A Fault Prediction System and Method for HVAC Equipment

Abstract

The present invention relates to the technical field of fault prediction, and specifically to a fault prediction system and method for heating, ventilation, and air conditioning (HVAC) equipment. The system includes: a data acquisition module that collects the temperature, pressure, humidity, and operating duration of the HVAC equipment to generate serialized input data; a model initialization module that establishes a fault prediction model based on this data and sets an initial probability distribution to form model parameters; and an online learning module that adjusts and updates the model parameters in real time to dynamically match changes in the equipment behavior. In the present invention, through real-time monitoring and the integration of continuous data streams, fault prediction becomes more accurate and immediate. By establishing a fault prediction model based on this data and initializing and continuously updating the model parameters, it is possible to dynamically adapt to changes in the equipment behavior, improving the prediction accuracy of future fault points. By dynamically adjusting the learning rate and model parameters, the system can adapt to environmental changes, avoid prediction failures caused by outdated models, improve the stability of equipment operation, and reduce maintenance costs.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and particularly to a fault prediction system and method for heating, ventilation, and air conditioning (HVAC) equipment. Background Art

[0002] The technical field of fault prediction involves using algorithms and data analysis tools to predict possible faults in equipment, so as to perform maintenance or replacement in a timely manner, thereby improving the reliability and efficiency of the equipment. This technology generally includes steps such as data collection, feature extraction, model training, and prediction output. By using real-time data collected by various sensors and analyzing the data through machine learning or statistical models, the future state of the equipment and possible fault points are predicted. Fault prediction technology is widely applied in multiple fields such as industrial production, energy management, and transportation. Especially for those equipment with high maintenance costs or serious consequences of faults, this technology has very important value.

[0003] Among them, a fault prediction system for HVAC equipment refers to a fault prediction solution specifically designed for HVAC systems. Its purpose is to continuously monitor the operating conditions and performance of the equipment, and analyze data in real time to predict possible faults or performance degradation. By comparing historical data and real-time data, the system uses a prediction model to warn of potential problems, enabling facility managers to take measures in advance to avoid large-scale faults and expensive emergency repairs, thereby ensuring the efficient and stable operation of the HVAC system.

[0004] The existing technology usually relies on preset fixed models and parameters. When the equipment state or environment changes, this often leads to inaccurate or delayed fault prediction. Fixed models are difficult to react to the real-time changes of the equipment and the environment in a timely manner, which may result in untimely fault diagnosis, increasing the difficulty and cost of equipment maintenance. For example, in the case of failing to adapt to rapidly changing temperature or pressure, the existing system may miss the warning of key components, leading to unexpected system shutdowns and affecting the overall operation efficiency. Summary of the Invention

[0005] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a fault prediction system and method for HVAC equipment.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A fault prediction system for HVAC equipment includes:

[0007] A data collection module collects the operating data of HVAC equipment, including temperature, pressure, humidity, and operating duration, monitors the data quality in real time, excludes outliers and interference in the data, integrates it into a data stream in a unified format, and generates serialized input data;

[0008] The model initialization module receives the serialized input data, establishes a fault prediction model, and performs initial parameter setting, including defining a target state transition matrix and emission probability for each device state, setting an initial probability distribution, and forming initial model parameters;

[0009] The online learning module continuously receives the serialized input data, uses the collected data to adjust and update the initial model parameters in real time, adjusts the state transition matrix and emission probability, enables the fault prediction model to dynamically match the changes in device behavior, and generates optimized model parameters;

[0010] The fault detection module uses the optimized model parameters to analyze the current data of the device, detects key points deviated from the fault prediction model, evaluates the occurrence probabilities of various faults in real time, determines the fault type based on the occurrence probability of the fault and the model output, and generates a fault detection result;

[0011] The system adaptation module dynamically adjusts the learning rate and parameters of the fault prediction model according to the fault detection result, including the state transition matrix, optimizes the model to match the current device state and environmental changes, and generates an adaptive adjustment result.

[0012] The serialized input data includes temperature data, pressure data, humidity data, and operation duration data. The initial model parameters include a state transition matrix, an emission probability, and an initial probability distribution. The optimized model parameters include an updated state transition matrix and an updated emission probability. The fault detection result includes a fault occurrence probability and a determined fault type. The adaptive adjustment result is specifically an updated learning rate and adjusted model parameters.

[0013] As a further solution of the present invention, the steps for obtaining the serialized input data are specifically as follows:

[0014] Collect real-time data of the operation of the HVAC equipment, including temperature, pressure, humidity, and operation duration, update the data points every minute, and generate an original dataset of the HVAC equipment;

[0015] Perform outlier detection and interference data removal on the original dataset of the HVAC equipment, use the Z-score method to determine the outlier threshold, and remove the data points exceeding three standard deviations to obtain a cleaned dataset;

[0016] Perform data serialization processing on the cleaned dataset, using the formula:

[0017] ;

[0018] Generate serialized input data ;

[0019] wherein, Represents the measurement value of a single data point, used to represent the actual readings monitored from the device. Represents the average value of the measurement values, used to balance the influence of outliers. Indicates the total number of data points, providing the scale of the data set. Is the weight of data point i, used to weight the key importance of each point in the overall data analysis.

[0020] As a further aspect of the present invention, the specific steps for obtaining the initialized model parameters are as follows:

[0021] Receive the serialized input data, identify different device operation modes and fault characteristics through running data analysis, define various device states, and generate a device state definition set.

[0022] Use the device state definition set to calculate the state transition probability and emission probability, construct a state transition matrix and an emission probability matrix, which are the probabilities of the device transitioning from one state to another and the generation probabilities of observed data under multiple states, to obtain a state transition and emission probability matrix set.

[0023] Starting from the state transition and emission probability matrix set, use the formula:

[0024] ;

[0025] Calculate the initial probability of each device state to form the initialized model parameters.

[0026] Wherein, Represents the initial probability of state i. Represents the distance from state i to state j, used to adjust the probability basis of state transition. Is the state transition weight, emphasizing the importance of the target state transition. Indicates the total number of states.

[0027] As a further aspect of the present invention, the specific steps for obtaining the optimized model parameters are as follows:

[0028] Receive the serialized input data, monitor the device behavior in real time, identify any deviations in the device state or the current fault mode, determine the device state change through data sampling and feature extraction, and generate an analysis result of the device state change.

[0029] Based on the analysis result of the device state change, refine the adjustment of the existing state transition matrix and emission probability, re-estimate the probability parameters and the probabilities between states through the Bayesian update method, and combine time series analysis to generate an adjusted state transition matrix and emission probability.

[0030] Combined with the adjusted state transition matrix and emission probability, use the formula:

[0031] ;

[0032] Generate optimized model parameters;

[0033] Among them, represents the newly optimized model parameters, is the previous model parameters, adjusted according to real-time data, and the weight reflects the actual observed criticality of state transition, ensuring computational stability, and is used to adjust the speed at which the model responds to new data.

[0034] As a further solution of the present invention, the steps for obtaining the fault detection result are specifically as follows:

[0035] Utilize the optimized model parameters to analyze the current operating data of the device, identify the deviation key points from the fault prediction model through fine-grained data feature extraction, monitor data fluctuations and abnormal indicators, and generate a deviation key point analysis result;

[0036] Based on the deviation key point analysis result, apply probability statistical methods to calculate the occurrence probabilities of various faults, combine data pattern recognition techniques to evaluate the influencing factors and probabilities of each fault mode, output the probability values of each fault type, and generate a fault probability calculation result;

[0037] Adopt decision analysis to process the fault probability calculation result and the model output, using the formula:

[0038] ;

[0039] Obtain a fault type evaluation and generate a fault detection result;

[0040] Among them, represents the comprehensive evaluation result, represents the calculated fault probability, used to measure the predictability of the occurrence of a fault, is the response intensity of the model to the fault, sets the baseline for fault determination, and adjusts the response speed and accuracy of the model to new data.

[0041] As a further solution of the present invention, the steps for obtaining the adaptive adjustment result are specifically as follows:

[0042] Analyze the fault detection result, identify the model parameters that need to be adjusted, including the learning rate and the state transition matrix, perform parameter analysis for changes in the device state and environment, and generate a parameter adjustment requirement;

[0043] Apply the parameter adjustment requirements, recalculate the values of the learning rate and the state transition matrix, adjust the parameters using an optimization algorithm, update the parameters, and generate a model adjustment result;

[0044] Integrate the model adjustment results, update the fault prediction model, using the formula:

[0045] ;

[0046] Optimize the parameters based on the gradient descent method, match the working state of the current device, and generate an adaptive adjustment result;

[0047] where, represents the current model parameters, is the original parameter, is the adjusted learning rate, is the parameter change amount calculated based on real-time data, guiding the direction and amplitude of model adjustment.

[0048] A method for predicting faults in HVAC equipment, the method for predicting faults in HVAC equipment is executed based on the above-mentioned HVAC equipment fault prediction system, and includes the following steps:

[0049] S1: Based on the collection of operation data, collect the details of temperature, pressure, humidity, and operation duration, remove outliers and interference factors from the data, unify the data into a standard format, perform data quality monitoring, and generate serialized input data;

[0050] S2: Based on the serialized input data, set the target state transition matrix and emission probability for each device state, initialize the probability distribution, perform parameter setting, and establish initial model parameters;

[0051] S3: Based on the initial model parameters, continuously receive the current serialized input data, monitor the changes in the device behavior in real time, perform real-time adjustment and update on the state transition matrix and emission probability to dynamically match the changes in the device behavior, and generate optimized model parameters through multiple parameter adjustment actions;

[0052] S4: Based on the optimized model parameters, analyze the current device data, monitor the key points deviating from the model, evaluate the probability of fault occurrence, judge the fault type according to the probability of fault occurrence, and generate a fault detection result.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are:

[0054] In the present invention, through the integration of real-time monitoring and continuous data streams, fault prediction becomes more accurate and immediate. By establishing a fault prediction model based on these data and initializing and continuously updating the model parameters, it can dynamically adapt to changes in device behavior, improving the prediction accuracy of future fault points. By dynamically adjusting the learning rate and model parameters, the system can adapt to environmental changes, avoiding prediction failures caused by outdated models, thereby enhancing the stability of device operation and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is the system flowchart of the present invention;

[0056] Figure 2 is the flowchart of the acquisition steps of the serialized input data of the present invention;

[0057] Figure 3 is the flowchart of the acquisition steps of the initialized model parameters of the present invention;

[0058] Figure 4 is the flowchart of the acquisition steps of the optimized model parameters of the present invention;

[0059] Figure 5 is the flowchart of the acquisition steps of the fault detection results of the present invention;

[0060] Figure 6 is the flowchart of the acquisition steps of the adaptive adjustment results of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0062] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. In addition, in the description of the present invention, the meaning of "a plurality" is two or more unless otherwise specifically defined.

[0063] Example 1, please refer to Figure 1 , the present invention provides a technical solution: A fault prediction system for heating, ventilation and air conditioning equipment includes:

[0064] The data acquisition module collects the operation data of HVAC equipment, including temperature, pressure, humidity, and operation duration, monitors the data quality in real-time, excludes outliers and interferences in the data, integrates them into a data stream in a unified format, and generates serialized input data;

[0065] The model initialization module receives the serialized input data, establishes a fault prediction model, and performs initial parameter setting, including defining the target state transition matrix and emission probability for each equipment state, setting the initial probability distribution, and forming the initialized model parameters;

[0066] The online learning module continuously receives the serialized input data, uses the collected data to adjust and update the initialized model parameters in real-time, adjusts the state transition matrix and emission probability, enables the fault prediction model to dynamically match the changes in equipment behavior, and generates optimized model parameters;

[0067] The fault detection module uses the optimized model parameters to analyze the current data of the equipment, detects the key points that deviate from the fault prediction model, evaluates the occurrence probability of multiple faults in real-time, determines the fault type based on the occurrence probability of the fault and the model output, and generates a fault detection result;

[0068] The system adaptation module dynamically adjusts the learning rate and parameters of the fault prediction model according to the fault detection result, including the state transition matrix, optimizes the model to match the current equipment state and environmental changes, and generates an adaptive adjustment result.

[0069] The serialized input data includes temperature data, pressure data, humidity data, and operation duration data. The initialized model parameters include the state transition matrix, emission probability, and initial probability distribution. The optimized model parameters include the updated state transition matrix and updated emission probability. The fault detection result includes the fault occurrence probability and the determined fault type. The adaptive adjustment result is specifically the updated learning rate and adjusted model parameters.

[0070] Please refer to Figure 2 , and the specific steps for obtaining the serialized input data are as follows:

[0071] Collect the real-time data of the operation of HVAC equipment, including temperature, pressure, humidity, and operation duration, update the data points every minute, and generate the original dataset of HVAC equipment;

[0072] During the process of collecting the operation data of HVAC equipment, first, the monitoring sensors are set at the key parts of the equipment to facilitate real-time collection of temperature, pressure, humidity, and operation duration. These data points are updated every minute to ensure the freshness and accuracy of the data. The key to this step is to correctly place and calibrate the sensors to ensure the accuracy of data collection. The selection of monitoring points is set according to the key performance parameters of the equipment operation, including setting temperature and pressure monitoring points at the inlet and outlet of the heat exchanger, setting a humidity monitoring point near the air handling unit, and the operation duration is directly read through the operating system of the equipment. Each data point is marked with a timestamp to facilitate time series analysis during subsequent data processing, thereby generating the original dataset of HVAC equipment.

[0073] Perform outlier detection and removal of interference data on the original dataset of HVAC equipment. Use the Z-score method to determine the outlier threshold and remove the data points that exceed three standard deviations to obtain the cleaned dataset.

[0074] During the process of data cleaning, the dataset first automatically screens for outliers through an algorithm. The key to this step is to use the Z-score method to determine whether the data exceeds the normal range. The data points that exceed three standard deviations will be regarded as outliers and removed. This method is based on statistical principles and can effectively identify and eliminate the data deviations caused by equipment failures or external interferences. Therefore, the Z-score calculation for each data point is based on the mean and standard deviation of the entire dataset. The key in this process is to maintain the consistency and accuracy of data processing to ensure that the dataset after each data cleaning can reflect the actual operating status of the equipment.

[0075] Perform data serialization processing on the cleaned dataset, using the formula:

[0076] ;

[0077] Generate serialized input data ;

[0078] where, represents the measured value of a single data point, used to represent the actual reading monitored from the equipment, represents the mean of the measured values, used to balance the influence of outliers, represents the total number of data points, providing the scale of the dataset, is the weight of data point i, used to weight the key importance of each point in the overall data analysis.

[0079] Formula: ;

[0080] The benefit of the formula is that by introducing the weight coefficient The square root operation enhances the sensitivity to abnormal data, and through weight adjustment, it can pay more attention to the changes of those key data points, effectively improving the accuracy and flexibility of data serialization processing.

[0081] Detailed formula explanation and formula calculation derivation process:

[0082] Consider a specific data set. Suppose there are the following temperature values for data points , average value , total number of data points , and assume that the weights of each data point are equal, that is . First, calculate the value of each data point:

[0083] ;

[0084] Then apply the square root and multiply by the weight :

[0085] ;

[0086] Add these values together to get: ;

[0087] Finally, divide by to obtain the serialized input data .

[0088] This result shows that the serialized input data represents the average value considering weights and abnormal data sensitivity, providing an accurate and weighted measure of the current operating state of the device. Through such calculations, the health status of the device can be monitored more precisely and necessary adjustments or maintenance can be carried out.

[0089] Please refer to Figure 3 , and the specific steps for obtaining the initialized model parameters are as follows:

[0090] Receive the serialized input data. Through running data analysis, identify the differentiated device operating modes and fault characteristics, define multiple device states, and generate a device state definition set;

[0091] The received serialized input data includes the operating information of multiple devices. These data are sorted by analyzing the device operating modes and fault characteristics, classified by clustering algorithms, and a set of states are defined for each device state. These states reflect the performance of the device under differentiated operating conditions. Through statistical analysis, the characteristics of multiple states are determined, and the device state set is defined accordingly. This process involves technical steps such as data preprocessing, feature extraction, and pattern recognition, ensuring the accuracy of the data set and the wide range of applications, and forming a device state definition set.

[0092] Using the device status definition set, calculate the state transition probability and emission probability, construct the state transition matrix and emission probability matrix, and obtain the probability of the device transferring from one state to another state and the generation probability of observation data under multiple states, so as to obtain the state transition and emission probability matrix set;

[0093] Using the device status definition set, combined with the state recognition algorithm and probability statistics method, calculate the transition probability from one state to another state, including the probability evaluation of the state transition event and the establishment of the probability distribution. Through the data-driven method, the state transition matrix M and the emission probability matrix E are established. These two matrices not only describe the state transition probability of the device, but also include the predictability of data observation under the target state. Through batch data analysis and probability model optimization, the prediction ability of the model and the practicality of the application are enhanced.

[0094] Starting from the state transition and emission probability matrix set, adopt the formula:

[0095] ;

[0096] Calculate the initial probability of each device state to form the initial model parameters;

[0097] Among them, represents the initial probability of state i, represents the distance from state i to state j, which is used to adjust the probability basis of state transition, is the state transition weight, emphasizing the importance of the target state transition, represents the total number of states.

[0098] Formula: ;

[0099] The benefit of the formula is that by introducing the distance and weight between states, the predictability of each state transition is adjusted, thus optimizing the accuracy of state prediction. Especially in complex device systems, it can better match the actual situation of multiple state transitions.

[0100] Detailed explanation of the formula and the derivation process of formula calculation:

[0101] First, set the distance from state i to state j. The distance is calculated through the change of the device's performance indicators, and the difference between the two states is measured by indicators such as current fluctuation and temperature change. Then, set the weight parameter . These weights are determined by the comprehensive factors of the device's working environment and state data, reflecting the key nature of the transition from state i to state j; by taking the logarithmic transformation of the distance for each pair of state transitions, the adjusted transition probability is obtained, and the initial probability of state i is obtained through weighted average 。

[0102] Set the distance from state 1 to state 2 , the distance from state 1 to state 3 , and the corresponding weights are respectively and , then the calculation process is as follows:

[0103] ;

[0104] The results show that the initial probability of state 1 is about 0.406, reflecting the criticality of state 1 in the device system and providing a probability basis for subsequent state transition prediction.

[0105] Please refer to Figure 4 , the steps for obtaining the optimized model parameters are specifically as follows:

[0106] Receive serialized input data, monitor the device behavior in real time, identify any deviations or current fault modes of the device state, determine the device state changes through data sampling and feature extraction, and generate an analysis result of the device state changes;

[0107] Receive serialized input data, monitor the device behavior in real time, identify any deviations or current fault modes of the device state through regular data sampling frequencies and feature extraction techniques. These data are preliminarily screened and processed to determine the details of the device state changes. Statistical analysis methods are used to analyze the deviation trends and abnormal patterns of the data. These analysis results will provide a basis for real-time updates of the fault prediction model, match the behavior changes of the device, and generate an analysis result of the device state changes.

[0108] Based on the analysis result of the device state changes, refine the adjustment of the existing state transition matrix and emission probability, re-estimate the probability parameters and the probability between states through the Bayesian update method, and combine time series analysis to generate an adjusted state transition matrix and emission probability;

[0109] Based on the analysis result of the device state changes, refine the adjustment of the existing state transition matrix and emission probability. This step involves using the Bayesian update method to re-estimate the state transition probability and emission probability. The key is to adjust these probability parameters in real time to match the current device behavior data, calculate the current transition probability through a probability statistical model. At the same time, combine the time series data of the device behavior to optimize the transition probability between each state. This not only requires calculating the probability value in the current state but also predicting future state changes, and generates an adjusted state transition matrix and emission probability.

[0110] Combine the adjusted state transition matrix and emission probability, and use the formula:

[0111] ;

[0112] Generate optimized model parameters;

[0113] Among them, represents the newly optimized model parameters, is the previous model parameter, adjusted according to real-time data, and the weight reflects the actual observed criticality of state transition, ensuring computational stability, used to adjust the speed at which the model responds to new data.

[0114] Formula: ;

[0115] The benefit of the formula is to provide a method for dynamically adjusting model parameters, allowing the model to match changes in device behavior. By introducing the concepts of weight and distance, the sensitivity and response ability of the model to new data are enhanced.

[0116] Detailed explanation of the formula and the derivation process of formula calculation:

[0117] Set the original model parameter to 0.6, the smoothing coefficient to 0.5, the total number of states to 3, the weights between states are set to 0.2, 0.3, 0.5 respectively, and the distances between states are set to 1, 2, 3, and the small constant is 0.01, and the calculation process is as follows:

[0118] ;

[0119] The result shows that the current model parameter is 0.5565, which indicates that the model adjusts its parameters after receiving new data to make it more suitable for the current state changes of the device and improves the prediction accuracy.

[0120] Please refer to Figure 5 , and the specific steps for obtaining the fault detection result are as follows:

[0121] Utilize the optimized model parameters to analyze the current operating data of the device, identify the deviation key points from the fault prediction model through fine-grained data feature extraction, monitor data fluctuations and abnormal indicators, and generate the deviation key point analysis result;

[0122] Analyze the current operating data of the device using the optimized model parameters. Through feature extraction and pattern recognition, identify the key deviation points from the fault prediction model, monitor data fluctuations and abnormal indicators, and conduct in-depth analysis of the data, including time series analysis and outlier detection, in order to accurately monitor the device status and respond to potential faults. This method relies on continuous data streams and real-time processing algorithms, ensuring the high sensitivity and responsiveness of the system. Generate the analysis results of key deviation points, providing the necessary information for the maintenance team to take preventive or repair measures.

[0123] Based on the analysis results of key deviation points, apply probability and statistics methods to calculate the occurrence probabilities of various faults, combine data pattern recognition technology to evaluate the influencing factors and probabilities of each fault mode, output the probability values of each fault type, and generate the calculation results of fault probabilities.

[0124] Based on the analysis results of key deviation points, use methods of statistics and probability theory to calculate the probability of fault occurrence, evaluate the influencing factors of each fault mode, and calculate the occurrence probabilities of various faults by using Bayesian networks and decision trees. These calculations rely on the data collected from the device and its performance indicators, combine the operating conditions and environmental factors of the device, output the probability values of each fault type, and generate the calculation results of fault probabilities, providing a scientific basis for subsequent fault type evaluation and fault response decision-making.

[0125] Adopt decision analysis to process the calculation results of fault probabilities and the model output, using the formula:

[0126] ;

[0127] Obtain the fault type evaluation and generate the fault detection results.

[0128] Among them, represents the comprehensive evaluation result, represents the calculated fault probability, used to measure the predictability of fault occurrence, is the response intensity of the model to the fault, sets the baseline for fault determination, and adjust the response speed and accuracy of the model to new data.

[0129] Formula: ;

[0130] The advantage of the formula is that by combining the fault probability and the model output , the weighted logarithmic transformation enhances the sensitivity of the model to low-probability faults, thus more accurately identifying and predicting rare events. Adjust the parameters and Provides additional flexibility to adjust the thresholds and responses of fault detection according to different operating environments.

[0131] Detailed formula explanations and formula calculation derivation processes:

[0132] Set the number of fault types , As the overall scaling factor, To avoid the case where the denominator is zero, set the probability of each fault , , , model output , , , fault threshold , , . Calculate the contributions of multiple fault items:

[0133] ;

[0134] ;

[0135] ;

[0136] Integrate multiple items: ;

[0137] The result shows that based on the current input parameters and model responses, the overall fault risk assessment value is 0.0119, which is the unified output result after the model calculates each fault item. This value is used as the final evaluation index for the step calculation and reflects the weighted influence intensity of each type of fault in the current state. This value can be further compared with the preset fault determination threshold to determine whether to trigger the alarm mechanism or take subsequent processing measures, thus completing the transition from quantitative calculation to actual determination steps.

[0138] Please refer to Figure 6 , the specific steps to obtain the adaptive adjustment result are as follows:

[0139] Analyze the fault detection results, identify the model parameters that need to be adjusted, including the learning rate and state transition matrix, perform parameter analysis according to the equipment state and environmental changes, and generate parameter adjustment requirements;

[0140] According to the fault detection results, the model parameters need to be adjusted, including the learning rate and the state transition matrix. The learning rate is obtained through experiments, and it is necessary to evaluate the convergence speed and stability of model training under different learning rates. The state transition matrix depends on the device state and environmental changes. It is necessary to set the transition probabilities between different states according to the device fault and environmental monitoring data. For the adjustment of the learning rate, first, it is necessary to collect previous data, including the device fault duration, fault frequency, and environmental change factors, calculate the initial learning rate value, and determine the optimal value through experimental tests. The setting of the state transition matrix needs to be based on the transition situation of the device under different working states, combined with the state transition model, and modeled through the Markov process to obtain the transition probabilities between states, and gradually adjusted and optimized through a data-driven method to improve the accuracy of the prediction model.

[0141] Apply the parameter adjustment requirements, recalculate the values of the learning rate and the state transition matrix, use the optimization algorithm to adjust the parameters, update the parameters, and generate the model adjustment results;

[0142] When applying the parameter adjustment requirements, first, it is necessary to recalculate the learning rate and the state transition matrix according to the previous fault detection results. The recalculation of the learning rate depends on the experimental results or the feedback of the optimization algorithm, including the common gradient descent method. By adjusting the learning rate, the optimal update of the model parameters is carried out to ensure the convergence speed and stability during the training process. The adjustment process of the state transition matrix is more complex, involving optimizing the transition probabilities between device states through the data from simulation and field tests. In this process, optimization algorithms such as genetic algorithms and particle swarm optimization can effectively adjust the parameters of the state transition matrix, so as to improve the matching ability and prediction accuracy of the model for device state transitions in a dynamically changing environment. Through cyclic optimization and feedback mechanisms, the currently adjusted parameters are obtained, enabling the model to still operate stably and provide state predictions under changing environmental conditions.

[0143] Integrate the model adjustment results, update the fault prediction model, using the formula:

[0144] ;

[0145] Optimize the parameters based on the gradient descent method, match the current working state of the device, and generate the adaptive adjustment results;

[0146] Among them, represents the current model parameters, is the original parameter, is the adjusted learning rate, is the parameter change amount calculated based on real-time data, guiding the direction and amplitude of model adjustment.

[0147] Formula: ;

[0148] The advantage of the formula is that by dynamically adjusting the learning rate and parameter update to match the changes in the device state, the prediction accuracy of the model for faults can be improved.

[0149] Detailed explanation of the formula and the derivation process of formula calculation:

[0150] Let the initial state transition matrix parameter be , the adjusted learning rate be , and the parameter adjustment amount obtained by analyzing the monitored data. Substitute these into the formula to calculate the current state transition matrix parameter:

[0151] ;

[0152] The results show that by fine-tuning the learning rate, the matching of the model to environmental changes is enhanced, which helps to improve the sensitivity and accuracy of prediction.

[0153] A method for predicting faults in heating ventilation and air conditioning (HVAC) equipment. The method for predicting faults in HVAC equipment is executed based on the above-mentioned HVAC equipment fault prediction system and includes the following steps:

[0154] S1: Based on the collection of operation data, collect details of temperature, pressure, humidity, and operation duration, remove outliers and interference factors from the data, unify the data into a standard format, perform data quality monitoring, and generate serialized input data;

[0155] S2: Based on the serialized input data, set the target state transition matrix and emission probability for each device state, initialize the probability distribution, perform parameter setting, and establish the initial model parameters;

[0156] S3: Based on the initial model parameters, continuously receive the current serialized input data, monitor the changes in the device behavior in real time, perform real-time adjustment and update of the state transition matrix and emission probability to dynamically match the changes in the device behavior, and generate optimized model parameters through multiple parameter adjustment actions;

[0157] S4: Based on the optimized model parameters, analyze the current device data, monitor the key points of deviation from the model, evaluate the probability of fault occurrence, determine the type of fault according to the probability of fault occurrence, and generate a fault detection result.

[0158] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the relevant art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A heating, ventilation and air conditioning equipment fault prediction system, characterized in that, The system includes: A data acquisition module that collects the operation data of HVAC equipment, including temperature, pressure, humidity, and operation duration, monitors the data quality in real time, eliminates outliers and interferences in the data, integrates them into a data stream in a unified format, and generates serialized input data; A model initialization module that receives the serialized input data, establishes a fault prediction model, and performs initial parameter setting, including defining a target state transition matrix and emission probability for each equipment state, setting the initial probability distribution, and forming initialized model parameters; An online learning module that continuously receives the serialized input data, uses the collected data to adjust and update the initialized model parameters in real time, adjusts the state transition matrix and emission probability, enables the fault prediction model to dynamically match the changes in equipment behavior, and generates optimized model parameters; A fault detection module that uses the optimized model parameters to analyze the current data of the equipment, detects the key points deviating from the fault prediction model, evaluates the occurrence probability of various faults in real time, determines the fault type based on the occurrence probability of the fault and the model output, and generates a fault detection result; A system adaptation module that dynamically adjusts the learning rate and parameters of the fault prediction model according to the fault detection result, optimizes the model to match the current equipment state and environmental changes, and generates an adaptive adjustment result; Among them, the specific steps for obtaining the initialized model parameters are as follows: Receive the serialized input data, identify different equipment operation modes and fault characteristics through operation data analysis, define the states of various equipment, and generate an equipment state definition set; Use the equipment state definition set to calculate the state transition probability and emission probability, construct a state transition matrix and an emission probability matrix, and obtain the state transition and emission probability matrix set for the probability of the equipment transitioning from one state to another and the generation probability of observed data under multiple states; Starting from the state transition and emission probability matrix set, use the formula: ； Calculate the initial probability of each equipment state to form the initialized model parameters; Among them, represents the initial probability of state i, represents the distance from state i to state j, which is used to adjust the probability basis of state transition, is the state transition weight, emphasizing the importance of target state transition, represents the total number of states; The specific steps for obtaining the optimized model parameters are as follows: Receive the serialized input data, monitor the equipment behavior in real time, identify any deviation or current fault mode of the equipment state, determine the equipment state change through data sampling and feature extraction, and generate an equipment state change analysis result; Based on the equipment state change analysis result, refine the adjustment of the existing state transition matrix and emission probability, re-estimate the probability parameters and the probability between states through the Bayesian update method, and combine time series analysis to generate an adjusted state transition matrix and emission probability; Combined with the adjusted state transition matrix and emission probability, use the formula: ； Generate the optimized model parameters; Among them, represents the optimized model parameters, is the previous model parameter, is a constant used to ensure computational stability, is a smoothing parameter used to adjust the speed at which the model responds to new data; The specific steps for obtaining the fault detection result are as follows: Use the optimized model parameters to analyze the current operation data of the equipment, identify the key points of deviation from the fault prediction model through fine-grained data feature extraction, monitor data fluctuations and abnormal indicators, and generate a key point of deviation analysis result; Based on the analysis results of the deviation key points, the probability of occurrence of various faults is calculated by applying probability statistics methods, and the influencing factors and probabilities of each fault mode are evaluated in combination with data pattern recognition technology, and the probability values of each fault type are output to generate the calculation results of fault probabilities. The calculation results of the fault probabilities and the model output are processed by decision analysis, using the formula: ； The fault type evaluation is obtained to generate the fault detection results. Among them, represents the comprehensive evaluation result, represents the calculated fault probability, which is used to measure the predictability of the occurrence of a fault, is the response intensity of the model to the fault, is the baseline for setting fault determination, and are used to adjust the response speed and accuracy of the model to new data.

2. The HVAC equipment fault prediction system according to claim 1, wherein The serialized input data includes temperature data, pressure data, humidity data, and operation duration data. The initialized model parameters include the state transition matrix, emission probability, and initial probability distribution. The optimized model parameters include the updated state transition matrix and updated emission probability. The fault detection results include the fault occurrence probability and the determined fault type. The adaptive adjustment results are specifically the updated learning rate and adjusted model parameters.

3. The HVAC equipment fault prediction system according to claim 2, wherein The specific steps for obtaining the serialized input data are as follows: Collect the real-time data of the operation of the HVAC equipment, including temperature, pressure, humidity, and operation duration, and update the data points every minute to generate the original dataset of the HVAC equipment. Perform outlier detection and interference data elimination on the original dataset of the HVAC equipment, use the Z-score method to determine the outlier threshold, and eliminate the data points that exceed three standard deviations to obtain the cleaned dataset. Perform data serialization processing on the cleaned dataset, using the formula: ； Generate serialized input data ; Among them, represents the measured value of a single data point, which is used to represent the actual reading monitored by the device, represents the average value of the measured values, which is used to balance the influence of outliers, represents the total number of data points, providing the scale of the data set, is the weight of data point i, which is used to weight the key importance of each point in the overall data analysis.

4. The HVAC equipment fault prediction system according to claim 1, wherein The specific steps for obtaining the adaptive adjustment results are as follows: Analyze the fault detection results, identify the model parameters that need to be adjusted, including the learning rate and the state transition matrix, perform parameter analysis for the equipment state and environmental changes, and generate parameter adjustment requirements. Apply the parameter adjustment requirements, recalculate the values of the learning rate and the state transition matrix, use the optimization algorithm to adjust the parameters, update the parameters, and generate the model adjustment results. Integrate the model adjustment results, update the fault prediction model, using the formula: ； Optimize the parameters based on the gradient descent method, match the working state of the current equipment, and generate the adaptive adjustment results. Among them, represents the current model parameters, is the original parameter, is the adjusted learning rate, is the parameter change calculated based on real-time data, guiding the direction and amplitude of model adjustment.

Images