Machine learning-based intelligent control method and system for water chiller

Abstract

The embodiment of the present application relates to the refrigeration technical field, disclose a kind of based on machine learning's water chilling unit intelligent control method, comprising: by each infrared camera of being arranged in monitoring area to obtain a series of infrared thermal image frames of monitoring area;The weighted center point of grid unit is calculated;Dynamic clustering algorithm is executed to the weighted center point in continuous frame, to identify one or more independent heat source and track its trajectory;For each tracked heat source, construct a multidimensional feature vector;Based on the multidimensional feature vector, calculate the kinematic parameters of the tracked heat source;And based on the calculated kinematic parameters, assign motion intensity level to the tracked heat source;Determine the motion intensity level and kinematic parameters, based on the received motion intensity level and kinematic parameters generate control command for the water chilling unit control system, and the control command is transmitted to corresponding water chilling unit control system to regulate its operation.It can adaptively carry out temperature regulation.

Description

Technical Field

[0001] This invention relates to the field of refrigeration technology, and specifically to a machine learning-based intelligent control method and system for water chillers. Background Technology

[0002] Currently, existing venues employ relatively simple methods for cooling control, typically using fixed settings. These methods are unable to adapt to changing conditions and cannot provide suitable temperatures for occupants. Therefore, designing a solution that can provide appropriate cooling temperatures and adapt to environmental changes has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0003] To address the aforementioned shortcomings, this invention discloses a machine learning-based intelligent control method for chiller units, which enables precise temperature control of venue areas.

[0004] The first aspect of this invention discloses a machine learning-based intelligent control method for chiller units, comprising: A series of infrared thermal image frames of the monitored area are acquired by various infrared cameras installed in the monitoring area. Thermal radiation data from each infrared thermal image frame is mapped onto an N×N grid representing the physical coordinates of the monitored area, and the weighted center point of the grid cell is calculated, with each weighted center point represented by coordinates and a normalized temperature value. A dynamic clustering algorithm is performed on the weighted center points in consecutive frames to identify one or more independent heat sources and track their trajectories; For each tracked heat source, a multidimensional feature vector is constructed, which contains information representing location, monitored temperature, inter-frame displacement, and temperature change rate. A set of kinematic parameters of the tracked heat source is calculated based on the multidimensional feature vector, including at least velocity and thermal diffusivity values; and a motion intensity level is assigned to the tracked heat source based on the calculated kinematic parameters. The motion intensity level and kinematic parameters are determined, and control commands for the chiller unit control system are generated based on the received motion intensity level and kinematic parameters. The control commands are then transmitted to the corresponding chiller unit control system to regulate its operation.

[0005] As an optional implementation, in the first aspect of the present invention, the step of calculating a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector includes: For each identified tracked heat source in the infrared thermal image frame, the temperature value of its corresponding grid is extracted, and the temperature range is divided into K temperature intervals. The frequency distribution vector is obtained by counting the number of radiation points of the tracked heat source within each temperature range. , where the frequency distribution vector It is the ratio of the number of radiation points in the i-th temperature range to the total number of radiation points; The corresponding thermal diffusivity value is calculated based on the frequency distribution vector and the thermal diffusivity value calculation formula, and the thermal diffusivity value is used as the kinematic parameter of the tracked heat source. The thermal diffusivity value calculation formula is as follows: H is the thermal diffusivity value. Let K be the frequency distribution vector, and K be the number of temperature ranges.

[0006] As an optional implementation, in the first aspect of the present invention, after calculating the corresponding thermal diffusivity entropy value according to the frequency distribution vector and the thermal diffusivity entropy value calculation formula, the method further includes: Obtain the spatial weight matrix of the corresponding monitoring area, and assign correction coefficients to the radiation point temperatures of different areas based on the spatial weight matrix; The corrected thermal diffusivity entropy value is determined based on the thermal diffusivity entropy value, the spatial weight matrix, and the entropy correction formula; the entropy correction formula is as follows: ; in, This is the spatially corrected thermal diffusivity value. This is the uncorrected thermal diffusivity value; Let M be the spatial weight of the j-th radiation point, and M be the number of effective radiation points. This is the frequency distribution vector.

[0007] As an optional implementation, in the first aspect of the present invention, after calculating the corresponding thermal diffusivity entropy value according to the frequency distribution vector and the thermal diffusivity entropy value calculation formula, the method further includes: A sliding window is used to perform sliding calculations on multiple infrared thermal image frames to obtain entropy value sequence data. The smoothed entropy value parameters are then determined based on this entropy value sequence data and an exponential smoothing formula, which is: ,in, The entropy value parameter is the smoothed value at the current moment. The entropy value parameter is the smoothed value from the previous time step. The entropy value parameter at the current moment. For smoothing coefficients; The entropy change rate is calculated, and the entropy change rate and the smoothed entropy parameter are matched with the set conditions to determine the current thermal state. The chiller unit control system is then adjusted according to the current thermal state.

[0008] As an optional implementation, in the first aspect of the present invention, the intelligent control method further includes: The current scene type information is determined, and the corresponding temperature range information is determined based on the scene type information. The temperature range information is obtained by training historical thermal radiation data of different scenes using the K-means clustering algorithm.

[0009] As an optional implementation, in the first aspect of the present invention, the dynamic clustering algorithm is the DBSCAN algorithm, and the intelligent control method further includes: If a clustered heat source has a displacement equal to or greater than a first threshold and a temperature equal to or greater than a second threshold within a preset number of frames, it is classified as an active human heat source.

[0010] Calculate the temperature decay rate at the previous location of the active human body after the heat source is removed, and use the temperature decay rate as a corresponding kinematic parameter. The intelligent control method further includes: Monitor multiple heat sources; A high-intensity mass event is detected when a predetermined number of heat sources simultaneously meet the conditions of having a velocity value exceeding a velocity threshold and a thermal diffusion entropy value exceeding an entropy threshold within a specified time window. Control commands for the chiller unit control system are generated based on the detected high-intensity events in the group.

[0011] As an optional implementation, in the first aspect of the present invention, the intelligent control method further includes: Construct a trained LSTM model; The constructed multidimensional feature vector is input into the LSTM model for identification to determine the motion intensity in the next stage; The control commands for the chiller unit control system are generated based on the intensity of the next stage of motion. Alternatively, the intelligent control method may further include: A corresponding heat source map is generated based on a series of infrared thermal image frames of the monitored area obtained by each infrared camera, and the heat source map is superimposed with the air outlet map of the chiller unit to obtain the control map. The process of generating control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmitting the control commands to the corresponding chiller unit control system to regulate its operation, includes: The received motion intensity level, kinematic parameters, and control map are input into the pre-built air supply control model to determine the optimal air supply parameters for each cooling air outlet. The optimal air supply parameters include air volume parameters, air direction parameters, and temperature parameters. Based on the optimal air supply parameters, control commands are generated for the chiller control system, and the control commands are transmitted to the corresponding chiller control system to regulate its operation.

[0012] A second aspect of this invention discloses an intelligent control system for a chiller unit based on machine learning, comprising: Acquisition module: used to acquire a series of infrared thermal image frames of the monitored area through various infrared cameras set up in the monitoring area; Calculation module: used to map thermal radiation data from each infrared thermal image frame onto an N×N grid representing the physical coordinates of the monitored area, and to calculate the weighted center point of the grid cell, each weighted center point being represented by coordinates and a normalized temperature value; Clustering module: Used to perform dynamic clustering algorithms on weighted centroids in consecutive frames to identify one or more independent heat sources and track their trajectories; Feature construction module: used to construct a multi-dimensional feature vector for each tracked heat source, the multi-dimensional feature vector containing location information, monitored temperature information, inter-frame displacement and temperature change rate; Motion calculation module: used to calculate a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector, the set of kinematic parameters including at least velocity value and thermal diffusivity value; and to assign a motion intensity level to the tracked heat source based on the calculated kinematic parameters; Control module: used to determine the motion intensity level and kinematic parameters, generate control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmit the control commands to the corresponding chiller unit control system to regulate its operation.

[0013] A third aspect of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute the machine learning-based intelligent control method for chiller units disclosed in the first aspect of the present invention.

[0014] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the machine learning-based intelligent control method for chiller units disclosed in the first aspect of the present invention.

[0015] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In this embodiment of the invention, the machine learning-based intelligent control method for chiller units classifies motion intensity levels according to kinematic parameters. The chiller unit control system can formulate appropriate control commands based on the motion intensity of the heat source. When the heat source motion intensity is high and heat diffusion is fast, cooling is accelerated; when the heat source distribution is stable and the intensity is low, the output is appropriately reduced, thereby flexibly adapting to real-time load requirements. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the intelligent control method for chiller units based on machine learning disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of the process for calculating kinematic parameters disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a machine learning-based intelligent control system for a chiller unit provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that the terms first, second, third, fourth, etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms used in the embodiments of this invention include and have, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0020] Example 1 Please see Figure 1 , Figure 1This is a flowchart illustrating the intelligent control method for chiller units based on machine learning disclosed in an embodiment of the present invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It may also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on devices located in a certain location. In some scenarios, it can also control multiple storage devices, which may be placed in the same location as the devices or in different locations. Figure 1 As shown, the machine learning-based intelligent control method for chiller units includes the following steps: S101: Acquire a series of infrared thermal image frames of the monitored area by various infrared cameras installed in the monitoring area; S102: Map the thermal radiation data from each infrared thermal image frame onto an N×N grid representing the physical coordinates of the monitored area, and calculate the weighted center point of the grid cell, each weighted center point being represented by coordinates and a normalized temperature value. S103: Perform a dynamic clustering algorithm on the weighted center points in consecutive frames to identify one or more independent heat sources and track their trajectories; S104: For each tracked heat source, construct a multi-dimensional feature vector, which contains information representing location, monitored temperature, inter-frame displacement, and temperature change rate. S105: Calculate a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector, the set of kinematic parameters including at least velocity value and thermal diffusivity value; and assign a motion intensity level to the tracked heat source based on the calculated kinematic parameters; S106: Determine the motion intensity level and kinematic parameters, generate control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmit the control commands to the corresponding chiller unit control system to regulate its operation.

[0021] This invention utilizes an infrared camera to acquire thermal images and maps the thermal radiation data to a physical coordinate grid and calculates a weighted center point, enabling precise location of the physical position and corresponding temperature of heat sources within the monitored area. Simultaneously, a dynamic clustering algorithm identifies multiple independent heat sources and tracks their trajectories, preventing misidentification of heat sources and allowing the chiller unit to respond specifically to the actual heat distribution.

[0022] Specifically, each N×N grid cell corresponds to a physical area within the monitoring region, such as a 1-square-meter square area. The coordinates of the weighted center point are typically two-dimensional planar coordinates, such as x and y values, representing the centroid position of all thermal radiation points within that grid cell. This is the average position calculated based on the temperature weights of each radiation point. Radiation points with higher temperatures contribute more to the center point coordinates. This coordinate precisely corresponds to a specific location in physical space, establishing a one-to-one correspondence between abstract thermal radiation data and the spatial layout of the actual monitoring area. The normalized temperature value of the weighted center point is a quantitative description of the overall thermal intensity within the grid cell. It is obtained by weighting the temperature values ​​of each radiation point within the grid according to their weights (e.g., higher temperatures have higher weights) and then normalizing it to a specific range (e.g., 0~1). This value reflects the average thermal level of the grid area and eliminates the differences in absolute temperature values ​​caused by different infrared cameras or ambient light, making the thermal states of different grids and at different times comparable. The weighted center point is essentially a feature point that combines spatial location and thermal intensity information. It represents the overall thermal distribution state within the corresponding grid cell, providing basic spatial anchoring data for subsequent heat source identification, trajectory tracking, and control decisions.

[0023] In this embodiment of the invention, the raw data of the infrared thermal image is a pixel-level temperature matrix (e.g., 1024×768 pixels), with each pixel corresponding to a temperature value and coordinates. Directly clustering these raw pixels would lead to the following problems: an explosion in data volume and excessively high computational costs; and because the raw pixels are discrete, adjacent pixels may be grouped into different clusters due to slight temperature differences. For example, the pixel temperature at the edge of a human heat source is slightly lower than that of the core area; direct clustering might divide a complete human thermal radiation into multiple small clusters, failing to identify it as "an independent heat source." Infrared images are susceptible to environmental interference (such as device reflections and air convection), containing isolated high-temperature pixels (noise points). Direct clustering would misclassify these noise points as "small heat sources," resulting in redundant clustering results and interfering with subsequent heat source tracking and control decisions.

[0024] The method of first dividing the data into grids and then performing clustering, as described in this embodiment of the invention, achieves the following effects: First, it significantly reduces the amount of data and improves computational efficiency. Assuming the monitored area is divided into a 100×100 grid (10,000 grid units), regardless of the number of pixels in the original image, only 10,000 weighted center points are output. The data volume is compressed from millions to tens of thousands, reducing the computational pressure on the clustering algorithm by more than 100 times, thus meeting real-time requirements. Second, it strengthens spatial correlation. Each grid unit corresponds to a continuous physical region (e.g., 1m×1m), and the weighted center points integrate the temperature and location information of all pixels within that region, effectively "condensing" the overall thermal characteristics of the region. For example, in multiple adjacent grids covered by a human heat source, the weighted center points of each grid will be biased towards the core of the human body, and their temperature characteristics are similar, making them easier to group into a complete cluster during clustering, avoiding segmentation. Third, filtering local noise improves clustering stability. The weighted calculation within the grid (temperature-weighted averaging) naturally smooths out local noise: a single high-temperature noise pixel has a negligible impact on the weighted center point of the entire grid, while the continuous thermal radiation from a real heat source will continuously affect multiple grids, and its characteristics are more prominent at the grid level. For example, a single high-temperature pixel caused by sunlight reflection will be "diluted" by the average calculation within the grid, while the continuous heat dissipation of the human body will cause the weighted center points of multiple grids to exhibit stable high-temperature characteristics.

[0025] Specifically, after identifying a heat source, a multi-dimensional feature vector encompassing location, temperature, displacement, and rate of temperature change can be constructed for that specific heat source. Simultaneously, kinematic parameters such as velocity and thermal diffusivity entropy are calculated, comprehensively reflecting the dynamic thermal characteristics of the heat source. This multi-dimensional description provides richer decision-making basis for the control of the chiller unit, avoiding the problems of untimely or inaccurate control caused by relying solely on a single temperature factor.

[0026] Based on kinematic parameters, the chiller unit's control system classifies heat source motion intensity levels and can formulate appropriate control commands accordingly. When the heat source motion intensity is high and heat diffusion is rapid, cooling is accelerated; when the heat source distribution is stable and the intensity is low, the output is appropriately reduced, thus flexibly adapting to real-time load requirements. The entire process relies on automated operation of data acquisition, algorithm analysis, and command generation, reducing the need for manual intervention. Through continuous analysis of thermal image data using machine learning algorithms, the chiller unit can automatically optimize its operating mode based on dynamic changes in the monitored area's thermal environment, significantly improving the level of intelligent control. Because the chiller unit can precisely adjust its operation according to the actual dynamic conditions of the heat source in the monitored area, it can prevent over-cooling or under-cooling, ensuring effective environmental temperature control while helping to reduce unnecessary energy consumption and achieve energy conservation and emission reduction.

[0027] More preferably, such as Figure 2 As shown, the calculation of a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector includes: S1051: For each identified tracked heat source in the infrared thermal image frame, extract the temperature value of the grid in which it is located, and divide the temperature range into K temperature intervals; S1052: Count the number of radiation points of the tracked heat source within each temperature range to obtain the frequency distribution vector. , where the frequency distribution vector It is the ratio of the number of radiation points in the i-th temperature range to the total number of radiation points; S1053: The corresponding thermal diffusivity value is calculated based on the frequency distribution vector and the thermal diffusivity value calculation formula, and the thermal diffusivity value is used as the kinematic parameter of the tracked heat source, wherein the thermal diffusivity value calculation formula is: H is the thermal diffusivity value. Let K be the frequency distribution vector, and K be the number of temperature ranges.

[0028] In this embodiment of the invention, by dividing the temperature range into K intervals and counting the number of radiation points in each interval, the thermal distribution characteristics of the heat source are transformed into a quantifiable frequency distribution vector, making the originally abstract thermal diffusion state concrete and calculable. This quantification method provides a data foundation for subsequent analysis of the thermal diffusion law of the heat source.

[0029] The thermal diffusion entropy value, calculated using the frequency distribution vector and entropy formula, effectively reflects the disorder and complexity of the heat distribution of a heat source. A higher entropy value indicates that the heat radiation from the heat source is more dispersed across different temperature ranges, resulting in a more complex thermal diffusion state; a lower entropy value indicates that the heat radiation is concentrated in a few temperature ranges, indicating relatively stable thermal diffusion. This precise characterization helps to gain a deeper understanding of the thermal behavior of heat sources. Incorporating the thermal diffusion entropy value into the kinematic parameter system enriches the descriptive dimensions of the tracked heat source's state. Compared to relying solely on parameters such as velocity, the thermal diffusion entropy value provides additional information from the perspective of heat distribution, allowing kinematic parameters to more comprehensively reflect the dynamic characteristics of the heat source and providing a more reliable basis for subsequent classification of motion intensity levels.

[0030] Since the thermal diffusivity entropy value accurately reflects the thermal diffusion of a heat source, the chiller control system can use this value to more accurately determine the degree of influence of the heat source on the ambient temperature. For example, for heat sources with high thermal diffusivity entropy, complex heat distribution, and rapid diffusion, a more aggressive cooling strategy can be adopted; for heat sources with low entropy and stable heat distribution, the cooling intensity can be appropriately adjusted, thereby further improving the targeting and effectiveness of the control.

[0031] By accurately grasping the thermal diffusion characteristics of the heat source, the chiller unit can more rationally match the cooling output, avoiding energy waste caused by inaccurate judgment of the thermal diffusion state. While ensuring a suitable temperature in the monitored area, energy consumption can be further optimized, improving the system's energy-saving performance.

[0032] More preferably, after calculating the corresponding thermal diffusivity value based on the frequency distribution vector and the thermal diffusivity value calculation formula, the method further includes: S1054: Obtain the spatial weight matrix of the corresponding monitoring area, and assign correction coefficients to the radiation point temperatures of different areas based on the spatial weight matrix; S1055: The corrected thermal diffusivity entropy value is determined based on the thermal diffusivity entropy value, the spatial weight matrix, and the entropy correction formula; the entropy correction formula is: ; in, This is the spatially corrected thermal diffusivity value. This is the uncorrected thermal diffusivity value; Let M be the spatial weight of the j-th radiation point, and M be the number of effective radiation points. This is the frequency distribution vector.

[0033] The solution of the present invention By assigning correction coefficients to the radiation point temperatures of different regions using a spatial weight matrix, the calculation of thermal diffusion entropy is no longer limited to simple temperature distribution statistics, but incorporates the spatial characteristics of the monitored area (such as heat dissipation requirements at different locations and the importance of equipment layout). For example, higher weights can be assigned to densely populated areas or critical equipment areas, making the entropy value more closely match the actual temperature control priorities of the space and enhancing the accuracy of the entropy value in representing the real thermal environment.

[0034] The corrected thermal diffusivity entropy (H') distinguishes the contribution of radiation points in different regions through spatial weighting, and can more sensitively capture the thermal diffusivity changes in key areas. Even if the original thermal diffusivity entropy (H) of two heat sources are the same, if their high-temperature radiation points are distributed in regions with different weights (such as the core region and the edge region), the corrected entropy values ​​will also produce significant differences, thus more accurately reflecting the actual impact of heat sources on environmental temperature control.

[0035] Based on the spatially corrected heat diffusion entropy value, the chiller control system can achieve more refined regional control. For example, when the corrected entropy value shows that the heat diffusion in high-weight areas is complex, the cooling in those areas can be prioritized; while the heat diffusion changes in low-weight areas have less impact on the control decision, avoiding a one-size-fits-all approach and improving the spatial adaptability of the control strategy.

[0036] The solution in this invention, through spatial weight correction, allows the system to focus on key areas that have a greater impact on ambient temperature. While meeting the temperature control needs of the core areas, it reduces excessive regulation of secondary areas. This differentiated control can further optimize energy allocation and reduce unnecessary energy consumption while ensuring overall environmental comfort, achieving a dual improvement in energy saving and temperature control.

[0037] The spatial layout of actual monitoring areas is often complex (e.g., there are partitions, uneven equipment distribution, etc.). The introduction of a spatial weight matrix allows the calculation of thermal diffusivity entropy to be adapted to the spatial characteristics of different scenarios. By adjusting the weight coefficients, this method can be flexibly applied to various places such as offices, shopping malls, and industrial workshops, improving the versatility and practicality of the technical solution.

[0038] More preferably, after calculating the corresponding thermal diffusivity value based on the frequency distribution vector and the thermal diffusivity value calculation formula, the method further includes: A sliding window is used to perform sliding calculations on multiple infrared thermal image frames to obtain entropy value sequence data. The smoothed entropy value parameters are then determined based on this entropy value sequence data and an exponential smoothing formula, which is: ,in, The entropy value parameter is the smoothed value at the current moment. The entropy value parameter is the smoothed value from the previous time step. The entropy value parameter at the current moment. For smoothing coefficients; The entropy change rate is calculated, and the entropy change rate and the smoothed entropy parameter are matched with the set conditions to determine the current thermal state. The chiller unit control system is then adjusted according to the current thermal state.

[0039] Specifically, infrared thermal image frames may be subject to noise due to environmental interference (such as instantaneous device heating or light fluctuations), causing abnormal fluctuations in the entropy value of a single frame. By using a sliding window to calculate the entropy value sequence and processing it with an exponential smoothing formula, the smoothed entropy value parameter can effectively filter instantaneous noise, weaken the influence of random factors on heat diffusion characteristics, and make the entropy value more stably reflect the long-term trend of the heat source, providing a more reliable data basis for subsequent judgment.

[0040] In this embodiment of the invention, by calculating the rate of change of entropy, the evolution speed and direction of the thermal diffusion state of the heat source can be quantified. For example, a rapid increase in entropy with a positive rate of change indicates that the complexity of thermal diffusion is increasing; a slow decrease in entropy with a negative rate of change indicates that the heat distribution is stabilizing. This capture of dynamic trends overcomes the limitation that entropy at a single moment can only reflect a static state, enabling the system to more sensitively perceive changes in the thermal environment.

[0041] By combining the smoothed entropy parameter with the rate of entropy change, the current thermal state can be more precisely classified, such as stable low entropy, rapid entropy increase, and slow entropy decrease. Different thermal states correspond to different degrees and rates of change in the impact of heat sources on the environment, allowing the system to match more precise control strategies—for example, initiating a pre-cooling mechanism for a rapidly increasing entropy state and maintaining basic cooling for a stable low entropy state, thereby improving the timeliness and foresight of control decisions.

[0042] The smoothed entropy parameter reduces misjudgments caused by instantaneous fluctuations. Combined with setting trigger conditions based on the entropy change rate (such as adjusting only when the change rate exceeds a threshold), it can prevent the chiller unit from frequently starting and stopping or drastically adjusting its output due to minor disturbances. This buffering mechanism reduces mechanical wear and tear on the equipment, extends its service life, and reduces the additional energy consumption caused by frequent adjustments.

[0043] In dynamic scenarios such as frequent personnel movement and intermittent equipment operation, the thermal diffusion state of a heat source often exhibits nonlinear changes. The combination of sliding window and exponential smoothing can adapt to this dynamic characteristic, preserving the influence of recent data (by adjusting sensitivity through the α coefficient) while also taking into account historical trends, enabling the system to maintain stable and accurate control performance even in complex scenarios.

[0044] More preferably, the intelligent control method further includes: The current scene type information is determined, and the corresponding temperature range information is determined based on the scene type information. The temperature range information is obtained by training historical thermal radiation data of different scenes using the K-means clustering algorithm.

[0045] The thermal environment characteristics of different scenarios (such as offices, conference rooms, stadiums, and production workshops) vary significantly (e.g., normal temperature range, heat source distribution patterns). By training historical thermal radiation data for different scenarios using the K-means clustering algorithm, temperature range information matching the scenario type is obtained. This eliminates the need for a uniform standard in dividing temperature ranges, instead tailoring them to the specific thermal characteristics of each scenario, thus improving the rationality and relevance of temperature range division.

[0046] Temperature ranges are fundamental for calculating frequency distribution vectors and thermal diffusivity entropy values. Contextualized temperature range information allows entropy calculations to better reflect the actual heat distribution patterns of the current scenario. For example, the temperature range division for a conference room scenario considers the heat dissipation characteristics of people gathering, while for a server room, it focuses on the temperature range of equipment operation, thus allowing the thermal diffusivity entropy value to more accurately reflect the heat source state in that scenario.

[0047] By identifying the current scenario type and calling the corresponding temperature range parameters, the system can flexibly respond to the temperature control requirements of different application scenarios. When the scenario changes (such as a meeting room changing from empty to full), it can automatically adapt to the temperature range standard of the new scenario, avoiding misjudgment of thermal status due to scenario changes, and ensuring that the control strategy of the chiller unit always matches the scenario characteristics.

[0048] Training the system with K-means clustering on historical thermal radiation data essentially allows the system to learn the thermal characteristics of different scenarios from historical operations, reducing the subjectivity and error of manually setting temperature ranges. As historical data accumulates, the temperature range information can be continuously optimized, enabling the system to gain a deeper understanding of the thermal environment in various scenarios and improving the robustness and adaptability of the overall control algorithm.

[0049] The association between scene type information and temperature range allows the control of chiller units to not only rely on real-time thermal data, but also to incorporate the inherent attributes of the scene (such as the comfort temperature preferences of office scenes and the equipment protection temperature thresholds of industrial scenes). This combination makes the generation of control commands more aligned with the functional requirements of the scene, further optimizing energy consumption and operating efficiency while ensuring core temperature control objectives (such as personnel comfort and equipment safety).

[0050] More preferably, the dynamic clustering algorithm is the DBSCAN algorithm, and the intelligent control method further includes: If a clustered heat source has a displacement equal to or greater than a first threshold and a temperature equal to or greater than a second threshold within a preset number of frames, it is classified as an active human heat source.

[0051] Calculate the temperature decay rate at the previous location of the active human body after the heat source is removed, and use the temperature decay rate as a corresponding kinematic parameter. The intelligent control method further includes: Monitor multiple heat sources; A high-intensity mass event is detected when a predetermined number of heat sources simultaneously meet the conditions of having a velocity value exceeding a velocity threshold and a thermal diffusion entropy value exceeding an entropy threshold within a specified time window. Control commands for the chiller unit control system are generated based on the detected high-intensity events in the group.

[0052] This invention utilizes the DBSCAN algorithm to dynamically cluster heat sources and, combined with displacement and temperature thresholds (a first threshold and a second threshold), classifies specific heat sources as active human heat sources. This effectively distinguishes the human body from other static heat sources (such as equipment). The human body, as a dynamic heat dissipation source, exhibits fundamentally different heat dissipation characteristics compared to equipment. Identifying it separately allows for optimization of cooling strategies for areas of human activity. Furthermore, calculating the temperature decay rate after the human body is removed reflects the rate of residual heat dissipation in that area, providing a more precise basis for determining whether to maintain cooling or adjust its intensity, thus avoiding over-cooling of unoccupied areas.

[0053] By monitoring multiple heat sources and setting time windows, a high-intensity mass event is identified when a predetermined number of heat sources simultaneously meet the threshold values ​​for both velocity and thermal diffusion entropy. These events typically lead to a rapid rise in ambient temperature and complex thermal diffusion. Based on this, the system can proactively activate an enhanced cooling mode to avoid the lag caused by temperature accumulation before subsequent adjustments, thus improving temperature control in densely populated areas.

[0054] The DBSCAN algorithm, as a density-based clustering method, does not require pre-setting the number of clusters and can adaptively identify heat source groups of different densities, making it particularly suitable for complex environments where human and device heat sources are mixed. Compared to traditional clustering algorithms, it has advantages in handling noisy points (such as instantaneous heat sources) and irregularly shaped heat sources, reducing the possibility of misclassifying non-human heat sources as active human bodies and improving the accuracy of heat source classification.

[0055] Based on the individual characteristics of active human heat sources (such as movement trajectory and temperature decay) and the overall characteristics of high-intensity group events, the system can formulate a tiered control strategy: providing precise localized cooling for individual active human areas; enhancing overall cooling for areas experiencing high-intensity group events; and reducing cooling output for areas without active human activity and where the temperature has decayed to a reasonable range. This differentiated control ensures comfort in key areas while minimizing ineffective energy consumption.

[0056] Whether it's the movement of a single person or a large-scale group activity, the system can respond quickly through mechanisms such as heat source clustering and feature threshold judgment. This method is applicable to various scenarios with frequent personnel flow, such as office areas, shopping malls, conference rooms, and production workshops. It solves the problem of insufficient response of traditional control methods to heat dissipation from dynamic crowds, and improves the versatility and practicality of intelligent control methods.

[0057] More preferably, the intelligent control method further includes: Construct a trained LSTM model; The constructed multidimensional feature vector is input into the LSTM model for identification to determine the motion intensity in the next stage; The control commands for the chiller unit control system are generated based on the intensity of the next stage of motion. Alternatively, the intelligent control method may further include: A corresponding heat source map is generated based on a series of infrared thermal image frames of the monitored area obtained by each infrared camera, and the heat source map is superimposed with the air outlet map of the chiller unit to obtain the control map. The process of generating control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmitting the control commands to the corresponding chiller unit control system to regulate its operation, includes: The received motion intensity level, kinematic parameters, and control map are input into the pre-built air supply control model to determine the optimal air supply parameters for each cooling air outlet. The optimal air supply parameters include air volume parameters, air direction parameters, and temperature parameters. Based on the optimal air supply parameters, control commands are generated for the chiller control system, and the control commands are transmitted to the corresponding chiller control system to regulate its operation.

[0058] The solution in this invention analyzes multi-dimensional feature vectors using a trained LSTM model to predict the intensity of heat source movement in the next stage. This predictive capability transforms chiller unit control from passively responding to the current state to proactively anticipating future demands. For example, by predicting the imminent appearance of a high-intensity heat source, it can proactively improve cooling preparations, avoiding discomfort caused by temperature fluctuations and significantly enhancing the timeliness and predictability of temperature control. In other words, it visualizes the heat distribution of the entire venue as a heat source map and overlays it with a wind vent map.

[0059] By overlaying heat source maps and air outlet maps to generate a control map, the spatial correspondence between heat source distribution and cooling equipment is visually presented. Based on this, the air supply control model can calculate the optimal air supply parameters (air volume, air direction, temperature) for each cooling air outlet, achieving precise control of where cooling is needed and where air is precisely supplied. For example, increasing air volume and adjusting air direction to directly blow on heat dissipation points in high-density heat source areas, and reducing output in low-temperature areas, can improve local temperature control while avoiding energy waste.

[0060] LSTM models excel at capturing long-term dependencies in time-series data and can effectively predict trends in heat source motion intensity, making them particularly suitable for scenarios with periodic or trend-based patterns, such as personnel movement and equipment operation. Meanwhile, spatial matching of heat source-air vent maps solves the problem of uneven cooling in traditional control systems. The combination of these two technologies enables the system to maintain stable and efficient control performance even in dynamically changing and complex environments.

[0061] The predicted output of the LSTM model and the parameter calculation of the air supply control model form a closed-loop decision chain, reducing the need for manual intervention. The system can autonomously complete the entire process of state perception, trend prediction, spatial matching, parameter optimization, and instruction generation, which not only improves control efficiency but also continuously optimizes the decision logic through continuous learning (such as iterative training of the LSTM model), achieving self-improvement in intelligent control.

[0062] By using LSTM models for advance control, energy peaks caused by temporary high-intensity cooling are avoided; and by precisely optimizing air outlet parameters, energy loss from ineffective air supply is reduced. Together, these two methods minimize the overall energy consumption of the chiller unit while ensuring accurate temperature control in each area, achieving the dual goals of precise cooling and energy-saving operation.

[0063] Specifically, the core components and parameter design of the LSTM model Input Layer: Temporal processing of multidimensional feature vectors. Input data format: The model input is a historical time sequence, not single-frame features. For example, multidimensional feature vectors from the past 10 frames (each frame 10 seconds apart) are used to form an input sequence of shape (10,D) (D is the feature dimension, such as position, temperature, displacement, entropy, etc., assumed to be D=8). Function: To standardize the feature vectors (e.g., normalize to the 0-1 range) and convert them into a tensor format that LSTM can process.

[0064] LSTM layer: captures temporal dependencies. Number of layers and units: usually 2 layers of LSTM are used (can be adjusted according to the amount of data). The first layer has 64 hidden units (captures local temporal features) and the second layer has 32 hidden units (extracts long-term trends).

[0065] Key parameters: return_sequences=True (first layer): Returns the output of each time step for processing by the second LSTM layer; return_sequences=False (second layer): Returns only the output of the last time step, representing the compressed features of the entire sequence; Add a Dropout layer (e.g., dropout=0.2): Prevents overfitting and enhances the model's generalization ability.

[0066] Fully connected layers: Feature integration and dimensionality compression. One to two fully connected layers are used. For example: Layer 1: 16 neurons, activation function is ReLU (enhancing nonlinear fitting ability); Layer 2: 8 neurons, further compressing feature dimensions. Function: Transforms the high-dimensional feature vector output by the LSTM into compact features relevant to the output target.

[0067] Output layer: Predicts motion intensity. Output format: Selected based on the definition of motion intensity (e.g., discrete levels 0-5 or continuous values): For discrete levels (classification tasks): Number of neurons in the output layer = number of levels (e.g., 6), Softmax activation function, outputting the probability of each level; For continuous values ​​(regression tasks): Output layer with 1 neuron, no activation function (or Linear), directly outputting the predicted intensity value. Loss function: Cross-entropy loss for classification tasks, mean squared error (MSE) for regression tasks. Then, historical training data is used as input and continuously optimized to obtain the final model parameters.

[0068] Whether it's a dynamic scenario relying on time-series prediction or a complex layout requiring precise spatial control, this method can adapt to the requirements by flexibly calling the prediction model or spatial matching mechanism, further expanding the application scenarios of the chiller control system and making it more applicable.

[0069] In this embodiment of the invention, the machine learning-based intelligent control method for chiller units classifies motion intensity levels according to kinematic parameters. The chiller unit control system can formulate appropriate control commands based on the motion intensity of the heat source. When the heat source motion intensity is high and heat diffusion is fast, cooling is accelerated; when the heat source distribution is stable and the intensity is low, the output is appropriately reduced, thereby flexibly adapting to real-time load requirements.

[0070] Example 2 Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of the intelligent control system for a chiller unit based on machine learning, as disclosed in an embodiment of the present invention. Figure 3 As shown, the machine learning-based intelligent control system for the chiller unit may include: Acquisition module 21: used to acquire a series of infrared thermal image frames of the monitored area through various infrared cameras set in the monitoring area; Calculation module 22: used to map the thermal radiation data from each infrared thermal image frame onto an N×N grid representing the physical coordinates of the monitored area, and to calculate the weighted center point of the grid cell, each weighted center point being represented by coordinates and a normalized temperature value; Clustering module 23: Used to perform a dynamic clustering algorithm on weighted centroids in consecutive frames to identify one or more independent heat sources and track their trajectories; Feature construction module 24: for each tracked heat source, construct a multi-dimensional feature vector, the multi-dimensional feature vector containing location information, monitored temperature information, inter-frame displacement and temperature change rate; Motion calculation module 25: used to calculate a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector, the set of kinematic parameters including at least velocity value and thermal diffusivity value; and to assign a motion intensity level to the tracked heat source based on the calculated kinematic parameters; Control module 26: Used to determine the motion intensity level and kinematic parameters, generate control commands for the chiller control system based on the received motion intensity level and kinematic parameters, and transmit the control commands to the corresponding chiller control system to regulate its operation.

[0071] In this embodiment of the invention, the machine learning-based intelligent control method for chiller units classifies motion intensity levels according to kinematic parameters. The chiller unit control system can formulate appropriate control commands based on the motion intensity of the heat source. When the heat source motion intensity is high and heat diffusion is fast, cooling is accelerated; when the heat source distribution is stable and the intensity is low, the output is appropriately reduced, thereby flexibly adapting to real-time load requirements.

[0072] Example 3 Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device can be a computer, a server, etc. Of course, in certain cases, it can also be a mobile phone, tablet computer, monitoring terminal, or other smart device, as well as an image acquisition device with processing capabilities. Figure 4 As shown, the electronic device may include: Memory 510 storing executable program code; Processor 520 coupled to memory 510; The processor 520 calls the executable program code stored in the memory 510 to execute some or all of the steps in the machine learning-based intelligent control method for chiller units in Embodiment 1.

[0073] This invention discloses a computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps in the machine learning-based intelligent control method for chiller units described in Embodiment 1.

[0074] This invention also discloses a computer program product, wherein when the computer program product is run on a computer, the computer executes some or all of the steps in the machine learning-based intelligent control method for chiller units in Embodiment 1.

[0075] This invention also discloses an application publishing platform, which is used to publish computer program products. When the computer program products are run on a computer, the computer executes some or all of the steps in the machine learning-based intelligent control method for chiller units in Embodiment 1.

[0076] In various embodiments of the present invention, it should be understood that the sequence number of each process does not necessarily imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they can be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0078] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0079] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several requests to cause a computer device (which can be a personal computer, server, or network device, specifically a processor in the computer device) to execute some or all of the steps of the methods described in the various embodiments of the present invention.

[0080] In the embodiments provided by this invention, it should be understood that B corresponding to A means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0081] Those skilled in the art will understand that some or all of the steps in the various methods of the embodiments described can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0082] The above provides a detailed description of the intelligent control method, system, electronic device, and storage medium for chiller units based on machine learning disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A machine learning-based intelligent control method for chiller units, characterized in that, include: A series of infrared thermal image frames of the monitored area are acquired by various infrared cameras installed in the monitored area. Thermal radiation data from each infrared thermal image frame is mapped onto an N×N grid representing the physical coordinates of the monitored area, and the weighted center point of the grid cell is calculated, with each weighted center point represented by coordinates and a normalized temperature value. A dynamic clustering algorithm is performed on the weighted center points in consecutive frames to identify one or more independent heat sources and track their trajectories; The dynamic clustering algorithm is the DBSCAN algorithm; if the displacement of a clustered heat source within a preset number of frames is equal to or greater than a first threshold and its temperature is equal to or greater than a second threshold, it is classified as an active human heat source; the temperature decay rate at its previous position after the active human heat source is moved away is calculated, and the temperature decay rate is also used as the corresponding kinematic parameter. For each tracked heat source, a multidimensional feature vector is constructed, which contains information representing location, monitored temperature, inter-frame displacement, and temperature change rate. A set of kinematic parameters of the tracked heat source is calculated based on the multidimensional feature vector, the set of kinematic parameters including at least velocity and thermal diffusivity values; and a motion intensity level is assigned to the tracked heat source based on the calculated kinematic parameters; the set of kinematic parameters of the tracked heat source calculated based on the multidimensional feature vector includes; For each identified tracked heat source in the infrared thermal image frame, the temperature value of its corresponding grid is extracted, and the temperature range is divided into K temperature intervals. The frequency distribution vector is obtained by counting the number of radiation points of the tracked heat source within each temperature range. , where the frequency distribution vector It is the ratio of the number of radiation points in the i-th temperature range to the total number of radiation points; The corresponding thermal diffusivity value is calculated based on the frequency distribution vector and the thermal diffusivity value calculation formula, and the thermal diffusivity value is used as the kinematic parameter of the tracked heat source. The thermal diffusivity value calculation formula is as follows: H is the thermal diffusivity value. Let K be the frequency distribution vector, and K be the number of temperature ranges. A sliding window is used to perform sliding calculations on multiple infrared thermal image frames to obtain entropy value sequence data. The smoothed entropy value parameters are then determined based on this entropy value sequence data and an exponential smoothing formula, which is: ,in, The entropy value parameter is the smoothed value at the current moment. The entropy value parameter is the smoothed value from the previous time step. The entropy value parameter at the current moment. For smoothing coefficients; Calculate the entropy change rate, match the entropy change rate and the smoothed entropy parameter with the set conditions to determine the current thermal state, and adjust the chiller unit control system according to the current thermal state. The motion intensity level and kinematic parameters are determined, and control commands for the chiller unit control system are generated based on the received motion intensity level and kinematic parameters. The control commands are then transmitted to the corresponding chiller unit control system to regulate its operation.

2. The intelligent control method for chiller units based on machine learning as described in claim 1, characterized in that, After calculating the corresponding thermal diffusivity value based on the frequency distribution vector and the thermal diffusivity value calculation formula, the method further includes: Obtain the spatial weight matrix of the corresponding monitoring area, and assign correction coefficients to the radiation point temperatures of different areas based on the spatial weight matrix. The corrected thermal diffusivity entropy value is determined based on the thermal diffusivity entropy value, the spatial weight matrix, and the entropy correction formula; the entropy correction formula is as follows: ; in, This is the spatially corrected thermal diffusivity value. This is the uncorrected thermal diffusivity value; Let M be the spatial weight of the j-th radiation point, and M be the number of effective radiation points. This is the frequency distribution vector.

3. The intelligent control method for chiller units based on machine learning as described in claim 1, characterized in that, The intelligent control method further includes: The current scene type information is determined, and the corresponding temperature range information is determined based on the scene type information. The temperature range information is obtained by training historical thermal radiation data of different scenes using the K-means clustering algorithm.

4. The intelligent control method for chiller units based on machine learning as described in claim 1, characterized in that, The intelligent control method further includes: Monitor multiple heat sources; A high-intensity mass event is detected when a predetermined number of heat sources simultaneously meet the conditions of having a velocity value exceeding a velocity threshold and a thermal diffusion entropy value exceeding an entropy threshold within a specified time window. Control commands for the chiller unit control system are generated based on the detected high-intensity events in the group.

5. The intelligent control method for chiller units based on machine learning as described in claim 1, characterized in that, The intelligent control method further includes: Construct a trained LSTM model; The constructed multidimensional feature vector is input into the LSTM model for identification to determine the motion intensity level in the next stage; The control commands for the chiller unit control system are generated based on the intensity level of the next stage of the exercise. Alternatively, the intelligent control method may further include: A corresponding heat source map is generated based on a series of infrared thermal image frames of the monitored area obtained by each infrared camera, and the heat source map is superimposed with the air outlet map of the chiller unit to obtain the control map. The process of generating control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmitting the control commands to the corresponding chiller unit control system to regulate its operation, includes: The received motion intensity level, kinematic parameters, and control map are input into the pre-built air supply control model to determine the optimal air supply parameters for each cooling air outlet. The optimal air supply parameters include air volume parameters, air direction parameters, and temperature parameters. Based on the optimal air supply parameters, control commands are generated for the chiller control system, and the control commands are transmitted to the corresponding chiller control system to regulate its operation.

6. A machine learning-based intelligent control system for a chiller unit, characterized in that, include: Acquisition module: Used to acquire a series of infrared thermal image frames of the monitored area through various infrared cameras set up in the monitored area; Calculation module: used to map thermal radiation data from each infrared thermal image frame onto an N×N grid representing the physical coordinates of the monitored area, and to calculate the weighted center point of the grid cell, each weighted center point being represented by coordinates and a normalized temperature value; Clustering module: Used to perform dynamic clustering algorithms on weighted centroids in consecutive frames to identify one or more independent heat sources and track their trajectories; The dynamic clustering algorithm is the DBSCAN algorithm; if the displacement of a clustered heat source within a preset number of frames is equal to or greater than a first threshold and its temperature is equal to or greater than a second threshold, it is classified as an active human heat source; the temperature decay rate at its previous position after the active human heat source is moved away is calculated, and the temperature decay rate is also used as the corresponding kinematic parameter. Feature construction module: used to construct a multi-dimensional feature vector for each tracked heat source, the multi-dimensional feature vector containing location information, monitored temperature information, inter-frame displacement and temperature change rate; Motion calculation module: used to calculate a set of kinematic parameters of the tracked heat source based on the multidimensional feature vector, the set of kinematic parameters including at least velocity value and thermal diffusivity value; And assign a motion intensity level to the tracked heat source based on the calculated kinematic parameters; the set of kinematic parameters of the tracked heat source calculated based on the multidimensional feature vector includes; For each identified tracked heat source in the infrared thermal image frame, the temperature value of its corresponding grid is extracted, and the temperature range is divided into K temperature intervals. The frequency distribution vector is obtained by counting the number of radiation points of the tracked heat source within each temperature range. , where the frequency distribution vector It is the ratio of the number of radiation points in the i-th temperature range to the total number of radiation points; The corresponding thermal diffusivity value is calculated based on the frequency distribution vector and the thermal diffusivity value calculation formula, and the thermal diffusivity value is used as the kinematic parameter of the tracked heat source. The thermal diffusivity value calculation formula is as follows: H is the thermal diffusivity value. Let K be the frequency distribution vector, and K be the number of temperature ranges. A sliding window is used to perform sliding calculations on multiple infrared thermal image frames to obtain entropy value sequence data. The smoothed entropy value parameters are then determined based on this entropy value sequence data and an exponential smoothing formula, which is: ,in, The entropy value parameter is the smoothed value at the current moment. The entropy value parameter is the smoothed value from the previous time step. The entropy value parameter at the current moment. For smoothing coefficients; Calculate the entropy change rate, match the entropy change rate and the smoothed entropy parameter with the set conditions to determine the current thermal state, and adjust the chiller unit control system according to the current thermal state. Control module: used to determine the motion intensity level and kinematic parameters, generate control commands for the chiller unit control system based on the received motion intensity level and kinematic parameters, and transmit the control commands to the corresponding chiller unit control system to regulate its operation.

7. An electronic device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the machine learning-based intelligent control method for chiller units as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to execute the machine learning-based intelligent control method for chiller units as described in any one of claims 1 to 5.

Images