Low-carbon refrigeration and snowmaking automated intelligent control system
The low-carbon refrigeration snowmaking system, which combines a mobile industrial camera and a deep convolutional neural network with a heat transfer model, solves the problems of decreased heat exchange efficiency and increased energy consumption caused by frost accumulation, and achieves efficient and low-carbon frost detection and defrosting optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ETERNAL STAR ICE & SNOW(BEIJING) TECH LTD CO
- Filing Date
- 2026-02-11
- Publication Date
- 2026-05-26
AI Technical Summary
In existing refrigeration and snowmaking systems, frost accumulation leads to decreased heat exchange efficiency and increased energy consumption. Traditional defrosting methods suffer from high misjudgment rates and difficulty in balancing energy efficiency improvement with low-carbon operation.
A mobile industrial camera combined with an automatic navigation rail is used for full-coverage scanning. Frost layers are identified through HSV color space conversion and deep convolutional neural network. An equivalent thermal resistance model of the fins is established based on the principles of heat and mass transfer. The defrosting timing is optimized based on an energy efficiency scheduling algorithm with a profit function.
It achieves high reliability and accuracy in frost detection, reduces energy consumption and carbon emissions, and improves the system's energy efficiency ratio and operational economy.
Smart Images
Figure CN122083570A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of refrigeration technology, specifically to a low-carbon refrigeration snowmaking automated intelligent control system. Background Technology
[0002] In existing refrigeration and snowmaking systems, frost easily forms on the surface of the air cooler. This is because the surface temperature of the air cooler is usually much lower than the dew point temperature of the ambient air, or even lower than 0°C, causing the moisture in the air to condense and freeze into a frost layer.
[0003] As frost gradually accumulates, it has a significant negative impact on system performance. Frost directly increases the thermal resistance between the refrigerant and air in the evaporator, leading to a substantial decrease in heat exchange efficiency. In terms of energy consumption, frost causes the system to run for longer periods to compensate for heat losses, increasing the compressor load and significantly reducing the overall coefficient of performance (COP). In high-load, continuous operation scenarios such as snowmaking machines, this can increase total system energy consumption by more than 20%, or even higher. Furthermore, the defrosting process itself consumes a significant amount of energy: traditional defrosting methods (such as electric heating, reverse-cycle hot air defrosting, or hot water defrosting) require suspending cooling, introducing additional heat loads (such as defrost water backflushing or defrosting energy consumption), and causing heat intrusion during shutdown, further amplifying energy losses.
[0004] Traditional defrosting methods mainly rely on timed defrosting or manual / semi-automatic intervention based on simple temperature / pressure thresholds. These methods have significant drawbacks:
[0005] (1) Timed defrosting is often a "one-size-fits-all" approach. Defrosting too early when the frost layer is not thick causes unnecessary energy consumption and downtime losses, while delaying intervention when the frost layer is too thick leads to a serious reduction in efficiency.
[0006] (2) Temperature sensor judgment is easily affected by environmental fluctuations, lighting or local unevenness, resulting in a high misjudgment rate;
[0007] (3) The lack of precise identification of the thickness, density and phase of the frost layer makes it impossible to achieve accurate intervention;
[0008] (4) It is difficult to achieve a balance between energy efficiency improvement and low-carbon operation. In application scenarios such as snowmaking sites, which have extremely high requirements for stable low temperature and large cooling capacity, energy waste and carbon emission problems are more prominent.
[0009] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0010] The purpose of this invention is to provide a low-carbon automated intelligent control system for refrigeration and snowmaking to solve the problems mentioned in the background art.
[0011] To achieve the above objectives, the present invention provides the following technical solution: a low-carbon refrigeration and snowmaking automated intelligent control system, specifically including: a data acquisition and processing module, a frost layer separation module, a frost layer identification module, a correlation analysis module, and a decision optimization module;
[0012] Acquisition and processing module: Acquires image data of the surface of the air cooler using a mobile industrial camera, and performs preprocessing on the image data;
[0013] Frost separation module: Converts the preprocessed image data to the HSV color space, performs adaptive threshold calibration based on the differences in HSV space components, and initially separates the frost area through binarization processing;
[0014] Frost layer recognition module: The frost layer candidate area is input into a deep convolutional neural network with three parallel feature extraction branches for color, contour and brightness. A three-dimensional mutual reward function is embedded in the frost physical model based on the principle of heat and mass transfer. The frost coverage rate is calculated by ratio calculation.
[0015] Correlation analysis module: Based on the frost coverage rate, establish the fin equivalent thermal resistance model, calculate the deviation of the actual refrigeration system energy efficiency ratio from the theoretical frost-free state, and calculate the influence coefficient of frost thickness on heat exchange efficiency in real time based on the multi-dimensional features output by the frost identification module.
[0016] Decision optimization module: The energy efficiency scheduling algorithm based on the revenue function takes minimizing the total operating energy consumption of the system as the objective function, dynamically calculates and outputs the best defrosting intervention command.
[0017] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0018] The mobile industrial camera is mounted on an automatic navigation rail directly in front of the air cooler. It performs a timed reciprocating full-coverage scan of the air cooler array according to a preset step value to acquire image data of the air cooler surface.
[0019] The preprocessing includes: denoising the image data using Gaussian filtering, and enhancing the contrast between the frost layer and the fin background on the surface of the air cooler using adaptive histogram equalization with limited contrast.
[0020] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0021] Based on the characteristic that the HSV color space is not sensitive to changes in lighting, the frost layer is decoupled from the background of the air cooler surface to obtain three independent channels: hue, saturation, and brightness.
[0022] Extract the grayscale and brightness distribution of image data;
[0023] Histogram statistics are performed on the saturation matrix channels based on the two distributions to identify significant peaks on the surface of the air cooler and the frost layer. The saturation component value with the largest inter-class variance is calculated using the Otsu method as the initial segmentation threshold. A dynamic correction coefficient is introduced based on the average brightness value of the brightness channel to compensate for the initial segmentation threshold and generate the final judgment threshold.
[0024] If a pixel satisfies the condition that the saturation component value is less than the final judgment threshold and its brightness is within a preset reasonable range, it is marked as a frost layer candidate area; otherwise, it is marked as a cold air blower background area, and a preliminary binarization mask is generated.
[0025] Topological correction is performed on the noise and holes generated after binarization, specifically including:
[0026] The opening operation first performs an erosion operation to eliminate small, isolated noise points, and then performs a dilation operation to restore the outline of the target body.
[0027] By using closing operations to fill the tiny holes inside the frost layer caused by uneven light and shadow, the recognition area becomes more coherent. The pixel area of each connected region is calculated, and noisy areas below a set threshold are removed.
[0028] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0029] The frost candidate region is input into a pre-constructed deep convolutional neural network, and three parallel feature extraction branches are configured, including a color dimension branch, a contour dimension branch, and a brightness dimension branch.
[0030] Based on the principles of heat and mass transfer, a physical model for frosting is pre-defined, and the coupling relationships of each dimension under the ideal frosting state are defined, including positive correlation constraints and consistency constraints.
[0031] The positive correlation constraint is that as the thickness of the frost layer increases, the complexity of its surface texture should increase synchronously; the consistency constraint is that the increase in the whiteness of the color component should have a linear logical relationship with the expansion trend of the contour.
[0032] A three-dimensional mutual reward function is additionally embedded into the training loss function of the deep convolutional neural network to calculate the matching degree between the feature extraction branch and the physical law model, and to perform dynamic rewards, specifically including:
[0033] When the color whiteness, contour geometric expansion and brightness gradient depth are simultaneously enhanced and conform to the preset physical evolution model, the activation factor is activated, the combination is given the highest feature weight, and the result is fed back to the neuron activation function.
[0034] If a feature in a certain dimension undergoes a sudden change, the reward function outputs zero, thus suppressing such non-frost interference.
[0035] Based on the combination of feature extraction branches with high reward weights, the edge reconstruction of the frost candidate region is performed, and the physical edge of the frost region is locked by feature map deconvolution to generate a high-confidence frost contour image.
[0036] Based on the high-confidence frosting contour image, the following operation is performed:
[0037] Calculate the total area of pixels with a pixel value of 1 in the mask image;
[0038] The ratio of the total pixel area to the total pixel area of the air cooler's heat exchange area is calculated to obtain the frost coverage rate, which is then transmitted to the decision optimization module in real time as a status parameter.
[0039] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0040] The correlation analysis module collects real-time pressure, temperature and power consumption parameters, establishes an equivalent thermal resistance model for the fins based on the frost coverage rate, and calculates the deviation between the actual energy efficiency ratio of the refrigeration system under the current state and the theoretical energy efficiency ratio of the frost-free refrigeration system.
[0041] The correlation analysis module monitors the offset between evaporation pressure and fan current, and calculates the influence coefficient of the current frost layer thickness on the heat exchange efficiency in real time based on the multidimensional features output by the frost layer identification module.
[0042] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0043] The decision optimization module adopts an energy efficiency scheduling algorithm based on a benefit function. When the expected improvement in cooling efficiency after defrosting is greater than the sum of the energy consumption cost of the defrosting process itself and the heat loss during shutdown, it is determined to be the optimal defrosting time.
[0044] The reward function incorporates the reward mechanism of reinforcement learning, and its expression includes:
[0045]
[0046] Where R represents the energy savings achieved after performing the defrosting action. This indicates the predicted power consumption under the current frosting condition. This indicates the baseline power consumption after defrosting. This indicates the instantaneous energy consumption of the defrosting operation. This indicates the intrusion of defrosting heat load during the shutdown period.
[0047] As a preferred embodiment of the low-carbon refrigeration and snowmaking automated intelligent control system described in this invention, wherein:
[0048] A closed-loop calibration is established. After each defrosting operation, the actual energy efficiency recovery data is compared with the model's predicted benefits, and the dimensional reward weights of the neural network are corrected through the backpropagation algorithm.
[0049] On the other hand, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements the steps of the low-carbon refrigeration snowmaking automated intelligent control system as described above.
[0050] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of the low-carbon refrigeration snowmaking automated intelligent control system as described above in the present invention.
[0051] The technical effects and advantages provided by the present invention in the above technical solution are as follows:
[0052] (1) A full-coverage scanning method using a mobile industrial camera combined with an automatic navigation rail replaces the traditional point sensor to achieve comprehensive and high spatial resolution visual perception of the frost distribution on the surface of the air cooler array fins. Compared with traditional methods that are easily affected by local unevenness, sensor drift, or installation position limitations, this method can capture the overall information of frost thickness, density, coverage uniformity, and phase evolution, avoiding misjudgment or omission, and improving the reliability and accuracy of frost detection.
[0053] (2) By combining HSV color space conversion with adaptive Otsu method threshold calibration and brightness correction coefficient binarization separation strategy, the interference of light change, shadow and fin metal reflection on frost layer segmentation is effectively decoupled.
[0054] (3) Adaptive precision adjustment of morphological opening / closing operations further eliminates noise, voids and noise, ensuring that the generated frost mask is of high quality and consistent.
[0055] (4) A three-dimensional mutual reward function mechanism based on the physical constraints of heat and mass transfer is introduced and embedded into the loss function of the deep convolutional neural network. Through the positive correlation and consistency constraints of multi-dimensional features such as color whiteness, contour complexity and brightness gradient, the fine identification of the phase transition of frost layer from "loose state to ice crystal state" is realized, and non-frost interference such as light flicker and local stains is effectively suppressed. This can significantly improve the generalization ability and confidence of the model in complex industrial environments.
[0056] (5) By combining the frost coverage rate with the pressure, temperature and power consumption parameters collected in real time, a dynamic model of the equivalent thermal resistance of the fins is established, and the influence coefficient of the frost thickness on the heat exchange efficiency is calculated. The energy efficiency ratio reduction and energy consumption increment caused by the frost layer can be quantified in real time.
[0057] (6) An energy efficiency scheduling model based on reinforcement learning is constructed with the goal of minimizing total operating energy consumption. A reward function is introduced to comprehensively consider the efficiency improvement benefits after defrosting, the instantaneous energy consumption of defrosting, and the heat loss during shutdown. At the same time, a power load-side response mechanism is embedded to dynamically adjust the decision threshold according to the peak and valley periods of external electricity prices and the priority of snowmaking demand, so as to realize the "peak shifting and valley filling" operation. Attached Figure Description
[0058] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0059] Figure 1 This is a flowchart of the method for the low-carbon refrigeration and snowmaking automated intelligent control system of the present invention.
[0060] Figure 2 This is a schematic diagram of the modules of the low-carbon refrigeration and snowmaking automated intelligent control system of the present invention. Detailed Implementation
[0061] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.
[0062] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a low-carbon refrigeration snowmaking automated intelligent control system, specifically including: a data acquisition and processing module, a frost layer separation module, a frost layer identification module, a correlation analysis module, and a decision optimization module;
[0063] Acquisition and processing module: Acquires image data of the surface of the air cooler using a mobile industrial camera, and performs preprocessing on the image data;
[0064] The mobile industrial camera is mounted on an automatic navigation rail directly in front of the air cooler. It performs a timed reciprocating full-coverage scan of the air cooler array according to a preset step value to acquire image data of the air cooler surface.
[0065] The preprocessing includes: denoising the image data using Gaussian filtering, and enhancing the contrast between the frost layer and the fin background on the surface of the air cooler using adaptive histogram equalization with limited contrast.
[0066] Frost separation module: Converts the preprocessed image data to the HSV color space, performs adaptive threshold calibration based on the differences in HSV space components, and initially separates the frost area through binarization processing;
[0067] Based on the characteristic that the HSV color space is not sensitive to changes in lighting, the frost layer is decoupled from the background of the air cooler surface to obtain three independent channels: hue, saturation, and brightness.
[0068] Extract the grayscale and brightness distribution of image data;
[0069] Histogram statistics are performed on the saturation matrix channels based on the two distributions to identify significant peaks on the surface of the air cooler and the frost layer. The saturation component value with the largest inter-class variance is calculated using the Otsu method as the initial segmentation threshold. A dynamic correction coefficient is introduced based on the average brightness value of the brightness channel to compensate for the initial segmentation threshold and generate the final judgment threshold.
[0070] If a pixel satisfies the condition that the saturation component value is less than the final judgment threshold and its brightness is within a preset reasonable range, it is marked as a frost layer candidate area; otherwise, it is marked as a cold air blower background area, and a preliminary binarization mask is generated.
[0071] Topological correction is performed on the noise and holes generated after binarization, specifically including:
[0072] The opening operation first performs an erosion operation to eliminate small, isolated noise points, and then performs a dilation operation to restore the outline of the target body.
[0073] By filling the tiny holes inside the frost layer caused by uneven light and shadow through closing operations, the recognition area becomes more coherent. The pixel area of each connected region is calculated, and noisy areas below a set threshold are removed.
[0074] It should also be noted that the opening and closing operations are morphological post-processing. Based on the energy efficiency priority fed back by the intelligent decision-making module, the mask reconstruction accuracy is adaptively selected to achieve an energy efficiency balance between recognition computing power consumption and recognition accuracy.
[0075] Frost layer recognition module: The frost layer candidate area is input into a deep convolutional neural network with three parallel feature extraction branches for color, contour and brightness. A three-dimensional mutual reward function is embedded in the frost physical model based on the principle of heat and mass transfer. The frost coverage rate is calculated by ratio calculation.
[0076] The frost candidate region is input into a pre-constructed deep convolutional neural network, and three parallel feature extraction branches are configured, including a color dimension branch, a contour dimension branch, and a brightness dimension branch.
[0077] Based on the principles of heat and mass transfer, a physical model for frosting is pre-defined, and the coupling relationships of each dimension under the ideal frosting state are defined, including positive correlation constraints and consistency constraints.
[0078] The positive correlation constraint is that as the thickness of the frost layer (i.e., the brightness feature gradient) increases, the complexity of its surface texture (i.e., the contour feature) should increase synchronously; the consistency constraint is that the increase in the whiteness of the color component should have a linear logical relationship with the expansion trend of the contour.
[0079] A three-dimensional mutual reward function is additionally embedded into the training loss function of the deep convolutional neural network to calculate the matching degree between the feature extraction branch and the physical law model, and to perform dynamic rewards, specifically including:
[0080] When the color whiteness, contour geometric expansion and brightness gradient depth are simultaneously enhanced and conform to the preset physical evolution model, the activation factor is activated, the combination is given the highest feature weight, and the result is fed back to the neuron activation function.
[0081] If a feature in a certain dimension undergoes a sudden change (such as a flashing light causing a sudden increase in brightness, but the color and outline remain unchanged), the reward function outputs zero and suppresses such non-frost interference.
[0082] Based on the combination of feature extraction branches with high reward weights, the edge reconstruction of the frost candidate region is performed, and the physical edge of the frost region is locked by feature map deconvolution to generate a high-confidence frost contour image.
[0083] Based on the high-confidence frosting contour image, the following operation is performed:
[0084] Calculate the total area of pixels with a pixel value of 1 in the mask image;
[0085] The ratio of the total pixel area to the total pixel area of the heat exchange area of the air cooler is calculated to obtain the frost coverage rate, which is then transmitted to the decision optimization module in real time as a status parameter.
[0086] It should also be noted that the dimensional reward function mechanism achieves phase transition identification of the frost layer evolving from a "loose state" to an "ice crystal state" by weighting the ideal dimensional combination, and adjusts the energy efficiency evaluation weights according to the different effects of different phase states on thermal resistance.
[0087] Correlation analysis module: Based on the frost coverage rate, establish the fin equivalent thermal resistance model, calculate the deviation of the actual refrigeration system energy efficiency ratio from the theoretical frost-free state, and calculate the influence coefficient of frost thickness on heat exchange efficiency in real time based on the multi-dimensional features output by the frost identification module.
[0088] The correlation analysis module collects real-time pressure, temperature and power consumption parameters, establishes an equivalent thermal resistance model for the fins based on the frost coverage rate, and calculates the deviation between the actual energy efficiency ratio of the refrigeration system under the current state and the theoretical energy efficiency ratio of the frost-free refrigeration system.
[0089] The correlation analysis module monitors the offset between evaporation pressure and fan current, and calculates the influence coefficient of the current frost layer thickness on the heat exchange efficiency in real time based on the multidimensional features output by the frost layer identification module.
[0090] Decision optimization module: The energy efficiency scheduling algorithm based on the benefit function takes minimizing the total operating energy consumption of the system as the objective function, dynamically calculates and outputs the best defrosting intervention command;
[0091] The decision optimization module adopts an energy efficiency scheduling algorithm based on a benefit function. When the expected improvement in cooling efficiency after defrosting is greater than the sum of the energy consumption cost of the defrosting process itself and the heat loss during shutdown, it is determined to be the optimal defrosting time.
[0092] The reward function is based on the reward mechanism of reinforcement learning, and its expression includes:
[0093]
[0094] Where R represents the energy savings achieved after performing the defrosting action. This indicates the predicted power consumption under the current frosting condition. This indicates the baseline power consumption after defrosting. This indicates the instantaneous energy consumption of the defrosting operation. This indicates the intrusion of defrosting heat load during the shutdown period;
[0095] It should also be noted that the decision optimization module introduces a power load-side response mechanism, and its decision threshold is dynamically adjusted according to the peak and valley periods of external electricity prices and the priority of snowmaking demand, so as to achieve synergistic optimization of economy and low carbon.
[0096] It should also be noted that a closed-loop calibration is established. After each defrosting operation, the actual energy efficiency recovery data is compared with the model's predicted benefits, and the dimensional reward weights of the neural network are corrected through the backpropagation algorithm.
[0097] Example 2
[0098] The following is another embodiment of the present invention, which provides a low-carbon refrigeration and snowmaking automated intelligent control system. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.
[0099] The baseline COP of the existing snowmaking refrigeration system is 3.5, the cooling requirement is 50kW, and the frost growth rate is 0.01 / hour (based on a typical low-temperature and high-humidity environment). The model for the decrease in energy efficiency ratio due to frost coverage is COP = 3.5 × (1 - 0.5 × coverage rate). Each defrosting operation consumes 5kWh of energy and has a heat loss of 3kWh. The total simulation duration is 720 hours (approximately 30 days).
[0100] Traditional timed defrosting method: defrost once every 48 hours, regardless of the state of the frost layer.
[0101] Total energy consumption: 11915.48 kWh;
[0102] Average energy efficiency ratio: 3.02;
[0103] Number of defrost cycles: 15;
[0104] Analysis: Due to the excessively long defrosting interval, the average frost coverage is high (the peak value is close to 1), resulting in prolonged operation in a low energy efficiency ratio state and overall low energy efficiency.
[0105] This invention employs an acquisition and processing module to acquire images in real time, a frost layer separation and identification module to calculate coverage, a correlation analysis module to assess energy efficiency degradation, and a decision optimization module to determine the timing of defrosting based on a benefit function.
[0106] Minimum defrost interval: 8 hours; coverage threshold: 0.2; predicted field of view: 24 hours.
[0107] Total energy consumption: 11153.70 kWh
[0108] Average energy efficiency ratio: 3.23
[0109] Number of defrost cycles: 36
[0110] Analysis: The system accurately identifies the frost layer status and intervenes in defrosting only when the expected benefits outweigh the costs, avoiding unnecessary downtime and preventing excessive frost buildup. Compared to traditional methods, while the number of defrosting operations increases, it is more optimized, resulting in lower average coverage and improved COP.
[0111] Comparison results:
[0112] Energy savings: (11915.48 - 11153.70) / 11915.48 × 100% = 6.39%
[0113] Energy efficiency ratio improvement: (3.23-3.02) / 3.02×100%≈6.95%
[0114] Carbon emission reduction estimate (assuming an electricity carbon factor of 0.5 kg CO2 / kWh): Energy savings of 761.78 kWh and a reduction in carbon emissions of approximately 380.89 kg.
[0115] Experimental results show that the present invention significantly reduces total operating energy consumption and improves the system energy efficiency ratio by optimizing decision-making through intelligent image recognition and reinforcement learning, thus verifying its superiority in low-carbon refrigeration and snowmaking applications.
[0116] By employing a mobile industrial camera combined with an automated guided rail for full-coverage scanning, this method replaces traditional point sensors to achieve comprehensive and high-spatial-resolution visual perception of the frost distribution on the surface of air cooler array fins. Compared to traditional methods that are susceptible to local unevenness, sensor drift, or installation location limitations, this method can capture comprehensive information on frost thickness, density, coverage uniformity, and phase evolution, avoiding false positives or false negatives and improving the reliability and accuracy of frost detection.
[0117] By combining HSV color space conversion with adaptive Otsu method threshold calibration and brightness correction coefficient binarization separation strategy, the interference of light changes, shadows and fin metal reflection on frost layer segmentation is effectively decoupled.
[0118] Adaptive precision adjustment of morphological opening / closing operations further eliminates noise, voids, and artifacts, ensuring high-quality and consistent frost masks.
[0119] By incorporating a three-dimensional mutual reward function mechanism based on physical constraints of heat and mass transfer into the loss function of a deep convolutional neural network, and through the positive correlation and consistency constraints of multi-dimensional features such as color whiteness, contour complexity, and brightness gradient, the model can accurately identify the phase transition of frost layer from "loose state to ice crystal state" and effectively suppress non-frost interference such as light flicker and local stains. This can significantly improve the model's generalization ability and confidence in complex industrial environments.
[0120] By combining the frost coverage rate with real-time collected pressure, temperature, and power consumption parameters, a dynamic model of the fin's equivalent thermal resistance is established, and the influence coefficient of frost thickness on heat exchange efficiency is calculated. This allows for real-time quantification of the energy efficiency ratio reduction and energy consumption increase caused by frost.
[0121] The energy efficiency scheduling model, built upon reinforcement learning, aims to minimize total operating energy consumption. It incorporates a reward function that comprehensively considers the efficiency improvement benefits after defrosting, the instantaneous energy consumption during defrosting, and the heat loss during shutdown. Simultaneously, it embeds a power load-side response mechanism to dynamically adjust the decision threshold based on external electricity price peak and valley periods and snowmaking demand priorities, achieving "peak-shifting and valley-filling" operation.
[0122] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A low-carbon refrigeration snowmaking automated intelligent management and control system, characterized in that, Specifically comprising: The acquisition processing module, the frost layer separation module, the frost layer identification module, the correlation analysis module and the decision optimization module; The acquisition processing module: acquiring image data of the surface of the air cooler through a mobile industrial camera, and performing preprocessing on the image data; The frost layer separation module: converting the preprocessed image data to HSV color space, performing adaptive threshold calibration based on the component difference of HSV space, and preliminarily separating the frost layer region through binarization processing; The frost layer identification module: inputting the frost layer candidate area into a deep convolutional neural network configured with three parallel feature extraction branches of color, contour and brightness, embedding a three-dimensional mutual reward function based on a pre-set frost physical model based on the principle of heat and mass transfer, and calculating the frost coverage rate through ratio operation; The correlation analysis module: establishing a fin equivalent thermal resistance model according to the frost layer coverage rate, calculating the deviation of the actual refrigeration system energy efficiency ratio relative to the theoretical frost-free state, and based on the multi-dimensional features output by the frost layer identification module, real-time solving the influence coefficient of the frost layer thickness on the heat exchange efficiency; The decision optimization module: based on the energy efficiency scheduling algorithm of the revenue function, taking the minimization of the total running energy consumption of the system as the objective function, dynamically calculating and outputting the best defrosting intervention instruction.
2. The low-carbon refrigeration and snowmaking automated intelligent management and control system according to claim 1, characterized in that: In the frost layer separation module, based on the characteristics of the HSV color space being insensitive to light changes, the frost layer decouples the background of the air cooler surface to obtain three independent channels of hue, saturation and brightness; Extracting the gray scale distribution and brightness distribution of the image data; Based on the two distributions, the saturation matrix channel is histogram counted to identify the significant wave peaks of the air cooler surface and the frost layer, and the component value of the saturation that maximizes the inter-class variance is calculated as the initial segmentation threshold value by the Otsu method. A dynamic correction coefficient is introduced based on the average brightness value of the brightness channel to compensate for the initial segmentation threshold value, and a final decision threshold is generated. If a pixel satisfies the component value of saturation less than the final decision threshold and its brightness is within a pre-set reasonable range, it is marked as a frost layer candidate area, otherwise it is marked as an air cooler background area, and a preliminary binary mask is generated.
3. The low-carbon refrigeration and snowmaking automated intelligent management and control system according to claim 1, characterized in that: In the frost layer identification module, the frost layer candidate area is input into a pre-constructed deep convolutional neural network, and three parallel feature extraction branches are configured, including color dimension branch, contour dimension branch and brightness dimension branch; Based on the principle of heat and mass transfer, a pre-set frost physical model is defined, and the coupling relationship of each dimension under ideal frost conditions is defined; A three-dimensional mutual reward function is additionally embedded in the training loss function of the deep convolutional neural network to calculate the matching degree of the feature extraction branch and the physical law model, and dynamic rewards are executed; Based on the combination of feature extraction branches with high reward weight, the edge of the frost layer candidate area is reconstructed, the physical edge of the frost area is locked through feature map deconvolution, and a high-confidence frost contour image is generated.
4. The low-carbon refrigeration and snowmaking automated intelligent management and control system according to claim Based on the high-confidence frost profile image, the following operations are performed: Statistical total pixel area of pixel value 1 in the mask image; The total pixel area is compared with the total pixel area of the cold air fan heat exchange area, and the frost coverage rate is calculated and transmitted to the decision optimization module as a state parameter.
5. The low-carbon refrigeration snowmaking automatic intelligent management and control system according to claim 1, characterized in that: In the correlation analysis module, by collecting real-time pressure, temperature and power consumption parameters, an equivalent thermal resistance model of the fin is established based on the frost coverage rate, and the deviation of the actual refrigeration system energy efficiency ratio under the current state from the theoretical frost-free refrigeration system energy efficiency ratio is calculated; By monitoring the deviation of the evaporation pressure and the fan current, based on the multi-dimensional features output by the frost layer identification module, the influence coefficient of the current frost layer thickness on the heat exchange efficiency is calculated in real time.
6. The low-carbon refrigeration snowmaking automatic intelligent management and control system according to claim 1, characterized in that: The decision optimization module adopts an energy efficiency scheduling algorithm based on a benefit function. When the expected refrigeration efficiency improvement benefit after defrosting is greater than the sum of the energy consumption cost and the shutdown heat loss of the defrosting process itself, it is determined as the optimal defrosting opportunity.
7. The low-carbon refrigeration snowmaking automatic intelligent management and control system according to claim 6, characterized in that: The benefit function combines the reward mechanism of reinforcement learning, and its expression includes: wherein R represents the energy saving margin obtained after the defrosting operation, represents the predicted power consumption in the current frosting state, represents the reference power consumption after defrosting, represents the instantaneous energy consumption of the defrosting operation, represents the defrost heat load intrusion during shutdown.
8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to realize the modules of the intelligent roof solar tracking optimization control system based on climate zone self-learning according to any one of claims 1 to 7.
9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the modules of the low-carbon refrigeration snowmaking automatic intelligent management and control system according to any one of claims 1 to 7.