Industrial air conditioning energy saving method based on neural network

Abstract

The present invention discloses an industrial air-conditioning energy-saving method based on a neural network, which relates to the technical field of industrial air-conditioning energy-saving, and specifically includes the following steps: when there is a failure of the cold and hot channels in the industrial air conditioner, locate the area where the cold and hot channel failure occurs, mark this area as the target area, and real-time obtain the airflow characteristic data of the target area; comprehensively analyze the real-time obtained airflow characteristic data of the target area in combination with the neural network model trained based on historical samples to evaluate the degree of airflow conflict in the target area; based on the evaluation result, execute the corresponding energy-saving regulation strategy and dynamically adjust the cold air flow direction. The present invention solves the problem that in the case of cold and hot channel failure conditions, the prior art cannot dynamically identify abnormal areas according to the degree of local airflow conflict and accurately regulate the cold air flow direction, realizes intelligent identification, adaptive evaluation and hierarchical response in the industrial air-conditioning energy-saving control process, and effectively improves the energy-saving efficiency and system operation stability.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial air - conditioning energy conservation, and particularly to an industrial air - conditioning energy - conservation method based on a neural network. Background Art

[0002] Industrial air conditioners are a type of large - scale air - conditioning equipment specifically used in industrial scenarios. They usually have high refrigeration capacity, forced ventilation ability, and good environmental adaptability, and are widely used in places with strict requirements for temperature, humidity, and air quality, such as computer rooms, factories, laboratories, and clean workshops. Compared with civilian air conditioners, industrial air conditioners have higher requirements in terms of operating intensity, power scale, usage duration, and operating continuity. Therefore, their energy consumption also increases significantly. Under the traditional control mode, industrial air conditioners often operate with preset parameters and static logic, making it difficult to perceive the dynamic changes of the environment and complex load characteristics in real - time, and prone to problems such as over - refrigeration or low energy efficiency. To solve this problem, introducing a neural network for energy - conservation control has become an effective means. Neural networks have powerful non - linear modeling and data - learning capabilities. After obtaining a large amount of historical operation data and environmental information, they can establish a prediction model for temperature and humidity change trends and load fluctuations, thereby accurately judging the optimal operating state. Based on this, industrial air conditioners can dynamically adjust key parameters such as compressor start - stop, air - supply mode, and refrigerant flow rate to achieve the goal of cooling on demand and intelligent energy control. This intelligent energy - conservation method not only significantly improves the energy utilization efficiency of air conditioners but also reduces the equipment operation cost, providing a reliable technical path for green and low - carbon operation in industrial scenarios.

[0003] Existing industrial air - conditioning energy - conservation technologies based on neural networks dynamically optimize the operating state of air - conditioning equipment through deep - learning models to achieve energy - conservation effects. Specifically, these technologies first deploy sensors in air - conditioning equipment and its environment to collect multi - dimensional data such as temperature, humidity, air flow velocity, and load changes in real - time. At the same time, it also includes operating parameters of the equipment such as the working state of the compressor, the pressures of the condenser and evaporator. Then, the neural network learns from this historical data to establish a non - linear relationship model between the environment and equipment operation, predicting load changes and energy - consumption demands in the next period of time. Based on these predictions, the neural - network algorithm can intelligently adjust the operating mode of air - conditioning equipment, such as the start - stop timing of the compressor, the speed adjustment of the fan, and the optimization of refrigerant flow rate, to adapt to different environmental and load demands. In addition, the neural network can continuously optimize the decision - making process through technologies such as reinforcement learning, and adjust the control strategy in real - time, so that the air conditioner always maintains the best energy - efficiency ratio in a changing working environment. In this process, the neural network can not only reduce over - refrigeration in a data - driven manner but also effectively avoid ineffective energy consumption, making the operation of industrial air conditioners more refined and intelligent, and ultimately achieving the goal of significantly reducing energy consumption and improving operation efficiency.

[0004] The prior art has the following deficiencies:

[0005] In some industrial air conditioning systems operating at high loads, a hot and cold aisle design is usually adopted to achieve efficient cold air distribution and hot air discharge. However, when the equipment layout changes or the number of equipment increases, the effective isolation of the hot and cold aisles may be damaged, resulting in cold air flowing into non-target areas, or hot air being misdirected into the cold air flow channels, thereby rendering the hot and cold aisles ineffective. Due to this ineffectiveness, the cold air fails to accurately reach the areas that need to be cooled, resulting in the cold load not being fully consumed, while some areas may be over-cooled or have uneven temperatures due to abnormal cold air flow. Existing neural network-based industrial air conditioning energy-saving technologies cannot dynamically adjust the cold air flow direction according to the degree of air flow conflict in the areas where the hot and cold aisle failures occur. Since the failure of the hot and cold aisles does not directly cause an overall temperature change, neural networks usually rely on global temperature and humidity data for regulation, but fail to timely identify the energy efficiency losses caused by local air flow conflicts and design failures. Therefore, the air conditioning system will still continue to operate at full load, resulting in cold air waste, decreased energy efficiency, extended system operation time, and increased energy consumption. Maintaining this failure state for a long time will not only exacerbate energy waste, but may also lead to an increased burden on the air conditioning equipment and an increase in system failure rates, ultimately affecting the overall performance and energy-saving goals of the air conditioning system.

[0006] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0007] The object of the present invention is to provide a neural network-based industrial air conditioning energy-saving method to solve the problems in the above background art.

[0008] To achieve the above object, the present invention provides the following technical solution: A neural network-based industrial air conditioning energy-saving method, specifically including the following steps:

[0009] Real-time obtain the air state information of the hot and cold aisle areas inside the industrial air conditioner through a sensor network arranged at the air supply outlets, air return outlets, and hot and cold aisle areas of the industrial air conditioner, and analyze it to determine whether there is a failure of the hot and cold aisles in the industrial air conditioner;

[0010] When there is a failure of the hot and cold aisles in the industrial air conditioner, locate the area where the hot and cold aisle failure occurs, mark this area as the target area, and real-time obtain the air flow characteristic data of the target area;

[0011] Comprehensively analyze the real-time obtained air flow characteristic data of the target area in combination with a neural network model trained based on historical samples to evaluate the degree of air flow conflict in the target area;

[0012] Based on the evaluation results, execute the corresponding energy-saving regulation strategies and dynamically adjust the cold air flow direction;

[0013] Input the environmental response data after the execution of the energy-saving regulation strategies into the neural network model, and enable the neural network model to continuously improve its recognition ability and regulation ability through continuous learning during operation.

[0014] Preferably, when there is a failure of the cold and hot channels in the industrial air conditioner, locate the area where the cold and hot channel failure occurs, specifically:

[0015] When there is a failure of the cold and hot channels in the industrial air conditioner, evenly divide the cold air channel into several sub-regions, and each sub-region uses a sensor network to continuously monitor the wind speed, wind direction, temperature, and humidity data in real time; for each sub-region, when the wind speed in the sub-region is lower than the preset threshold of the normal wind speed, and the included angle between the wind direction in the sub-region and the preset wind direction exceeds the set maximum deviation angle, and the temperature change rate and humidity change rate in the sub-region both exceed their respective set normal fluctuation ranges, determine that the sub-region is the area where the cold and hot channel failure occurs.

[0016] Preferably, comprehensively analyze the airflow characteristic data of the target area obtained in real time in combination with the neural network model trained based on historical samples to evaluate the degree of airflow conflict in the target area, specifically including the following steps:

[0017] Preprocess the airflow characteristic data of the target area obtained in real time, and extract the cold air directivity characteristic data and local disturbance response data from it after preprocessing;

[0018] Respectively combine the extracted cold air directivity characteristic data and local disturbance response data with the neural network model trained based on historical samples for comprehensive analysis, and generate the cold air offset index and the airflow interference coefficient respectively;

[0019] Construct an airflow conflict evaluation model, input the generated cold air offset index and airflow interference coefficient into the model, and generate an airflow conflict index through weighted summation;

[0020] Determine the preset airflow conflict index threshold interval, and compare it with the generated airflow conflict index after determination, and evaluate the degree of airflow conflict in the target area according to the comparison result.

[0021] Preferably, the training process of the neural network model trained based on historical samples includes the following steps:

[0022] Collect historical samples, where the historical samples include the cold air directivity feature data, local disturbance response data, and the corresponding air flow conflict degree level labels in different operating states of the target area within a historical time period; perform format standardization and numerical normalization processing on the historical samples, and divide them into a training set, a validation set, and a test set according to a set ratio; construct a multi-layer neural network model including an input layer, a hidden layer, and an output layer, where the input layer receives the feature data vector and the output layer generates the adjustment factor result; use the supervised learning method to continuously update the network weights based on the error backpropagation algorithm by minimizing the error between the conflict evaluation index generated after the adjustment factor and the true level label until the training error meets the preset convergence condition, and complete the training of the neural network model; solidify and save the trained model parameter structure for subsequent generation and evaluation analysis of the adjustment factor for the real-time extracted feature data.

[0023] Preferably, the acquisition logic of the cold air offset index is as follows:

[0024] Extract the cold air directivity feature data from the preprocessed air flow feature data of the target area, specifically including the angle between the cold air wind direction and the preset reference wind direction at different moments within a period of time, the average cold air wind speed, and the cold air reference wind speed under normal operating conditions, and calibrate them as 、 and respectively, represents the angle between the cold air wind direction and the preset reference wind direction at the target area at the moment within a period of time, represents the average cold air wind speed at the target area at the moment within a period of time, represents the cold air reference wind speed under normal operating conditions, , is a positive integer;

[0025] Determine the values of the adjustment factors and based on the neural network model trained with historical samples. The adjustment factor is used to adjust the influence weight of the wind direction offset term in the cold air offset index, and the adjustment factor is used to adjust the influence weight of the wind speed deviation term in the cold air offset index;

[0026] Calculate the cold air offset index, and the specific calculation formula is as follows:

[0027] In the formula, is the cold air offset index.

[0028] Preferably, the acquisition logic of the air flow interference coefficient is as follows:

[0029] Extract local disturbance response data from the airflow characteristic data of the preprocessed target area, specifically including the temperature change rate of the target area at different times within a period of time, the air pressure difference between the air supply inlet and outlet, and the static pressure value, the average cold air velocity, and the standard deviation of the cold air velocity at the cold air outlet of the target area within this period of time, and calibrate them respectively as , , , and , represents the temperature change rate of the target area at the moment of within a period of time, represents the air pressure difference between the air supply inlet and outlet of the target area at the moment of within a period of time, represents the static pressure value at the cold air outlet of the target area within this period of time, represents the average cold air velocity of the target area within this period of time, represents the standard deviation of the cold air velocity of the target area within this period of time, , is a positive integer;

[0030] Determine the values of the adjustment factors , and based on the neural network model trained with historical samples. The adjustment factor is used to adjust the weight of the temperature disturbance term in the overall interference evaluation. The adjustment factor is used to adjust the weight of the wind speed fluctuation term in the overall interference evaluation. The adjustment factor is used to adjust the weight of the pressure difference term in the overall interference evaluation;

[0031] Calculate the airflow interference coefficient. The specific calculation formula is as follows:

[0032] In the formula, is the airflow interference coefficient.

[0033] Preferably, construct an airflow conflict evaluation model, and input the generated cold air offset index and the airflow interference coefficient into this model, and generate an airflow conflict index through weighted summation. The specific calculation formula is as follows:

[0034] In the formula, is the airflow conflict index, and are respectively the cold air offset index and the airflow interference coefficient non-zero weight coefficients, and .

[0035] Preferably, a preset air flow conflict index threshold range is determined, and after determination, it is compared with the generated air flow conflict index to evaluate the air flow conflict degree of the target area according to the comparison result. The specific comparison and analysis are as follows:

[0036] If , the air flow conflict degree of the target area is low;

[0037] If , the air flow conflict degree of the target area is medium;

[0038] If , the air flow conflict degree of the target area is severe.

[0039] Preferably, based on the evaluation result, the corresponding energy-saving regulation strategy is executed, and the cold air flow direction is dynamically adjusted. Specifically:

[0040] When the evaluation result is low, the energy-saving regulation strategies executed include: keeping the current settings of the air supply wind speed, air supply direction, and output power in the target area unchanged, and continuously collecting air state information at preset time intervals for verification; the corresponding dynamic adjustment method of the cold air flow direction is: maintaining the air supply direction parameter at the initial setting without performing direction correction operations;

[0041] When the evaluation result is medium, the energy-saving regulation strategies executed include: modifying the air supply direction setting of the target area and adjusting the air supply wind speed to concentrate the cold air flow towards the air flow conflict risk area; the corresponding dynamic adjustment method of the cold air flow direction is: updating the guiding parameter of the air supply unit, adjusting the cold air path to focus on the target cold load area, and keeping the air supply direction stable after adjustment;

[0042] When the evaluation result is severe, the energy-saving regulation strategies executed include: terminating the air supply operation in the non-target area, centrally adjusting the air supply pressure difference and direction setting in the target area to improve the stable delivery ability of the cold air; the corresponding dynamic adjustment method of the cold air flow direction is: closing the air supply path of the non-target channel, and jointly adjusting the air supply direction and wind speed setting to make the cold air focus on the air flow conflict area and improve the air flow organization efficiency.

[0043] In the above technical solution, the technical effects and advantages provided by the present invention:

[0044] 1. By collecting and extracting multi-dimensional air state information such as wind speed, wind direction, temperature, and humidity in local areas of the cold and hot channels of industrial air conditioners in real time, the present invention successfully overcomes the problem that traditional regulation based on global temperature and humidity data cannot accurately identify local air flow conflicts. By dividing the cold air channel into multiple sub-regions and making sub-region judgments by combining refined parameters such as angle deviation, wind speed deviation, and temperature and humidity change rate, the present invention can accurately identify the failure areas of the cold and hot channels caused by factors such as equipment layout changes and air flow field instability, providing a reliable identification basis for subsequent energy-saving control and effectively solving the problem of energy efficiency loss caused by the inability of the prior art to timely detect local failures of the cold and hot channels.

[0045] 2. The present invention innovatively constructs a dual-index evaluation system with the cold air offset index and the air flow interference coefficient as the core. By using a neural network model to learn historical samples to generate adjustment factors, it realizes the weighted analysis of factors such as wind direction deviation, wind speed fluctuation, temperature disturbance, and pressure difference abnormality, and further generates an air flow conflict index through a weighted summation model, enabling the air flow conflict degree in the target area to have a quantifiable, comparable, and evaluable calculation basis. This design significantly enhances the sensitivity and judgment accuracy of conflict identification, breaks through the fuzzy identification method of traditional manual threshold setting or empirical judgment, and realizes the intelligent evaluation ability of high precision and scene adaptability.

[0046] 3. On the basis of identification and evaluation, the present invention further constructs a hierarchical response regulation mechanism, which can automatically execute different cold air flow adjustment strategies according to the air flow conflict level (low, medium, severe), realizing a closed-loop energy-saving control process from "identification → evaluation → decision → execution". At the same time, the environmental response data after introducing the energy-saving regulation strategy is used as feedback to input into the neural network model, and the model parameters are continuously optimized through an online continuous learning mechanism, enabling the system to have the ability of model self-update and control strategy adaptability during long-term operation, effectively coping with non-linear disturbance problems such as environmental changes, load drift, and structural transformation, thereby greatly improving the energy-saving efficiency, stability, and intelligent level of industrial air conditioner operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a flow schematic diagram of the industrial air conditioner energy-saving method based on neural network of the present invention.

[0049] Figure 2 It is a method mind map of the industrial air conditioner energy-saving method based on neural network of the present invention Detailed Implementation Modes

[0050] Example implementation modes will now be described more fully with reference to the accompanying drawings. However, the example implementation modes can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these example implementation modes are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the example implementation modes to those skilled in the art.

[0051] The present invention provides an industrial air conditioner energy-saving method based on a neural network as shown in Figure 1 and Figure 2 which specifically includes the following steps:

[0052] Real-time obtain the air state information in the hot and cold channel area inside the industrial air conditioner through a sensor network arranged at the air supply outlet, air return outlet and hot and cold channel area of the industrial air conditioner, and analyze it to determine whether there is a situation of hot and cold channel failure in the industrial air conditioner;

[0053] A multi-type environmental sensor can be arranged at the key airflow path nodes inside the industrial air conditioner to form a high-density sensor network covering the air supply outlet, air return outlet, cold air channel and hot air channel, so as to realize the real-time acquisition of the air state. This sensor network is usually composed of temperature sensors, humidity sensors, wind speed sensors and wind direction sensors. Among them, the temperature and humidity sensors are used to monitor the heat and moisture exchange of local cold and warm air, and the wind speed and wind direction sensors are used to collect key aerodynamic parameters such as wind flow speed and air flow direction deviation. The data of all sensors are uploaded to the central processing unit or edge computing device in real time through a wireless or wired communication module (such as Modbus, RS485 or Wi-Fi Internet of Things gateway). In the software system, there is a sensor data acquisition module, which synchronously receives the air state information uploaded by each node at a set time interval and stores it uniformly in the time series database for subsequent airflow behavior analysis. The air state information includes: local temperature, relative humidity, wind speed magnitude, wind direction angle, wind speed gradient change, etc., which are used to reflect the airflow operation state of the hot and cold channels.

[0054] After obtaining the real-time air status information, the software system constructs a multi-parameter benchmark model for the normal operation of the cold and hot channels, and uses feature comparison and anomaly detection algorithms to determine whether there is a failure of the cold and hot channels at present. Specifically, the software compares the currently collected wind speed, wind direction, temperature and humidity data with the preset reference intervals to determine whether there are abnormal features such as the deviation of the cold air flow direction, the rapid attenuation of the wind speed, the countercurrent of the wind direction, and the enhancement of local hot gas disturbance. For example, when the wind speed of a certain cold channel drops significantly and the temperature of the target area does not decrease significantly, the system can judge that the cold air has not reached the target area accordingly; another example is that when the included angle between the inlet direction of the cold air flow and the preset flow direction exceeds the set threshold (such as 30°), it is judged that there is a cold air deviation phenomenon. In addition, through the analysis of the multi-point temperature difference change, if the temperature and humidity of a certain area fluctuate abnormally while the overall system temperature remains stable, it can also be inferred that there is a mixture of cold and hot air flows in this area. This judgment process is completely executed by the software, and the "failure / normal" judgment can be carried out by using conditional rule triggering, time series analysis based on a sliding window or a simple neural network binary classification model, and finally output whether there is a failure in this state and mark the failure area.

[0055] The fundamental purpose of realizing the automatic judgment of whether the cold and hot channels of industrial air conditioners fail is to solve the problem of energy consumption waste caused by the traditional energy-saving control relying on the overall environmental parameters and being unable to identify local anomalies. In actual operation, the failure of the cold and hot channels often manifests as the ineffective coverage of the target area by the cold air flow, or the misentry of the hot air flow into the cold air channel. However, since these phenomena occur locally, their impacts may not be immediately reflected in the overall temperature and humidity data, resulting in the failure of the traditional regulation strategy based on the "global average temperature". Only by constructing a sensor network to monitor the status of each area in real time and performing dynamic analysis in software can we accurately capture whether the cold and hot channels fail. This judgment mechanism not only improves the sensitivity of the cold and hot air flow regulation, but also provides a prerequisite basis for the subsequent evaluation of the degree of air flow conflict and the selection of energy-saving strategies based on neural networks. If this step is not carried out, the system will not be able to identify problems such as the mixture of cold and hot air and the deviation of cold air in which areas, and thus misjudge the global operation status as "normal", causing the air conditioner system to continue to operate at full load, resulting in the waste of cold air and ineffective cooling of the target area, ultimately affecting the energy efficiency ratio and the service life of the equipment. Therefore, this judgment is the key first link to achieve refined energy-saving control.

[0056] When there is a failure of the cold and hot channels in the industrial air conditioner, locate the area where the cold and hot channels fail, mark this area as the target area, and obtain the air flow characteristic data of the target area in real time;

[0057] In this embodiment, when there is a failure of the cold and hot channels in the industrial air conditioner, the area where the cold and hot channels fail is located specifically as follows:

[0058] When there is a failure in the cold and hot channels of an industrial air conditioner, the cold air channel is evenly divided into several sub-regions, and each sub-region uses a sensor network to monitor the wind speed, wind direction, temperature, and humidity data in real time. For each sub-region, when the wind speed in the sub-region is lower than the preset threshold of the normal wind speed, and the angle between the wind direction in the sub-region and the preset wind direction exceeds the set maximum deviation angle, and the temperature change rate and humidity change rate in the sub-region both exceed their respective set normal fluctuation ranges, it is determined that the sub-region is the area where the cold and hot channel failure occurs.

[0059] When there is a failure in the cold and hot channels of an industrial air conditioner, to achieve fine perception and dynamic evaluation of the cold air flow state, the cold air channel can be evenly divided into several sub-regions in the spatial dimension through software. The boundary of each sub-region is generated by a preset spatial grid division algorithm, and a unique number and coordinate range are assigned to each sub-region in the database. Subsequently, according to the position relationship of the sub-regions, the wind speed sensors, wind direction sensors, temperature sensors, and humidity sensors arranged at the air supply path, return air path, and the junction of the cold and hot channels are bound to the corresponding sub-regions. During operation, the software acquisition module calls the sensor interfaces corresponding to each sub-region according to the set time period to obtain the real-time wind speed, wind direction, temperature, and humidity data in each sub-region, and synchronously records them in the time series data buffer for subsequent dynamic judgment and model evaluation. Through this structured spatial division and binding acquisition mechanism, not only can the "global average" misjudgment of the air flow state be avoided, but also the cold and hot channel operation state can have a high-resolution perception ability, providing a basic support for accurately identifying the cold and hot channel failure areas and implementing local energy-saving control strategies. This method completely completes the area division, data acquisition task assignment, and real-time monitoring process organization through software logic, and has good portability and scalability, and is applicable to multi-region cold air control scenarios in complex industrial environments.

[0060] In each sub-region, the judgment of hot and cold channel failure can be achieved through a multi-condition joint judgment algorithm built into the software system. This algorithm performs dynamic analysis and feature extraction on various physical quantities based on the original data of wind speed, wind direction, temperature, and humidity collected by sensors in real time. The judgment method for wind speed is as follows: Compare the real-time wind speed value at the current moment t with the average wind speed recorded in the normal operating state of this sub-region. If the current wind speed is lower than the set proportional threshold of this historical average (e.g., 80%), it is regarded as an abnormal wind speed. The judgment method for wind direction deviation is as follows: Calculate the included angle between the current wind direction vector and the preset wind direction vector (based on the vector cosine formula). If the included angle exceeds the maximum allowable deviation angle (e.g., 30°), it is judged as an abnormal wind direction. The temperature change rate and humidity change rate are based on the differential calculation of consecutive time slices: By calculating the difference between the temperature value and humidity value between the current moment and the previous sampling moment, and dividing by the time interval, the temperature and humidity change rates are obtained. If the absolute value exceeds their respective set normal fluctuation thresholds (such as 0.5 °C / minute, 2%RH / minute), it is regarded as there being abnormal thermal and humidity disturbances in this sub-region. When the above four conditions are simultaneously met, the system determines that this sub-region is the area where hot and cold channel failure occurs. This judgment method is fully implemented by software, does not rely on hardware control logic, has high real-time performance and scalability, and by introducing a multi-variable cross-validation mechanism, effectively reduces the misjudgment caused by the fluctuation of a single index, improves the accuracy of identifying local hot and cold channel failures, and thus provides a reliable data basis for subsequent intelligent energy-saving control.

[0061] After a certain sub-region is determined to be the area where hot and cold channel failure occurs, the system will immediately trigger the area calibration logic to achieve targeted management of this area. The specific implementation method is as follows: After the software system completes the failure identification, it writes the unique number of this sub-region in the area division index, the corresponding spatial coordinate range, and its judgment status into the marked data table in a structured record manner, and generates a target status label for this area in the central data structure (such as the "target area list"). This label will bind all subsequent data flow entrances of this area to ensure that the neural network model only processes the feature data from valid target areas and avoids interference from non-critical areas to the decision-making. The reason for explicitly calibrating the failure area as a target area is that the abnormal phenomena in the cold air channel usually have spatial locality and dynamics. Only through explicit marking can targeted operations be achieved in subsequent data collection, analysis, neural network inference, etc., improving the utilization efficiency of algorithm resources, reducing redundant calculations, and achieving precise triggering of energy-saving control strategies.

[0062] After completing the calibration of the target area, the data acquisition module of the system will call the bound sensor data interface based on the area number and spatial coordinates, and continuously obtain key airflow parameters such as wind speed, wind direction, temperature, and humidity at a set time interval. The specific implementation method is as follows: The system establishes a "target area data stream mapping table" to bind the target area with its corresponding sensor nodes one by one, and preferentially obtains the latest data of the corresponding sensors from the mapping table in each round of data polling. The acquisition results will be parsed into a structured data format (such as key-value pairs or multi-dimensional vectors) and written into the feature data buffer in sequence according to the time stamp for subsequent calls by the neural network model. To ensure the timeliness and accuracy of the data, the system can also set up a data integrity check mechanism, such as packet loss monitoring and outlier filtering. Through the above methods, the system can accurately and continuously obtain the real-time airflow feature data of the target area without interfering with the overall operation of the air conditioner, providing dynamic and high-quality input support for subsequent airflow conflict assessment and energy-saving control strategies.

[0063] Combine the airflow feature data of the target area obtained in real time with the neural network model trained based on historical samples for comprehensive analysis to evaluate the degree of airflow conflict in the target area;

[0064] In this embodiment, combining the airflow feature data of the target area obtained in real time with the neural network model trained based on historical samples for comprehensive analysis to evaluate the degree of airflow conflict in the target area specifically includes the following steps:

[0065] Preprocess the airflow feature data of the target area obtained in real time, and extract cold air directivity feature data and local disturbance response data from it after preprocessing;

[0066] Before comprehensively analyzing the airflow characteristic data of the target area, it is necessary to preprocess it first to improve the data quality, enhance the stability and reliability of feature expression, and ensure that the input of the subsequent neural network model has a unified structure and mathematical operability. Since the original data collected from the air supply outlet, air return outlet and inside the channel often have problems such as noise interference, data missing, inconsistent dimensions, and uneven sampling frequencies, it is necessary to clean, repair and standardize the data through software methods. The specific preprocessing process includes the following steps: First, handle missing values. For instantaneously lost data points, interpolation methods (such as linear interpolation, spline interpolation) can be used to complete the filling; second, remove outliers. Identify abnormal fluctuation points by setting dynamic thresholds based on historical statistical features and mask them; then perform normalization or standardization processing on different physical quantities, and map data such as wind speed, temperature, humidity, and air pressure to the interval [0,1] or the standard normal distribution interval with a mean of 0 and a standard deviation of 1 to eliminate the influence of unit scales; finally, perform time series resampling and alignment to ensure that various features have synchronous corresponding relationships at the same moment. All these operations are automatically executed by the data preprocessing module set in the software to form a structured, unified and high-quality input data set.

[0067] After completing the data preprocessing, the system will structurally extract the airflow characteristic data of the target area according to the established feature classification rules to construct two core feature sets required for the cold air offset index and the airflow interference coefficient. When extracting the cold air directivity feature data, the software system first obtains the real-time wind speed value and the reference wind speed value of the target area at the current moment, calculates the wind speed ratio between the two, and at the same time uses the cosine value of the angle between the current wind speed direction and the reference direction to extract the wind direction offset degree index; these two values constitute the cold air directivity feature data, which is used to reflect whether the cold air is accurately delivered to the target area. When extracting the local disturbance response data, the system performs differential calculations on the temperature value, humidity value and wind speed value within the sliding time window to obtain the temperature change rate, humidity change rate and wind speed standard deviation, and calculates the air pressure difference between the air supply inlet and outlet of this area, and then constitutes a feature set reflecting the local flow field disturbance degree. All feature extractions are encapsulated in the software through feature engineering algorithms and automatically executed according to data field classification, time window management and mathematical calculation rules, without manual intervention, ensuring real-time and high consistency.

[0068] Combine the extracted cold air directivity feature data and local disturbance response data with the neural network model trained based on historical samples for comprehensive analysis, and generate the cold air offset index and the airflow interference coefficient respectively;

[0069] Construct an airflow conflict assessment model, input the generated cold air offset index and airflow interference coefficient into this model, and generate the airflow conflict index through weighted summation;

[0070] Determine the pre-set air flow conflict index threshold range, and after determination, compare it with the generated air flow conflict index, and evaluate the air flow conflict degree of the target area according to the comparison result.

[0071] To achieve the automatic grading evaluation of the air flow conflict degree of the target area, the system needs to set a reasonable air flow conflict index threshold range in a software manner before model establishment, which is used to divide the finally generated conflict index into multiple levels. This threshold range can be analyzed and dynamically optimized offline based on historical operation data samples. The specific steps are as follows: First, the system extracts the corresponding cold air offset index and air flow interference coefficient from the datasets that have marked or evaluated the air flow state under multiple different working conditions in the past, and calculates the historical distribution sequence of the air flow conflict index through weighted fusion; Then, based on clustering algorithms (such as K-means or hierarchical clustering), segmental clustering is performed on the distribution of the conflict index, and the data is naturally divided into three grade intervals, such as low conflict, medium conflict, and high conflict; Next, the system automatically extracts the boundary values of each clustering interval as the initial threshold; Finally, combined with expert experience rules or manual intervention windows, strategic fine-tuning is performed on the clustering results to make the threshold division have both statistical significance and meet the engineering practical controllability. All analysis processes are implemented through the threshold generation module in the software system, forming a parameter structure that can be persistently stored, for online model comparison and calling, and supporting self-learning optimization according to newly added operation data regularly to achieve the long-term adaptive ability of the threshold range.

[0072] In this embodiment, the training process of the neural network model trained based on historical samples includes the following steps:

[0073] Collect historical samples, where the historical samples include the cold air directivity feature data, local disturbance response data of the target area under different operating states in the historical time period, and the corresponding air flow conflict degree level labels; perform format standardization and numerical normalization processing on the historical samples, and divide them into a training set, a validation set, and a test set according to a set ratio; construct a multi-layer neural network model including an input layer, a hidden layer, and an output layer, where the input layer receives the feature data vector and the output layer generates the adjustment factor result; use the supervised learning method to continuously update the network weights based on the error backpropagation algorithm by minimizing the error between the conflict evaluation index generated after the adjustment factor and the true level label until the training error meets the preset convergence condition, and complete the training of the neural network model; solidify and save the trained model parameter structure for subsequent generation and evaluation analysis of the adjustment factor for the real-time extracted feature data.

[0074] To achieve the intelligent judgment of the cold air supply offset and local disturbance intensity under different airflow states by the neural network model, it is necessary to construct a neural network training process based on historical samples through software. First, the system calls the historical operation database in the background scheduling module to extract the target area data covering different environmental loads, equipment configurations, and air supply strategies, including cold air directivity feature data (such as wind speed changes, wind direction offset degrees) and local disturbance response data (such as temperature change rates, wind speed fluctuations, air pressure differences, etc.), and correspondingly extracts the airflow conflict degree level labels manually marked or recorded by the strategy feedback system during the corresponding time period. Subsequently, the data preprocessing module standardizes the sample format, normalizes the values, and interpolates the missing values to ensure that the input data has a unified scale and integrity among different dimensions. The processed sample set is divided into a training set, a validation set, and a test set according to a preset ratio and input into the multi-layer feedforward neural network model for supervised learning training. Through the error backpropagation mechanism, the weight update is continuously iterated, enabling the model to learn the non-linear relationship and relative importance of various features in the conflict assessment. After training, the model output structure will be used to generate adjustment factors in real time to drive the airflow conflict index calculation process, ultimately realizing the intelligent and quantitative assessment of cold air offset and disturbance intensity. This process is fully realized by the software system through a data-driven method without manual intervention, supports periodic self-training to adapt to different engineering conditions, and has extremely high adaptability and scalability.

[0075] The neural network model training process defined aims to establish an evaluation engine with self-learning ability to realize the mapping relationship between airflow state characteristics and actual conflict levels. Among them, historical samples not only contain the original airflow state data, but also contain label information directly related to airflow stability, making the training process have a clear supervision signal. Through format standardization and normalization processing, it can ensure the consistency of different data dimensions at the numerical level and avoid the problem of biased learning caused by differences in unit dimensions during the model learning process; the division of the training set and the validation set helps to control the balance between the model fitting ability and the generalization ability. The neural network structure adopted can include one or more layers of hidden neurons, and the number of neurons can be dynamically adjusted according to the sample complexity to adapt to different feature dimensions and non-linear intensities; the backpropagation algorithm realizes the optimal fitting of the feature-label relationship by gradually reducing the loss value between the model output and the conflict level label. After training, the connection weights formed inside the model are used as the importance reference for each input feature, and are used to automatically generate adjustment factors for parameter calculation during real-time data analysis, realizing an intelligent evaluation closed-loop that extracts rules from historical experience and makes responses from real-time data.

[0076] In this embodiment, the acquisition logic of the cold air offset index is as follows:

[0077] Extract the cold air directivity feature data from the airflow feature data of the preprocessed target area, specifically including the angle between the cold air wind direction and the preset reference wind direction at different moments within a period of time, the average cold air wind speed, and the cold air reference wind speed under normal operating conditions, and calibrate them respectively as , and , indicating the angle between the cold air wind direction and the preset reference wind direction in the target area at the moment within a period of time, indicating the average cold air wind speed in the target area at the moment within a period of time, indicating the cold air reference wind speed under normal operating conditions, , where

[0078] is a positive integer; To achieve the dynamic evaluation of the cold air offset index, key data such as the cold air wind direction, cold air wind speed, and reference wind speed at multiple moments within a period of time in the target area need to be obtained in real time. First, through a multi-point wind speed sensor array (such as an ultrasonic anemometer or a thermal anemometer) deployed in the cold air channel of the industrial air conditioner, the instantaneous wind speed value and wind direction angle of each sub-area are collected in real time. The system calculates the angle between the actual cold air flow direction and the set reference air supply direction using the built-in trigonometric function model of the software based on the component information output by the wind speed sensor, and records it as , which reflects whether the cold air deviates. The average cold air wind speed is obtained by weighted averaging the data of multiple wind speed measurement points in the same sub-area within a set time window (such as every 10 seconds or 30 seconds), and is used to represent the cold air flow intensity in this area at the current moment. As for the cold air reference wind speed , it is generated by the system based on the statistical mean of the wind speed in the optimal energy-saving period without disturbance in the historical operation data of this area, and is preset as the reference wind speed in the software parameter module for dynamically comparing the difference between the current state and the ideal state. The acquisition of these data can be accessed to the central processing module through the communication interface, and the full-automatic real-time processing is completed through the interaction between the data acquisition layer and the calculation module, ensuring that the entire offset index evaluation process has high accuracy and timeliness without manual intervention.

[0079] Determine the values of the adjustment factors and based on the neural network model trained with historical samples. The adjustment factor is used to adjust the influence weight of the wind direction offset term in the cold air offset index, and the adjustment factor is used to adjust the influence weight of the wind speed deviation term in the cold air offset index;

[0080] To achieve an adaptive adjustment mechanism for the cold air offset index during the evaluation process, the system dynamically determines the adjustment factors through a neural network model trained based on historical samples. and values. Specifically, during the training phase of the neural network model, the cold air directivity feature data extracted from historical samples and the corresponding conflict level labels are input. The model learns the influence intensity of feature variables (such as wind direction angle, wind speed deviation) on the conflict result, and gradually optimizes the weights of the hidden layer through the backpropagation algorithm. Finally, the factors and generated by the output layer can accurately reflect the relative contribution degree of each feature in the calculation of the conflict index. During real-time operation, the model receives the preprocessed current feature data, maps it into the neural network, and outputs a set of adjustment factors applicable at the current moment according to the trained connection weights. The adjustment factor is used to control the influence intensity of the wind direction offset term on the final cold air offset index. If the current features indicate that the wind direction deviation is highly correlated with the air flow conflict, then the output value will increase; the adjustment factor is used to control the contribution degree of the wind speed deviation term. In actual operation, the influence of wind speed changes on energy efficiency under different loads and different structures is not constant, so it is necessary to dynamically assign weights according to the working conditions. The core purpose of introducing these two adjustment factors is to enable the offset index evaluation model to have "data-driven self-adaptability", breaking through the problems of large evaluation deviation and weak generalization ability of traditional fixed-weight models when facing different environments, thereby improving the accurate perception of complex air flow behavior and energy-saving judgment ability.

[0081] Calculate the cold air offset index, and the specific calculation formula is as follows:

[0082] In the formula, is the cold air offset index.

[0083] To accurately measure the deviation of the cold air supply direction and wind speed in the target area during continuous operation, this cold air offset index calculation formula is used for quantitative evaluation. This formula analyzes the wind direction and wind speed offsets at each sampling moment within a set time period point by point, and performs weighted processing based on the adjustment factors generated by the neural network, so as to reflect the overall offset trend. The summation part in the formula represents the gradual accumulation of the local cold air states at all moments within the time period, reflecting the stability analysis under time continuity. The first term in the expression is used to measure the angle offset degree between the cold air wind direction and the preset reference wind direction at the moment. The value range is between 0 and 2. The cosine function can convert the direction deviation into a unitless numerical difference to ensure continuous quantification. This offset degree is multiplied by the adjustment factor output by the neural network is used to dynamically adjust the weight of the wind direction in the overall offset evaluation according to the historical sample learning results. The second term is used to measure the attenuation degree of the current wind speed compared with the ideal reference wind speed. The logarithmic function is used to compress the influence brought by the extreme wind speed ratio and enhance the sensitivity of the model to small deviations. The addition of 1 operation is to avoid division by zero and improve numerical stability. This wind speed deviation term is also multiplied by the adjustment factor generated by the neural network to reflect the dynamic importance of the wind speed change in the overall offset. Finally, the entire expression is divided by to calculate the average value of the offset degree at each moment within the time period, and obtain the cold air offset index representing the overall deviation degree of the target area . The larger the index value, the more serious the deviation of the cold air in terms of direction and flow rate, indicating that the system needs to adopt an energy-saving regulation strategy for correction.

[0084] Cold air offset index is positively correlated with the degree of air flow conflict in the target area, that is the larger the value, the more serious the deviation of the cold air flow from the ideal air supply state in the two key dimensions of direction and wind speed, which means that there is more likely to be abnormal cold air transmission paths, ineffective delivery of cold air to the target area, or interference between cold and hot air flows in this area, constituting local air flow conflicts. Specifically, when the wind direction offset term in is continuously at a high value, it indicates that the cold air flow deviates from the target channel direction, which is likely to cause cold air reflux, bypass flow, or hot air intrusion into the cold area, reducing the heat transfer efficiency; while a higher wind speed deviation term indicates that the kinetic energy of the cold air weakens and the air supply effect declines, which will also cause uneven cold and hot distribution, thus inducing response behaviors such as local air supply and overcooling in the system that amplify energy consumption. Therefore, the cold air offset index not only reflects the physical degree of the cold air flow deviating from the reference path, but can also be used as one of the key indicators for evaluating the efficiency of air flow organization and the effectiveness of local energy-saving strategies. When the index reaches or exceeds the preset threshold, it can be used to judge that there is a high conflict risk in the target area, thereby triggering a dynamic regulation strategy to optimize the air flow structure and reduce energy consumption.

[0085] In this embodiment, the acquisition logic of the air flow interference coefficient is as follows:

[0086] Extract local disturbance response data from the preprocessed air flow characteristic data of the target area, specifically including the temperature change rate of the target area at different moments within a period of time, the pressure difference between the air supply inlet and outlet, and the static pressure value, the average cold air speed, and the standard deviation of the cold air speed at the cold air outlet of the target area within this time period, and respectively calibrate them as , , , and , represents the temperature change rate of the target area at a moment within a period of time and represents the air pressure difference between the air supply inlet and outlet of the target area at a moment within a period of time represents within a period of time the air pressure difference between the air supply inlet and outlet of the target area at a moment represents the static pressure value of the cold air outlet of the target area within this time period represents the average value of the cold air speed of the target area within this time period represents the standard deviation of the cold air speed of the target area within this time period , is a positive integer;

[0087] To achieve the dynamic quantitative evaluation of the airflow disturbance characteristics of the target area, the system needs to continuously obtain a series of physical quantities characterizing the airflow state changes within the set time window, including temperature change rate, air pressure difference, outlet static pressure value, average cold air speed, and wind speed standard deviation. The temperature change rate is obtained by the digital temperature sensors arranged at multiple sub-positions in the target area to obtain the temperature sequence at regular intervals, and the software system calculates the temperature change rate at each moment by dividing the temperature difference between two consecutive moments by the time interval , which is used to reflect the local thermal disturbance trend; the air pressure difference between the air supply inlet and outlet is obtained by the differential pressure sensors respectively deployed at the starting section of the air supply and the outlet of the end channel to obtain the static pressure readings, and the system automatically calculates the difference between the two; while the static pressure value of the cold air outlet of the target area is continuously collected by the static pressure sensor arranged at the cold air outlet and is used as the reference for differential pressure normalization; the average cold air speed and the wind speed standard deviation are obtained by the wind speed sensor array arranged in the ventilation channel of the target area to obtain the continuous time series data. The sliding average and standard deviation are calculated by the software module within each time period, respectively reflecting the current flow velocity intensity and fluctuation stability. All the above data are reported to the central control platform in real time through the sensor acquisition system, archived and called uniformly by the data acquisition interface of the software, and standardized processing such as time alignment, interpolation completion, and unit normalization is completed by the data processing module to ensure that the input data for subsequent neural network model analysis and airflow interference coefficient calculation has continuity, consistency, and high timeliness, thus supporting the dynamic energy-saving regulation judgment based on the disturbance behavior

[0088] The neural network model trained based on historical samples determines the values of the adjustment factors , and . The adjustment factor is used to adjust the weight of the temperature disturbance term in the overall interference evaluation, and the adjustment factor is used to adjust the weight of the wind speed fluctuation term in the overall interference evaluation, and the adjustment factor Used to adjust the weight of the pressure difference term in the overall interference assessment;

[0089] To achieve the adaptive quantitative assessment of the air flow interference coefficient under different operating conditions, the system dynamically determines the adjustment factors through a neural network model trained based on historical samples. , and During the training process, the system collects historical sample data of multiple known disturbance situations. These samples include disturbance indicators such as the temperature change rate, wind speed fluctuation intensity, and supply air pressure difference in the target area under different loads, climates, and equipment configurations, as well as the air flow conflict level labels generated by manual annotation or operation feedback corresponding to them. Under the supervised learning framework, the neural network inputs the disturbance data, outputs the adjustment factor vector, and optimizes the error backpropagation based on the conflict level label, continuously adjusting the model weights until the training converges. After the training is completed, the neural network model can receive the latest disturbance data during real-time operation and output the , and values that are most suitable for the current working conditions through forward calculation. The adjustment factor is used to control the influence intensity of the temperature disturbance term on the air flow interference coefficient. If the system detects that the temperature fluctuation is highly sensitive to conflicts in the historical samples, then the weight is increased; characterizes the relative weight of the wind speed fluctuation term and reflects the contribution degree of the flow velocity instability to the air flow disorder; is used to adjust the importance of the pressure difference disturbance in the overall interference assessment, and its weight value increases significantly especially when the cold and hot channel isolation is poor or there is a sudden change in wind resistance. The core purpose of introducing these three adjustment factors is to enhance the robustness and flexibility of the interference assessment model, enabling it to dynamically assign weights according to the changes of disturbance factors in different scenarios, overcoming the problems of large evaluation deviation and high strategy misjudgment rate of the traditional fixed weight model under complex air conditioning operating conditions, so as to achieve high-precision and intelligent air flow conflict identification and energy-saving control response.

[0090] Calculate the air flow interference coefficient, and the specific calculation formula is as follows:

[0091] In the formula, is the air flow interference coefficient.

[0092] To accurately depict the disturbance degree of the air flow state in the target area over a period of time, this air flow interference coefficient calculation model is used for evaluation. This model accumulates and sums based on the disturbance data at multiple moments and performs nonlinear processing through a logarithmic compression function to enhance the sensitivity to severe disturbances and at the same time suppress the amplification effect of extreme outliers. In this formula, represents the The absolute value of the temperature change rate at a certain moment reflects the degree of thermal disturbance caused by the intersection of cold and hot gases at that moment. The higher the value, the more drastic the local temperature fluctuation. represents the normalized term of the fluctuation intensity of the cooling air speed, where is the standard deviation of wind speed, is the mean wind speed, which is used to quantify the fluctuation of the cold air flow stability during this period. The greater the fluctuation, the more unstable the flow field. It is used to express the pressure difference disturbance between the air inlet and outlet at that moment, and is expressed by the outlet static pressure Normalization processing not only ensures the uniformity of physical dimensions, but also enhances the adaptability of the model under different pressure conditions. The three disturbance terms are multiplied by the adjustment factors output by the neural network model. , and , in order to achieve the model's dynamic weighted reflection of different disturbance sources, ensuring that the evaluation system has a high degree of adaptability and self-adjustment capabilities. Finally, all disturbance values ​​are compressed through a logarithmic function, and then the average value within the time period is taken and output as the airflow disturbance coefficient , which is used to quantitatively judge the overall airflow turbulence degree of the area in the current cycle. This model not only retains the independent influence of multi-source disturbance factors, but also reflects their overall superposition effect through a nonlinear fusion mechanism, and has strong engineering versatility and energy-saving judgment adaptability.

[0093] Airflow interference coefficient The value of is positively correlated with the degree of airflow conflict in the target area, that is, The higher the value, the more significant the airflow disturbance in the area during the monitored period, which means that the local area is more likely to have problems such as mixing of cold and hot gases, wind speed disturbance or abnormal air supply path, which will lead to reduced cold air distribution efficiency and increased energy consumption. The composition includes key disturbance indicators such as temperature change rate, wind speed fluctuation intensity and pressure difference abnormality. These factors directly affect the stability and transmission efficiency of airflow organization in the cold and hot channel structure of air conditioners. When the temperature change rate increases significantly, it means that there may be crossover of cold and hot gases in the area or the cold air flow is back pressed by the hot air, causing local thermal disturbances; when the wind speed standard deviation increases, it reflects the discontinuity of the air supply flow field or the enhancement of turbulence, which reduces the uniformity of air supply coverage; and large pressure difference fluctuations indicate the risk of cold air backflow or channel blockage. After weighted fusion of the influence intensity of these disturbance factors through the adjustment factors generated by the neural network, The damage degree of the above disturbances to the system stability under specific working conditions is comprehensively evaluated. It can be used as an important evaluation basis for judging whether there is moderate or severe air flow conflict in the target area. When this value continuously exceeds the upper limit of the preset threshold range, it indicates that there are very likely problems such as a decrease in the air supply efficiency and inaccurate transfer of the cooling load in this area, and it is necessary to immediately trigger the corresponding energy-saving control strategy for intervention.

[0094] In this embodiment, an air flow conflict evaluation model is constructed, and the generated cold air offset index and the air flow interference coefficient are input into this model, and the air flow conflict index is generated through weighted summation. The specific calculation formula is as follows:

[0095] In the formula, is the air flow conflict index, and are the non-zero weight coefficients of the cold air offset index and the air flow interference coefficient respectively, and .

[0096] The air flow conflict evaluation model is constructed by using a weighted fusion method, aiming to unify and integrate the two evaluation dimensions of the cold air offset index and the air flow interference coefficient to obtain a quantitative index - the air flow conflict index that can comprehensively reflect the degree of air flow conflict in the current target area. This evaluation model is constructed based on a linear weighted summation structure, and the expression is , where and are two non-zero weight coefficients, and . The two non-zero weight coefficients and in the air flow conflict evaluation model are respectively used to control the influence ratio of the cold air offset index and the air flow interference coefficient in the overall evaluation result, and their values are dynamically determined by a pre-trained neural network model in the system. Specifically, the neural network uses a large number of historical samples as input during the training stage, including cold air directional feature data, local disturbance response data under different working conditions, and corresponding conflict level labels, and continuously optimizes the mapping relationship between the feature input and the conflict level through the supervised learning process. After the model converges, the system inputs the currently extracted feature data into this neural network model during real-time operation, and the neural network outputs a set of normalized weight coefficients and to adjust the relative weights of CADI and ADC in the process of calculating the air flow conflict index (ACI) to reflect which type of factor has a higher dominance in the formation of conflicts under the current working conditions. Among them, represents the sensitivity weight of the wind direction offset type feature in the overall evaluation, It represents the comprehensive influence intensity of local interference factors such as temperature fluctuation, wind speed disturbance, and air pressure anomaly. The sum of the two is always 1. By introducing this dynamic weight mechanism, the evaluation model can be made to have environmental adaptability, effectively avoiding evaluation biases caused by fixed weight settings, thereby achieving more accurate and reliable identification of conflict levels.

[0097] In this example, a pre-set threshold range of the air flow conflict index is determined. After determination, it is compared with the generated air flow conflict index to evaluate the air flow conflict degree of the target area according to the comparison result. The specific comparison and analysis are as follows:

[0098] If , the air flow conflict degree of the target area is at a low level;

[0099] This situation indicates that the deviation degree of the cold air supply direction in the current target area is small, and the disturbance indicators such as temperature, wind speed, and air pressure are within the stable fluctuation range. The overall air flow organization state is good, the cold and hot channel isolation structure basically functions normally, and there are no obvious turbulent flow, countercurrent, or reflux phenomena. At this time, the system can be judged as a "low conflict state", and there is no need to trigger any energy-saving compensation actions. Only maintaining the existing operating parameters is beneficial to reducing the control frequency, lowering the system load, and reflecting the steady-state tolerance and regulation economy of the energy-saving model.

[0100] If , the air flow conflict degree of the target area is at a medium level;

[0101] This situation indicates that there are certain degrees of air flow disturbance behaviors in the target area, such as local cold air deviation, unstable wind speed, or short-term drastic temperature changes. Although it has not evolved into significant turbulent flow or energy efficiency out of control, it has already affected the air supply efficiency and heat exchange effect. Such a "medium conflict state" is usually reflected in the inaccurate transfer of the cooling load or the existence of overcooling / overheating problems in some areas. At this time, it is recommended that the system start a "lightweight energy-saving regulation strategy", such as appropriately fine-tuning the air supply direction, wind speed, or pressure difference setting, to suppress the potential deterioration trend as early as possible, ensure the operation stability of the system, and improve the energy efficiency utilization rate.

[0102] If , the air flow conflict degree of the target area is at a severe level.

[0103] This situation indicates that there is a serious disorder in the air distribution in the target area. Common manifestations include the serious deviation of the cold air direction to non-target areas, cold air reflux or being counter-pressed by hot air, strong fluctuations in the air pressure at the air supply or return air vents, and local temperature out of control. Such a "serious conflict state" will lead to the failure of the cold and hot channels, a significant decrease in energy utilization efficiency, and may cause frequent start-stop of air conditioning equipment and an increase in failure rate. At this time, the system needs to immediately trigger a "strong response energy-saving strategy", such as reconstructing the air supply path, restricting the cooling supply in abnormal areas, and temporarily switching the operating mode, etc., to quickly restore the channel isolation structure and the balance of the air flow field, prevent further deterioration of energy consumption, and ensure the stability and safety of the system.

[0104] Based on the evaluation results, execute the corresponding energy-saving control strategy and dynamically adjust the cold air flow direction;

[0105] In this example, based on the evaluation results, execute the corresponding energy-saving control strategy and dynamically adjust the cold air flow direction. Specifically:

[0106] When the evaluation result is at a low level, the energy-saving control strategies to be executed include: keeping the current settings of the air supply speed, air supply direction, and output power in the target area unchanged, and continuously collecting air state information at a preset time interval for verification; the corresponding dynamic adjustment method for the cold air flow direction is: maintaining the air supply direction parameter at the initial setting without performing direction correction operations;

[0107] In the case where the evaluation result is at a low level, the conservative control strategy template can be called through the software configuration policy selector. The parameters in this strategy template remain at the current operating values, and the system does not execute adjustment instructions. Only at a set time interval, the data acquisition module is called to regularly read the air supply speed, air supply direction, and output power data from the sensors, and compare them with the fluctuation tolerance interval in the historical operation data. If the comparison result shows a stable state, the status quo is retained without intervention. This method helps to reduce the control frequency, reduce energy consumption waste, keep the system running at the lowest dynamic power consumption state, and avoid the phenomenon of over-compensation for energy saving caused by excessive adjustment.

[0108] When the cold air flow direction does not need to be adjusted, the software control platform will keep the original air supply angle parameter unchanged in the air supply direction setting module. The system will continuously monitor the real-time output data of the air supply direction sensor and perform deviation analysis with the initial air supply direction setting value. If the direction error does not exceed the preset threshold, the system determines that the air supply direction is stable and does not generate control instructions. At this time, the control execution logic remains in an "empty instruction state", that is, no directional adjustment commands are issued, so as to maintain the continuous delivery of cold air in the original path. This strategy avoids unnecessary adjustment actions, ensures the stability of cold air supply, and conforms to the "minimum intervention principle" of energy-saving control.

[0109] When the evaluation result is medium, the energy-saving control strategies implemented include: modifying the air supply direction setting in the target area and adjusting the air supply speed so that the cold air flow converges towards the area with high air flow conflict risk; the corresponding dynamic adjustment method for the cold air flow direction is: updating the guiding parameters of the air supply unit, adjusting the cold air path to focus on the target cooling load area, and keeping the air supply direction stable after adjustment;

[0110] After detecting a medium conflict level, the software will automatically load the policy template with the adjustment level of "medium-level response" and start the wind speed and direction adjustment subroutine. In the specific implementation, the wind speed adjustment module dynamically increases the output frequency value of the air supply unit according to the wind speed attenuation trend analyzed from the preprocessed feature data; the direction adjustment module corrects the air supply direction setting through real-time guiding angle offset analysis to make the air supply trajectory closer to the target cooling load area. This process generates the target adjustment vector through the collaborative analysis of the cooling load distribution map and the disturbance response map. This strategy achieves a balance between precision control and local optimization, strengthening the cold air focusing effect while avoiding the increase in energy consumption caused by global resource reallocation.

[0111] For a medium conflict level, the software will enable the flow direction redirection mechanism. In this mechanism, the angle parameters of the air supply units in the target area are reset, and the scheduler issues adjustment commands by calling the wind direction control interface. The new setting value is calculated based on the preset offset curve and is ensured to be smooth and stable during the air supply direction adjustment process through a first-order lag filtering method. After the adjustment is completed, the control logic will set the new parameter to the locked state and will not perform a secondary direction modification before the end of the evaluation period. This method effectively improves the heat load response ability of the cold air to the key area by precisely redirecting the cold air path, while maintaining the coherence and predictability of the control.

[0112] When the evaluation result is severe, the energy-saving control strategies implemented include: terminating the air supply operation in non-target areas, and centrally adjusting the air supply pressure difference and direction setting in the target area to improve the stable delivery ability of the cold air; the corresponding dynamic adjustment method for the cold air flow direction is: closing the air supply paths of non-target channels, and jointly adjusting the air supply direction and speed setting to make the cold air focus on the air flow conflict area and improve the air flow organization efficiency.

[0113] When the conflict level is severe, the scheduler triggers the high-priority control strategy path and loads the emergency response strategy group. First, the air supply tasks outside the target area will be immediately aborted, specifically by setting the air supply output parameters to zero through control instructions; at the same time, the air supply pressure difference setting in the target area will be significantly increased by increasing the fan load frequency and the control parameters of the flow guiding structure; the direction setting is modified to the strong guidance mode, that is, forcing the cold air to bypass the interference path and directly reach the core of the conflict area. This strategy quickly re-concentrates the system resources at the conflict point, improves the concentrated penetration ability of the cold air flow, and suppresses the diffusion of countercurrent, backflow and cross-interference.

[0114] In a severe conflict state, the software control logic enters the forced reconstruction mode. In this mode, the air supply path of non-target channels is first closed, and the judgment criterion is that the air supply efficiency in this area is lower than the set value. Subsequently, for each air supply unit in the target area, the air supply direction parameter and the wind speed output setting are jointly adjusted to make the cold air path concentrate on the core position where the air flow conflict occurs. This operation realizes path reconstruction by establishing a "target path priority mapping table", and at the same time combines the wind speed gradient reverse feedback mechanism to continuously correct the air flow speed and angle until the path converges and stabilizes. This mechanism has extremely strong cold air organization ability, which helps to quickly relieve the imbalance state of the air flow structure and restore the basis of energy-saving operation.

[0115] Input the environmental response data after the implementation of the energy-saving regulation strategy into the neural network model, and enable the neural network model to continuously improve its recognition ability and regulation ability through continuous learning during operation.

[0116] In order to improve the adaptability and evaluation accuracy of the neural network model under different working conditions, after the implementation of the energy-saving regulation strategy, the environmental response data before and after the implementation can be automatically recorded through the data acquisition mechanism, including key parameters such as air supply wind speed, wind direction deviation, temperature change, pressure difference fluctuation, and cold load response change. These data are used as "regulation result samples" and are input into the online learning module in the neural network model through the set data channel. Specifically, the system will periodically collect these parameters according to the set feedback sampling interval and construct input-output pairs: the input is the feature data before regulation, and the output is the change in the environmental response index after regulation. Through these new sample data, the model can trigger the incremental training mechanism during operation, that is, perform small-scale and low-rate weight updates based on the current model weight structure without affecting the real-time operation efficiency. The update operation is still based on the backpropagation algorithm, but a lower learning rate and batch sampling frequency are used to avoid overfitting or system oscillation, so that the model can continuously optimize the recognition boundary and the adaptation ability of the regulation strategy without deviating from the online task.

[0117] The core purpose of introducing the continuous learning mechanism is to address the issues of operating condition drift and scenario changes existing in industrial air conditioners during actual operation, such as the change in cooling load demand with seasons, the response time delay caused by equipment aging, and the change in disturbance patterns caused by local renovation of the air circulation structure. Traditional static training models build evaluation capabilities only based on historical data and are prone to misjudgment in recognition or lag in control response in the above-changing environments. By continuously inputting the latest environmental response data and performing real-time incremental training, the neural network model can "self-adjust" during operation, dynamically updating its weight assignment logic for input features, thereby continuously enhancing the recognition ability for core features such as cold air offset and airflow interference. At the same time, the feedback of the control effect can also be perceived by the model, enabling it to gradually learn "what type of conflict is suitable for what type of control method", and ultimately achieving intelligent closed-loop regulation based on "effect orientation". This design significantly enhances the generalization ability and effectiveness of the model for complex and changing operating scenarios and is the key for intelligent energy-saving control to move from static evaluation to dynamic decision-making.

[0118] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0119] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0120] It should be understood that in various embodiments of the present application, the magnitudes of the serial numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0121] Those of ordinary skill in the art will realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0122] In several embodiments provided by this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the above-described embodiments are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0123] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0124] In addition, the functional units in each embodiment of this application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0125] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. An industrial air conditioning energy saving method based on a neural network, characterized in that, Specifically, it includes the following steps: The air state information of the hot and cold channel area inside the industrial air conditioner is obtained in real time through the sensor network arranged at the air supply outlet, air return outlet and hot and cold channel area of the industrial air conditioner, and it is analyzed to judge whether there is a failure of the hot and cold channel in the industrial air conditioner; When there is a failure of the hot and cold channel in the industrial air conditioner, locate the area where the hot and cold channel fails, mark this area as the target area, and obtain the airflow characteristic data of the target area in real time; The airflow characteristic data of the target area obtained in real time is comprehensively analyzed in combination with the neural network model trained based on historical samples to evaluate the degree of airflow conflict in the target area; Specifically, it includes the following steps: The airflow characteristic data of the target area obtained in real time is preprocessed, and the cold air directivity characteristic data and local disturbance response data are extracted from it after preprocessing; The extracted cold air directivity characteristic data and local disturbance response data are respectively comprehensively analyzed in combination with the neural network model trained based on historical samples to generate a cold air offset index and an airflow interference coefficient respectively; The acquisition logic of the cold air offset index is as follows: Extract the cold air directivity feature data from the preprocessed airflow feature data of the target area, specifically including the angle between the cold air wind direction and the preset reference wind direction at different times within a period of time, the average cold air wind speed, and the cold air reference wind speed under normal operating conditions, and calibrate them respectively as , and , represents the angle between the cold air wind direction and the preset reference wind direction of the target area at the moment within a period of time, represents the average cold air wind speed of the target area at the moment within a period of time, represents the cold air reference wind speed under normal operating conditions, , is a positive integer; Determine the adjustment factor based on the neural network model trained with historical samples and the value of, the adjustment factor is used to adjust the influence weight of the wind direction offset term in the cold air offset index, and the adjustment factor is used to adjust the influence weight of the wind speed deviation term in the cold air offset index; Calculate the cold air offset index, and the specific calculation formula is as follows: In the formula, is the cold air deviation index; The acquisition logic of the airflow interference coefficient is as follows: The local disturbance response data are extracted from the preprocessed airflow characteristic data of the target area, including the temperature change rate of the target area at different times within a period of time, the pressure difference between the air supply inlet and outlet, and the static pressure value, the mean value of the air conditioning wind speed and the standard deviation of the air conditioning wind speed at the air conditioning outlet of the target area within the period of time, and are calibrated as , , , and , Indicates that within a period of time The temperature change rate of the target area at the moment, Indicates that within a period of time The air pressure difference between the air supply inlet and outlet of the target area at any moment, Indicates the static pressure value of the cold air outlet in the target area during this time period. Indicates the average air conditioning wind speed in the target area during this time period. Indicates the standard deviation of the cooling air speed in the target area during this time period. , is a positive integer; Determine the adjustment factor based on the neural network model trained with historical samples , and The value of, the adjustment factor is used to adjust the weight of the temperature perturbation term in the overall interference evaluation, and the adjustment factor is used to adjust the weight of the wind speed fluctuation term in the overall interference evaluation, and the adjustment factor is used to adjust the weight of the differential pressure term in the overall interference evaluation; Calculate the airflow interference coefficient, and the specific calculation formula is as follows: In the formula, is the air flow interference coefficient; Construct an airflow conflict evaluation model, input the generated cold air offset index and airflow interference coefficient into this model, and generate an airflow conflict index through weighted summation; Determine the preset airflow conflict index threshold interval, and compare it with the generated airflow conflict index after determination, and evaluate the degree of airflow conflict in the target area according to the comparison result; Based on the evaluation result, execute the corresponding energy-saving regulation strategy and dynamically adjust the cold air flow direction; Input the environmental response data after the execution of the energy-saving regulation strategy into the neural network model, and enable the neural network model to continuously improve the recognition ability and regulation ability through continuous learning during operation.

2. The industrial air conditioning energy saving method based on a neural network according to claim 1, characterized in that, When there is a failure of the hot and cold channel in the industrial air conditioner, locate the area where the hot and cold channel fails, specifically: When there is a failure of the hot and cold channel in the industrial air conditioner, evenly divide the cold air channel into several sub-areas, and each sub-area monitors the wind speed, wind direction, temperature and humidity data in real time through the sensor network; for each sub-area, when the wind speed in this sub-area is lower than the preset threshold of the normal wind speed, and the included angle between the wind direction in this sub-area and the preset wind direction exceeds the set maximum deviation angle, and the temperature change rate and humidity change rate in this sub-area both exceed their respective set normal fluctuation ranges, determine that this sub-area is the area where the hot and cold channel fails.

3. The industrial air-conditioning energy-saving method based on a neural network according to claim 2, characterized in that, The training process of the neural network model trained based on historical samples includes the following steps: Collect historical samples, where the historical samples include cold air directivity feature data, local disturbance response data, and corresponding air flow conflict degree level labels in different operating states of the target area during a historical time period; perform format standardization and numerical normalization processing on the historical samples, and divide them into a training set, a validation set, and a test set according to a set ratio; construct a multi-layer neural network model including an input layer, a hidden layer, and an output layer, where the input layer receives a feature data vector and the output layer generates a regulation factor result; use the supervised learning method to continuously update the network weights based on the error backpropagation algorithm by minimizing the error between the conflict evaluation index after generating the regulation factor and the true level label until the training error meets the preset convergence condition to complete the training of the neural network model; solidify and save the trained model parameter structure for subsequent generation and evaluation analysis of the regulation factor for the real-time extracted feature data.

4. The industrial air conditioning energy saving method based on a neural network according to claim 3, characterized in that, Construct an air flow conflict evaluation model and use the generated cold air offset index and the air flow interference coefficient as inputs to this model, and generate an air flow conflict index through weighted summation. The specific calculation formula is as follows: In the formula, is the air flow conflict index, and are respectively the non-zero weight coefficients of the cold air offset index and the air flow interference coefficient , and .

5. The industrial air-conditioning energy-saving method based on a neural network according to claim 4, wherein Determine the pre-set air flow conflict index threshold range , and after determination, compare it with the generated air flow conflict index , and evaluate the air flow conflict degree of the target area according to the comparison result. The specific comparison analysis is as follows: If , the degree of airflow conflict in the target area is low; If , the degree of air flow conflict in the target area is medium; If , the degree of air flow conflict in the target area is severe.

6. The industrial air-conditioning energy-saving method based on a neural network according to claim 5, characterized in that, Based on the evaluation result, execute the corresponding energy-saving regulation strategy and dynamically adjust the cold air flow direction, specifically: When the evaluation result is at a low level, the energy-saving regulation strategy to be executed includes: keeping the current settings of the air supply speed, air supply direction, and output power in the target area unchanged, and continuously collecting air state information for verification at a preset time interval; the corresponding dynamic adjustment method of the cold air flow direction is: maintaining the air supply direction parameter at the initialization setting without performing direction correction operations; When the evaluation result is at a medium level, the energy-saving regulation strategy to be executed includes: modifying the air supply direction setting of the target area and adjusting the air supply speed to concentrate the cold air flow towards the air flow conflict risk area; the corresponding dynamic adjustment method of the cold air flow direction is: updating the guiding parameter of the air supply unit, adjusting the cold air path to focus on the target cooling load area, and keeping the air supply direction stable after adjustment; When the evaluation result is at a severe level, the energy-saving regulation strategy to be executed includes: terminating the air supply operation in non-target areas, centrally adjusting the air supply pressure difference and direction setting in the target area to improve the stable delivery ability of the cold air; the corresponding dynamic adjustment method of the cold air flow direction is: closing the air supply path of non-target channels and jointly adjusting the air supply direction and speed setting to focus the cold air on the air flow conflict area and improve the air flow organization efficiency.

Images