to the next step; if not satisfied, an alarm will be issued and the operation process will be exited; Step 2: Determine whether the unit is in the maintenance and drainage period. If so, automatically open the inlet and outlet water valves of all maintenance pumps remotely through the LCU program; if not, issue a command to open the inlet and outlet water valves of some maintenance pumps to keep the water level in the collection well matching the number of pumps started.

[0034] Step 3: Data collection: Real-time acquisition of water level in the water collection well, water level rise rate, equipment health (vibration / insulation monitoring), operating hours of each pump, weather forecast (current rainfall intensity / forecast rainfall), and historical records (most recent pump start threshold / current pump start / stop status). Step 4, normalization processing: Normalize the multi-source real-time data and normalize the sample amplitude to [0,1] as follows: ; Where, is the normalized value, is the eigenvalue, is the minimum value, is the maximum value.

[0035] Step 5: Dynamically adjust the pumping threshold based on the DQN algorithm; During the drainage training process, the quadruple of sampled data is obtained ( , , , ), and then update the network parameters; the current state The action corresponding to the maximum Q value is selected by the estimation network: ; This action is then fed into the target network to calculate the estimated Q-value of the policy update reward: ; Where, is the reward value, is the discount factor, denotes the parameters of the estimated network, represents the parameters of the target network, Execute actions for algorithmic decisions, To estimate the maximum Q-value action of the network output, To use the target network Estimate the maximum Q value of the next state, Current status The action corresponding to the maximum Q value is selected by the estimation network, The estimated Q-value for calculating the policy update reward.

[0036] Preferably, the estimated Q value of the policy update reward is used to obtain the value function iteration process of the DQN algorithm: ; in, is the learning rate; The loss function is for: .

[0037] Preferably, the process of dynamically adjusting the pump start-up based on the DQN algorithm is as follows: 1) First enter Status of the water collection well at all times , output The Q value of all actions in the state; 2) Based on probability Randomly select the drainage pump action with probability Use Current State The maximum Q value action output by the prediction network ; 3) After the drainage pump is started, the water level in the water collection well drops and the state of the water collection well changes accordingly. , the four-tuple obtained by starting the drainage pump and interacting with the environment ( , , , ) as network training data.

[0038] 4) When training the network, ( , ) input into the estimation network and calculate the current Q value; Enter the target network and select the action corresponding to the maximum Q value ; Then the action and the next state Input the target network and calculate the target Q value ; Then calculate the loss function , by minimizing the above loss function, the parameters of the policy network are updated, and this cycle is repeated until the drainage is completed.

[0039] Step 6: Determine whether the fire protection system is activated; When firefighting systems (such as sprinklers or fire hydrants) are activated, they instantly release a large amount of water. This water must be quickly drained through drains into seepage sump wells. Insufficient drainage capacity can lead to flooding in the factory and submersion of electrical equipment (such as cable corridors and distribution cabinets). Therefore, when using firefighting water sprinklers, a seepage drainage pump should be activated simultaneously.

[0040] Step 7: Determine whether the humidity is greater than or equal to 60%.

[0041] When water accumulates in the cable trench, the humidity will soar, which will cause the insulation of the components to deteriorate or short circuit. Therefore, when the humidity is high, the dehumidification system should be started together.

[0042] In summary, the main innovations of the present invention are: 1. For hydropower plants, the drainage system methods are quite different during the unit maintenance period (non-flood season) and the non-maintenance period (flood season). Hydropower plant operators are required to adjust the drainage pump control method by opening or closing the drainage pump inlet and outlet valves, adjusting the control mode switching handle, etc., to reduce the amount of operation.

[0043] 2. The DQN algorithm addresses the challenges of threshold rigidity and multi-objective conflicts faced by traditional methods in dynamic environments. Its core advantages include the ability to dynamically adjust the pump activation threshold, intelligently determine pump activation timing based on real-time data and environmental predictions, and continuously evolve the system, adapting to long-term changes through online learning and reducing the need for manual maintenance.

[0044] 3. The drainage system can be linked with the fire protection system and dehumidification system according to the action conditions and humidity range requirements specified in the linkage procedure.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automatically controlling a drainage system of a hydropower station building based on adaptive operation conditions, characterized in that: The following steps are involved: Step 1: Check whether the automatic control conditions of the drainage system are met. If so, the operation process proceeds to the next step; If not satisfied, alarm will be given and the operation process will be exited; Step 2: Determine whether the unit is in the maintenance and drainage period. If so, the LCU program automatically opens the inlet and outlet water valves of all maintenance pumps remotely. If not, a command is issued to open the inlet and outlet water valves of some maintenance pumps to keep the water level in the water collection well consistent with the number of pumps in operation. Step 3: Collect data: Real-time collection of multi-source real-time data related to water level changes in water collection wells; Step 4, normalization processing: Normalize the data collected in step 3; Step 5: Dynamically adjust the pumping threshold based on the DQN algorithm; Step 6: Determine whether the fire protection system is activated; Step 7: Determine whether the humidity is greater than or equal to 60%.

2. The automatic control method for the drainage system of a hydropower station building based on operating condition adaptation according to claim 1 is characterized in that: The automatic control conditions of the drainage system in step 1 specifically include: there is no overall fault in the drainage system; all valves are in normal status; the control mode of the drainage joint control cabinet is "remote"; and the control mode of each deep well pump is "remote".

3. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 2, characterized in that: During the specific judgment process of the total fault of the drainage system, the total fault light of the drainage system will light up when one of the following conditions is not met: total fault of X# pump; water level sensor fault; AC220V power supply disappears; DC220V power supply disappears; 1-way DC24V power supply disappears; 2-way DC24V power supply disappears.

4. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 3 is characterized in that: During the specific judgment process of the X# pump total fault, the X# pump total fault light will light up when one of the following conditions is not met: X# pump power is lost; X# pump soft start failure; X# pump circuit breaker trips; X# pump starts failure; X# pump lubrication water is interrupted.

5. The automatic control method for the drainage system of a hydropower station building based on operating condition adaptation according to claim 1 is characterized in that: The multi-source real-time data collected in step 3 includes: Real-time water level in the water collection well, water level rise rate, equipment health, operating hours of each pump, weather forecast, and historical records; The equipment health includes vibration data and insulation detection status; The weather forecast includes the current rainfall intensity and the rainfall forecast amount; The historical records include a pump start threshold and the current pump start and stop status.

6. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 1, characterized in that: The specific process in step 4 is: Normalize the multi-source real-time data and normalize the sample amplitude to [0,1] as follows: ； Where, is the normalized value, is the eigenvalue, is the minimum value, is the maximum value.

7. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 1, characterized in that: The specific process in step 5 is: During the drainage training process, the quadruple of sampled data is obtained ( , , , ), and then update the network parameters; Current Status The action corresponding to the maximum Q value is selected by the estimation network: ； This action is then fed into the target network to calculate the estimated Q-value of the policy update reward: ； Where, is the reward value, is the discount factor, denotes the parameters of the estimated network, represents the parameters of the target network, Execute actions for algorithmic decisions, To estimate the maximum Q-value action of the network output, To use the target network Estimate the maximum Q value of the next state, Current status The action corresponding to the maximum Q value is selected by the estimation network, The estimated Q-value for calculating the policy update reward.

8. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 7, characterized in that: The estimated Q value of the updated policy return is used to obtain the value function iteration process of the DQN algorithm: ； in, is the learning rate; The loss function is for: 。 9. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 8, characterized in that: The process of dynamically adjusting the pump start based on the DQN algorithm is as follows: 1) First enter Status of the water collection well at all times , output The Q value of all actions in the state; 2) Based on probability Randomly select the drainage pump action with probability Use Current State The maximum Q value action output by the prediction network ; 3) After the drainage pump is started, the water level in the water collection well drops and the state of the water collection well changes accordingly. , the four-tuple obtained by starting the drainage pump and interacting with the environment ( , , , ) as network training data. 4) When training the network, ( , ) input into the estimation network and calculate the current Q value; Enter the target network and select the action corresponding to the maximum Q value ; Then the action and the next state Input the target network and calculate the target Q value ; Then calculate the loss function , by minimizing the above loss function, the parameters of the policy network are updated, and this cycle is repeated until the drainage is completed.

10. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 9, characterized in that: The specific process in step 6 is: When the fire protection system is started, a large amount of water will be released instantly. This water needs to be quickly discharged into the leakage collection well through the drainage ditch. If the drainage capacity is insufficient, it may cause: flooding in the factory and submergence of electrical equipment. Based on this, when the fire water is sprayed, the leakage drainage pump is started together.

11. The method for automatically controlling a drainage system of a hydropower station building based on operating condition adaptation according to claim 9, characterized in that: The specific process in step 7 is: When water accumulates in the cable trench, the humidity will soar, which will cause the insulation of components to deteriorate or short circuit. Based on this, the dehumidification system is jointly started when the humidity is high.

Images