Cooling tower fan energy-saving optimization method and system based on reinforcement learning

Abstract

The invention belongs to the technical field of cooling towers, and particularly relates to a cooling tower fan energy-saving optimization method and system based on reinforcement learning. The method comprises the following steps: acquiring historical operation data of a cooling tower fan, and processing the historical operation data to divide a time range covered by the historical operation data into at least one operation stage; for each operation stage, obtaining the target power of the operation stage; and when it is monitored that the current operation power of the cooling tower fan deviates from the target power, a structured compensation signal is generated according to the current operation deviation value, and the operation power of the cooling tower fan is adjusted based on the structured compensation signal until the operation power is stabilized in a preset target power interval. According to the method, the updated regulation and control reference path is generated based on the regulation and control result, the regulation and control stability and reliability are enhanced, and energy conservation is achieved on the premise that the cooling effect is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of cooling tower technology, specifically relating to a method and system for optimizing energy saving of cooling tower fans based on reinforcement learning. Background Technology

[0002] As a key power device in industrial cooling systems, cooling tower fans are essential for facilitating heat exchange between circulating cooling water and ambient air through forced air convection, ensuring stable temperatures in core production processes. With the increasing intelligence of industry, there is a corresponding need to improve the operating efficiency and intelligence level of cooling tower fans.

[0003] In existing energy-saving control technologies for cooling tower fans, the control strategies lack the ability to adapt to dynamic operating conditions. The control methods mostly rely on preset operating curves or fixed temperature and pressure thresholds for passive adjustment. They cannot respond in real time to complex system changes caused by a combination of factors such as ambient temperature and humidity, atmospheric pressure, and production load fluctuations. This results in energy redundancy and waste during low-load periods, and sluggish response during sudden load increases, failing to guarantee the best cooling effect.

[0004] To address the aforementioned problems, this invention provides a method and system for optimizing energy conservation in cooling tower fans based on reinforcement learning. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for energy-saving optimization of cooling tower fans based on reinforcement learning, which can set an independent power adjustment mechanism for each operating stage to ensure the adjustment accuracy of the fan's operating status.

[0006] The specific technical solution adopted by this invention is as follows: A method for energy-saving optimization of cooling tower fans based on reinforcement learning includes the following steps: When the current operating power of the cooling tower fan deviates from the preset target power for the current operating stage, execute: Based on the current operating deviation between the current operating power and the target power, a structured compensation signal is generated; Based on the structured compensation signal, the operating power of the cooling tower fan is adjusted until the operating power stabilizes within the preset target power range; Before adjusting the operating power of the cooling tower fan, the following steps are performed: acquiring historical operating data of the cooling tower fan, including operating power data and heat exchange data corresponding to the operating power data; and processing the historical operating data to divide the time range covered by the historical operating data into at least one operating stage.

[0007] Preferably, processing historical operational data to divide the time range covered by the historical operational data into at least one operational stage includes: clustering the historical operational data to determine inflection points; and dividing the data into at least one operational stage based on the inflection points.

[0008] Preferably, clustering is performed on historical operational data to determine inflection points, including: Establish a functional relationship with heat exchange data as the independent variable and operating power data as the dependent variable; calculate the second derivative of the functional relationship; and determine the inflection point as the point where the absolute value of the second derivative is greater than a preset threshold.

[0009] Preferably, based on the inflection point, dividing the operation into at least one phase includes: Establish a functional relationship with heat exchange data as the independent variable and operating power data as the dependent variable; calculate the first derivative of the functional relationship; extract the extreme points where the first derivative is zero; and determine adjacent extreme points as the start and end points of the operating phase, respectively.

[0010] Preferably, the method further includes: For each operating phase, the average operating power in the historical operating data corresponding to that operating phase is calculated; and the average operating power is multiplied by a preset amplitude adjustment coefficient to obtain the target power for that operating phase.

[0011] Preferably, the structured compensation signal includes: The first compensation signal is calculated based on reference parameters preset for the current operating phase; The second compensation signal is calculated based on real-time feedback data from the operation of the cooling tower fan.

[0012] Preferably, adjusting the operating power of the cooling tower fan based on the structured compensation signal includes: The power regulation range of the current operation phase is divided into multiple sub-ranges; within each sub-range, an updated regulation reference path is generated based on the regulation results of the cooling tower fan. The structured compensation signal is adjusted based on the updated regulatory reference path.

[0013] This invention also discloses a cooling tower fan energy-saving optimization system based on reinforcement learning, comprising the following modules: The operation phase segmentation module is used to acquire historical operation data of the cooling tower fan, divide the time range covered by the historical operation data into at least one operation phase, and determine the target power for each operation phase. The operating power monitoring module is used to monitor the current operating power of the cooling tower fan and determine the current operating deviation value between the current operating power and the target power determined by the operating stage division module. A compensation signal generation module, configured in response to the current operating deviation value determined by the operating power monitoring module, is used to generate a structured compensation signal; The operating power control module is used to adjust the operating power of the cooling tower fan based on the structured compensation signal generated by the compensation signal generation module. And a control path update module, which monitors the control results performed by the operating power control module to generate an updated control reference path and feeds the updated control reference path back to the compensation signal generation module for adjusting the structured compensation signal.

[0014] Preferably, the operation phase division module is configured to: perform clustering processing on historical operation data to determine inflection points; and divide the operation into at least one operation phase based on the inflection points; The operation phase division module is also configured to: calculate the average operating power in the historical operating data corresponding to each operation phase; and multiply the average operating power by a preset amplitude adjustment coefficient to obtain the target power for that operation phase.

[0015] Preferably, the structured compensation signal includes: The first compensation signal is calculated based on reference parameters preset for the current operating phase; And a second compensation signal calculated based on real-time feedback data from the operation of the cooling tower fan.

[0016] Beneficial effects: 1. This invention obtains historical operating data of cooling tower fans, analyzes the functional relationship between heat exchange data and operating power data to determine inflection points, and divides the time range covered by historical operating data into at least one operating stage based on the inflection points. This allows for the setting of differentiated control strategies for operating stages with different load characteristics, overcoming the shortcomings of traditional single models that are difficult to adapt to changing operating conditions. This enables power control to closely match the actual needs of each operating stage, thereby improving the accuracy and adaptability of energy-saving optimization.

[0017] 2. This invention compares the current operating power with the target power and generates a structured compensation signal based on the current operating deviation value. This compensation signal adjusts the operating power of the cooling tower fan by setting a negative offset value in the sub-interval and updating the control reference path, thereby achieving rapid response and precise fine-tuning of the operating power. This keeps the power continuously stable within the target power range that meets the heat exchange requirements, reducing energy consumption while ensuring the cooling effect and achieving energy saving. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0020] Example 1: See Figure 1 This embodiment provides a method for optimizing energy saving of cooling tower fans based on reinforcement learning. The specific steps are as follows: S1. Execution of data acquisition and phase division; Initially, historical operating data of the cooling tower fan is acquired. The historical operating data includes at least the operating power data collected in multiple historical operating cycles and the heat exchange data corresponding to the operating power data in time.

[0021] The acquired historical operating data is processed and organized into time series records according to the time dimension to establish a time series curve that characterizes the correspondence between heat exchange data and operating power data. This time series curve is the energy efficiency characteristic curve.

[0022] Furthermore, critical state transition points are identified based on the changing trends of the energy efficiency characteristic curves. Mathematically, critical state transition points are inflection points, which are used to divide the operation into phases, achieving precise segmentation of the operation process. Specifically: Calculate the second derivative of the energy efficiency characteristic curve, and identify the points where the absolute value of the second derivative is greater than a preset threshold as critical state transition points. These critical state transition points indicate that the system operating state has undergone a significant nonlinear change. After identifying the critical state transition points, the first derivative of the energy efficiency characteristic curve is calculated, and the points where the first derivative is zero are identified as local extreme points of the energy efficiency curve to accurately define the boundaries of each operating stage. Adjacent local extreme points are determined as the starting and ending points of the operation phase, respectively.

[0023] The above method divides the complete operating cycle into several continuous operating stages with unique operating trends and power characteristics, laying the foundation for subsequent targeted power regulation.

[0024] The operating phase refers to a continuous operating interval defined by adjacent local extreme points on the energy efficiency characteristic curve within a complete operating cycle.

[0025] S2. Determine the initial operating power; After the cooling tower fan has undergone a complete shutdown and received a reset signal, the cooling tower fan is restarted. After the cooling tower fan restarts and enters a stable operating state, its operating power is collected within a preset time period, and the stable operating power value within this time period is determined as the initial operating power of the current optimization cycle.

[0026] Furthermore, the initial operating power is used as a baseline reference for subsequent power regulation strategies.

[0027] S3. Set the power adjustment range and target power. Based on each operating stage, set an independent power adjustment range for it. The power adjustment range is the power control range set for a specific operating phase, and its upper and lower limits are determined by the lowest and highest operating power values ​​in the historical data of that phase, respectively.

[0028] Retrieve the historical operating data corresponding to this operating phase, set the lowest recorded operating power value as the starting point of the power adjustment range, and set the highest operating power value as the ending point of the power adjustment range. This power adjustment range is used to constrain the range of variation of subsequent power adjustment operations.

[0029] Meanwhile, in order to set reasonable energy-saving targets for each operating stage, the average operating power in the historical operating data corresponding to that operating stage is calculated; the average operating power is multiplied by the preset amplitude adjustment coefficient, and the result is determined as the target power for that operating stage, which is the ideal operating power value set for a specific operating stage.

[0030] S4. Construct the power regulation sequence and determine the final power node; Within the power regulation range of each operating phase, based on the preset power increase change value, starting from the initial power point of the power regulation range, a series of ordered discrete power regulation nodes are constructed in an incremental manner, thereby forming the power regulation sequence to be evaluated.

[0031] Among them, the power increase change value refers to the preset fixed power increment, which is used as the basic step size for constructing discrete power adjustment nodes within the power adjustment range.

[0032] Furthermore, by identifying the first peak power point in the power regulation sequence, a reasonable endpoint power node is determined within the power regulation sequence to prevent unlimited power increases during load expansion. The specific identification and determination process is as follows: Identify all local peak power points in the power adjustment sequence, and calculate the power difference between each local peak power point and the starting power point of the power adjustment sequence in ascending order of power value; determine the first local peak power point that makes the power difference greater than the preset adjustment amplitude threshold as the first peak power point; Once the first peak power point is determined, it is set as the end power node of the current operating phase, and subsequent load increases will not cause the operating power to exceed the power value corresponding to this node.

[0033] The adjustment amplitude threshold is a preset power difference value, which is used as a criterion for determining whether a local peak power point in the power adjustment sequence is the first peak power point.

[0034] S5. Implement dynamic power regulation; During the operation of the cooling tower fan, its current operating power is dynamically sampled and continuously compared with the target power of the current operating stage. Based on the comparison results, the current operating deviation value between the current operating power and the target power is calculated. Based on the current operating deviation value, a structured compensation signal is generated to adjust the operating power of the cooling tower fan.

[0035] Furthermore, if the current operating power is greater than the target power, the cooling tower fan is determined to enter the unloading state; if the current operating power is less than the target power, it is determined to enter the overload state.

[0036] The structured compensation signal consists of two logically independent parts: The first compensation signal part is a fixed adjustment amount calculated based on preset reference parameters, which provides basic, open-loop power regulation; The second compensation signal is a dynamic adjustment quantity that is dynamically calculated based on the real-time feedback data of the cooling tower fan operation, which realizes a precise closed-loop feedback control method.

[0037] Furthermore, the power adjustment process employs a segmented adjustment strategy, specifically: The power adjustment range of the current operating phase is divided into multiple sub-ranges, and negative offset values ​​are set for each sub-range. The negative offset value refers to the adjustment parameter set for each sub-range in the segmented adjustment strategy, which decreases monotonically in multiple sub-ranges as the power approaches the target power. Within each sub-interval, the control results are input into a preset threshold judgment rule set for evaluation. The evaluation results are used to generate or select an updated control reference path.

[0038] The updated control reference path is used to optimize the generation logic of the second compensation signal, making the feedback control more adaptive. The operating power of the cooling tower fan is continuously adjusted through this process until it stabilizes within a preset target power range.

[0039] Throughout the entire control process, all operations are conducted with the primary premise of ensuring that the equipment operates within the preset temperature control range. If any critical temperature parameter exceeds the safe range, the preset safety protection logic will be executed immediately, i.e., the power adjustment operation will be suspended or terminated. This allows for dynamic adjustment of the fan power output based on the energy efficiency characteristics of different operating stages, thereby achieving an overall reduction in the operating energy consumption of the cooling tower system.

[0040] The target power range is a preset allowable fluctuation range around the target power. When the fan's operating power stabilizes within this range, the control is considered complete. In practical applications, this embodiment can be integrated with a SCADA system or industrial automation control platform, and communicate with the cooling tower fan in real time through an edge computing gateway to achieve efficient operation control.

[0041] Furthermore, the temperature control zone refers to the permissible operating range of the equipment as defined by the safety threshold of key temperature parameters such as coolant temperature.

[0042] Example 2: See Figure 2 This embodiment provides a cooling tower fan energy-saving optimization system based on reinforcement learning, including: The operation phase segmentation module is used to perform offline analysis and modeling of the operating characteristics of cooling tower fans based on historical operating data, providing benchmarks and targets for online real-time control.

[0043] Furthermore, the operation phase segmentation module is configured to acquire historical operating data of the cooling tower fan, which includes at least time-series recorded operating power data and heat exchange data corresponding to the operating power data.

[0044] After acquiring historical operational data, the module processes the historical operational data to divide the entire time range covered by the historical operational data into at least one operational phase with different operational characteristics.

[0045] In practical applications, this module establishes a functional relationship with heat exchange data as the independent variable and operating power data as the dependent variable to characterize the correlation between fan power and system load. Clustering is performed on historical operational data to identify key points of change in the data distribution. Specifically: Calculate the second derivative of the functional relationship, and identify the points where the absolute value of the second derivative is greater than a preset threshold as inflection points describing significant changes in operational characteristics; based on one or more inflection points, divide the entire historical data time range into multiple independent operational phases.

[0046] Furthermore, this module can also identify the turning point of the running trend by calculating the first derivative of the functional relationship and extracting the extreme points where the first derivative is zero; and determine the adjacent extreme points as the starting point and ending point of the running stage, thus completing the division of the running stage.

[0047] Furthermore, after completing the operational phase division, the operational phase division module is also configured to determine the target power for each operational phase, specifically: For each defined operating phase, the average operating power in the historical operating data corresponding to that operating phase is calculated. This average operating power is then multiplied by a preset amplitude adjustment coefficient used to balance energy saving effect and cooling demand to obtain the target power for that operating phase.

[0048] Furthermore, this target power will serve as a benchmark for subsequent online regulation.

[0049] The operating power monitoring module is responsible for continuously monitoring the operating status of the cooling tower fan. During system operation, this module monitors the current operating power of the cooling tower fan in real time through the data interface. At the same time, based on the current time or operating conditions, it obtains the target power of the current operating stage from the operating stage segment. This module compares the monitored current operating power with the target power to determine the current operating deviation value between the two. When the module detects that the current operating power deviates from the preset target power range around the target power, it determines that there is a deviation that needs to be adjusted, and immediately transmits the current operating deviation value to the compensation signal generation module to trigger subsequent adjustment actions.

[0050] The compensation signal generation module is configured to respond to the current operating deviation value determined by the operating power monitoring module. Upon receiving the current operating deviation value, it generates a structured compensation signal, which consists of a first compensation signal part and a second compensation signal part.

[0051] The first compensation signal is calculated based on reference parameters preset for the current operating phase. These reference parameters are generated by the operating phase division module during the offline analysis phase and reflect the typical operating characteristics of the phase. This part of the signal provides a stable and forward-looking basic adjustment amount. The second compensation signal is dynamically calculated based on real-time feedback data from the operation of the cooling tower fan. It is used to quickly respond to and compensate for real-time disturbances and unmodeled dynamics. By combining the first and second compensation signal parts, the generated structured compensation signal combines the stability of the model with the flexibility of real-time feedback.

[0052] In addition, the compensation signal generation module also interacts with the control path update module, receives the updated control reference path generated by it, and adjusts the generation strategy of the structured compensation signal based on the updated control reference path, thereby realizing the continuous optimization of the control strategy.

[0053] For example, adjusting the weights or calculation model of the first compensation signal part and the second compensation signal part.

[0054] The power regulation module receives the structured compensation signal generated by the compensation signal generation module and parses the compensation signal into specific control instructions for the cooling tower fan hardware, such as adjusting the output frequency or voltage.

[0055] By executing specific control commands, the operating power of the cooling tower fan is directly adjusted, driving it to converge toward the target power until the operating power stabilizes within the preset target power range.

[0056] The control path update module is used to continuously monitor the control results executed by the operating power control module, that is, the actual response of the wind turbine power after receiving the compensation signal.

[0057] Furthermore, the module logically divides the power regulation range of the current operating phase into multiple sub-ranges. Within each sub-range, it analyzes the causal relationship and efficiency between the regulation action and the actual power change, and generates an updated regulation reference path based on the regulation results of the cooling tower fan. The updated control reference path is fed back to the compensation signal generation module to adjust the subsequent structured compensation signal, thereby forming a closed-loop learning and optimization process.

[0058] The generated updated regulatory reference path can be understood as a better mapping strategy of states and actions to guide the system to make more efficient adjustments when it encounters similar deviations in the future.

Claims

1. A method for optimizing energy saving of cooling tower fans based on reinforcement learning, characterized in that, Includes the following steps: When the current operating power of the cooling tower fan deviates from the preset target power for the current operating stage, execute: Based on the current operating deviation between the current operating power and the target power, a structured compensation signal is generated; Based on the structured compensation signal, the operating power of the cooling tower fan is adjusted until the operating power stabilizes within the preset target power range; Before adjusting the operating power of the cooling tower fan, the following steps are performed: acquiring historical operating data of the cooling tower fan, including operating power data and heat exchange data corresponding to the operating power data; and processing the historical operating data to divide the time range covered by the historical operating data into at least one operating stage.

2. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 1, characterized in that, Processing historical operational data to divide the time range covered by the historical operational data into at least one operational stage includes: clustering the historical operational data to determine inflection points; and dividing the data into at least one operational stage based on the inflection points.

3. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 2, characterized in that, Clustering of historical operational data to identify inflection points includes: Establish a functional relationship with heat exchange data as the independent variable and operating power data as the dependent variable; calculate the second derivative of the functional relationship; and determine the inflection point as the point where the absolute value of the second derivative is greater than a preset threshold.

4. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 1, characterized in that, Based on the inflection point, at least one operational phase is defined, including: Establish a functional relationship with heat exchange data as the independent variable and operating power data as the dependent variable; calculate the first derivative of the functional relationship; extract the extreme points where the first derivative is zero; and determine adjacent extreme points as the start and end points of the operating phase, respectively.

5. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 1, characterized in that, The method further includes: For each operating phase, the average operating power in the historical operating data corresponding to that operating phase is calculated; and the average operating power is multiplied by a preset amplitude adjustment coefficient to obtain the target power for that operating phase.

6. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 1, characterized in that, Structured compensation signals include: The first compensation signal is calculated based on reference parameters preset for the current operating phase; And a second compensation signal calculated based on real-time feedback data from the operation of the cooling tower fan.

7. The energy-saving optimization method for cooling tower fans based on reinforcement learning according to claim 1, characterized in that, Adjusting the operating power of cooling tower fans based on structured compensation signals includes: The power regulation range of the current operation phase is divided into multiple sub-ranges; within each sub-range, an updated regulation reference path is generated based on the regulation results of the cooling tower fan. The structured compensation signal is adjusted based on the updated regulatory reference path.

8. A cooling tower fan energy-saving optimization system based on reinforcement learning, characterized in that, Includes the following modules: The operation phase segmentation module is used to acquire historical operation data of the cooling tower fan, divide the time range covered by the historical operation data into at least one operation phase, and determine the target power for each operation phase. The operating power monitoring module is used to monitor the current operating power of the cooling tower fan and determine the current operating deviation value between the current operating power and the target power determined by the operating stage division module. A compensation signal generation module, configured in response to the current operating deviation value determined by the operating power monitoring module, is used to generate a structured compensation signal; The operating power control module is used to adjust the operating power of the cooling tower fan based on the structured compensation signal generated by the compensation signal generation module. And a control path update module, which monitors the control results performed by the operating power control module to generate an updated control reference path and feeds the updated control reference path back to the compensation signal generation module for adjusting the structured compensation signal.

9. A cooling tower fan energy-saving optimization system based on reinforcement learning according to claim 8, characterized in that, The operation phase segmentation module is configured to: perform clustering processing on historical operation data to determine inflection points; and divide the operation into at least one operation phase based on the inflection points. The operation phase segmentation module is also configured to: calculate the average operating power in the historical operation data corresponding to each operation phase; And the average operating power is multiplied by a preset amplitude adjustment coefficient to obtain the target power for this operating stage.

10. A cooling tower fan energy-saving optimization system based on reinforcement learning according to claim 8, characterized in that, Structured compensation signals include: The first compensation signal is calculated based on reference parameters preset for the current operating phase; And a second compensation signal calculated based on real-time feedback data from the operation of the cooling tower fan.

Images