Power failure detection method for an air conditioning apparatus

Abstract

This invention relates to the field of fault detection technology and provides a method for detecting power supply faults in air conditioning equipment. The method includes: deploying a power monitoring module and a status detection module in the air conditioning equipment to collect power supply data and operational data in real time; fusing the collected real-time operational data with historical fault data using data fusion technology to form a comprehensive dataset, and employing an association rule mining algorithm to reveal the correlation between equipment performance and power supply faults; establishing a power supply fault prediction model based on deep learning methods, and inferring the probability of a power supply fault by analyzing the changing trends of historical fault data and current operational data; when the fault prediction model identifies a power supply fault risk reaching a preset threshold, automatically issuing a warning signal through a linkage response mechanism, notifying the operator through a preset notification mechanism, and simultaneously triggering the equipment's adaptive adjustment mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and specifically to a method for detecting power supply faults of air conditioning equipment. Background Art

[0002] Fault detection refers to the process of identifying and locating abnormal states that occur in equipment, systems, or facilities. The purpose is to find out whether the system is working properly and indicate the possible types and locations of faults. In the fault detection of air conditioning power supplies, it is often regarded as an auxiliary or background problem, resulting in insufficient coverage.

[0003] In Chinese Patent Application No. CN202411421333.1, there is a method for detecting power supply faults of air conditioning equipment. This configuration method includes: respectively obtaining the main board voltage detection value, transformer voltage detection value, and plug voltage detection value of the air conditioning equipment under the power-on state, and performing data cleaning processing to obtain the preprocessed air conditioning operation data. Then, draw the power phase waveforms between the main board, transformer, and power plug of the air conditioning equipment and analyze the power phase sequence balance relationship of the air conditioning equipment. Combine the historical fault data of the air conditioning equipment and the power phase sequence balance relationship to analyze the abnormal power phase sequence. According to the analysis results, construct a power fault analysis model for analyzing the power supply faults of the air conditioning equipment. Through the power fault analysis model, perform power fault analysis on the power phase data under the power supply fault state of the air conditioning equipment to obtain the power fault detection data corresponding to the current power fault phase.

[0004] In the technical field of fault detection, although there is a technical solution of obtaining the power phase data under the power supply fault state of the air conditioning equipment, inputting the power phase data into the power fault analysis model for power fault analysis, and obtaining the power fault detection data corresponding to the current power fault phase, in the scenario of detecting power supply faults of air conditioners in a data center, when the power supply of the air conditioning equipment fails, it may cause high-density and highly reliable data center servers to shut down or reduce performance due to overheating. At the same time, due to the short fault recovery time and small influence range, accurate early warning is required, and the existing power fault detection methods are difficult to achieve faster, more reliable, and more accurate positioning;

[0005] To solve the above problems, the present invention proposes a method for detecting power supply faults of air conditioning equipment. This method collects data by installing sensing devices at the air conditioner, fuses the fault data and air conditioning data with a data fusion algorithm, and establishes a prediction model to infer the fault probability, and finally issues an early warning with a linkage response mechanism. Summary of the Invention

[0006] In view of the above existing problems, a method for detecting power supply faults of air conditioning equipment is proposed.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problem is: a power supply fault detection method for air conditioning equipment, comprising:

[0008] By deploying power monitoring and status detection modules in air conditioning equipment, power and operation data of the air conditioning equipment can be collected in real time, and the working status of the air conditioning equipment can be comprehensively monitored, providing data support for fault diagnosis and prediction.

[0009] The collected real-time operating data of air conditioning equipment and historical fault data are fused together using data fusion technology to form a comprehensive dataset. Then, an association rule mining algorithm is used to reveal the correlation between equipment performance and power failure.

[0010] Based on deep learning methods, a power failure prediction model is established. By analyzing the changing trends of historical failure data and current operating data, the probability of power failure of the equipment is inferred, and fault warning information is provided to equipment maintenance personnel.

[0011] When the fault prediction model identifies a power failure risk that reaches a preset threshold, it automatically issues a warning signal through a linkage response mechanism, notifies the operator through a preset notification mechanism, and triggers the equipment's adaptive adjustment mechanism to maintain equipment operation when a fault risk occurs.

[0012] As a preferred embodiment, the specific steps for real-time acquisition of power data and operating data of the air conditioning equipment are as follows:

[0013] A power monitoring module is deployed on the power input side of the air conditioning equipment. The power monitoring module is connected in parallel with the power input terminal of the air conditioning equipment. The power monitoring module includes a voltage sensor, a current sensor, a power sensor, and a frequency sensor. A status detection module is installed at the functional components of the air conditioning equipment. The status detection module includes a temperature sensor, a pressure sensor, a speed sensor, and a flow sensor.

[0014] As a preferred embodiment, the specific steps for real-time acquisition of power data and operating data of the air conditioning equipment further include:

[0015] The power monitoring module and the status detection module are electrically connected to the main control board of the air conditioning equipment. The collected power data and operating data are periodically sent to the main control board in digital form through the CAN bus interface. Parameters are read from the power monitoring module and the status detection module in sequence according to the preset sampling period, and a timestamp and device identification information are added to each set of data.

[0016] As a preferred embodiment, the specific steps of data fusion using data fusion technology are as follows:

[0017] First, real-time operating data and historical power failure data of air conditioning equipment are aligned and marked. Using equipment identification and timestamp as a joint index, the operating data of the same equipment on the same time axis is matched with its corresponding power failure record. The data sample corresponding to the time of power failure is marked. According to the preset time window, the sampling period before the power failure occurs and the sampling period after the failure ends are marked as the pre-failure stage, the failure stage, and the failure recovery stage. The remaining time period is marked as the normal stage. The operating status of different failure stages is explicitly distinguished in the same dataset. The operating data within the continuous sampling period is merged into a time window sample. Statistical characteristics and change characteristics of each parameter within the window are calculated. The characteristics are combined into a multi-dimensional feature vector according to a fixed field order and the corresponding failure stage label and power failure type label are attached to form a comprehensive data sample for power failure analysis.

[0018] As a preferred implementation, the specific steps of using an association rule mining algorithm to reveal the correlation between equipment performance and power failure are as follows:

[0019] The continuous features in the comprehensive data sample are discretized. For numerical parameters, they are divided into multiple interval levels according to statistical distribution. Each feature vector is converted into a symbolic set of feature terms, where each feature term consists of a parameter name plus an interval level. At the same time, the fault stage label of the sample is retained, forming a transaction dataset suitable for association rule mining.

[0020] An association rule mining algorithm is executed on the transaction dataset, using the power failure stage label as a consequent candidate and the multi-parameter combination state as an antecedent candidate. Minimum support and minimum confidence thresholds are set to mine association rules that meet the conditions. The support formula is:

[0021] ,

[0022] Where X represents the antecedent candidate, Y represents the consequent candidate, t represents the transaction, and D represents the aggregated dataset;

[0023] The confidence formula is: ;

[0024] As a preferred embodiment, the specific steps for establishing the power failure prediction model are as follows:

[0025] Training samples are constructed based on a comprehensive dataset. Air conditioning equipment operation data within a continuous sampling period is selected as input, using time windows as units. A power failure prediction model is established using deep learning methods. The model input receives a multi-dimensional feature sequence arranged in chronological order. Internally, a multi-layer LSTM network is used to extract features from the joint changing trends of power supply parameters and equipment performance parameters. An attention mechanism layer is added to strengthen the focus on key time segments before a failure. The model output layer uses a Sigmoid activation function to output the probability value of power failure risk. During the training phase, labeled samples in the comprehensive dataset are divided into training and validation sets. Power failure labels are used as supervision signals. Cross-entropy loss function and gradient descent optimization algorithms are used to iteratively update the network parameters until the loss on the validation set converges. The loss function formula is:

[0026] ,

[0027] Where i represents the sample label and N represents the total number of samples. This represents the label of the i-th sample; a value of 1 indicates a fault, and a value of 0 indicates normal operation. This represents the failure probability of the i-th sample;

[0028] As a preferred implementation, the specific steps for analyzing the changing trends of historical fault data and current operating data to infer the probability of a power supply failure in the equipment are as follows:

[0029] The trained power failure prediction model is embedded in the main control board of the air conditioning equipment. During the actual operation of the air conditioning equipment, the main control board constructs the feature sequence of the current time period in real time according to the same time window and feature organization method as the training phase and inputs it into the power failure prediction model. The model outputs the power failure probability value corresponding to the current time window.

[0030] As a preferred embodiment, the specific steps for automatically issuing a warning signal through the linkage response mechanism are as follows:

[0031] When the power failure prediction model outputs the power failure risk probability value p corresponding to the current time window in each prediction period, according to the pre-set multi-level risk threshold table, the first risk threshold is set to 0.3, the second risk threshold is set to 0.6 and the third risk threshold is set to 0.8, which correspond to three risk levels: attention level, warning level and high risk level, respectively. After each new risk value is obtained, it is compared with the thresholds at each level and combined with the risk level at the previous moment. The current risk level is determined by state machine logic with hysteresis interval to avoid frequent switching caused by the probability value fluctuating around the threshold.

[0032] As a preferred implementation, the specific steps of informing the operator through a preset notification mechanism and simultaneously triggering the device's adaptive adjustment mechanism are as follows:

[0033] Based on the current risk level, a linkage event object is generated. For events reaching the warning level or above, the event object is submitted to the preset notification mechanism for distribution. The power failure risk information is sent to the operator by popping up alarm information on the local display screen and sending alarm messages to the operation and maintenance platform. At the same time as completing the warning notification, the adaptive adjustment mechanism is invoked according to the risk level to adjust the operating parameters of the air conditioning equipment in stages. The adaptive adjustment mechanism has a pre-set control strategy table corresponding to the risk level. The control strategy is converted into control commands that can be recognized by the air conditioning main control board and sent to the compressor drive circuit, fan drive circuit and actuator control circuit through the internal control bus. Without relying on hardware protection actions, the equipment operating status is adjusted in advance based on the predicted power failure risk.

[0034] Beneficial effects

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] 1. By unifying the modeling and analysis of power supply anomalies and changes in air conditioning cooling performance, when a power supply to a certain air conditioning system in a data center experiences a short-term drop, fluctuation, or local failure, the specific air conditioning equipment affected and its operating status can be quickly identified. This avoids the problem of traditional methods that can only make a rough judgment but cannot accurately pinpoint the specific equipment.

[0037] 2. By aligning real-time operating data with historical fault data and combining association rule mining algorithms, implicit correlation patterns between multi-parameter combinations and power supply faults can be extracted. When the power supply is in the pre-fault stage, potential risks can be identified based on the combination of operating characteristics, enabling early detection of short-term, small-scale power supply faults. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0039] Figure 1 This is a flowchart illustrating the present invention;

[0040] Figure 2 This is a comparison diagram of the effects of the present invention and the prior art, where gray bars represent the prior art and black bars represent the present invention. Detailed Implementation

[0041] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0042] Example 1:

[0043] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a method for detecting power supply faults in air conditioning equipment, the method comprising the following steps:

[0044] Step S1: Install sensor equipment to collect air conditioning data;

[0045] Step S2: Merge fault data with air conditioning data;

[0046] Step S3: Establish a predictive model to infer the probability of failure;

[0047] Step S4: Issue an early warning through the linkage response mechanism.

[0048] This method is implemented in the order of S1–S4, and its overall process is as follows:

[0049] By deploying power monitoring and status detection modules in air conditioning equipment, power and operation data of the air conditioning equipment can be collected in real time, and the working status of the air conditioning equipment can be comprehensively monitored, providing data support for fault diagnosis and prediction.

[0050] The collected real-time operating data of air conditioning equipment and historical fault data are fused together using data fusion technology to form a comprehensive dataset. Then, an association rule mining algorithm is used to reveal the correlation between equipment performance and power failure.

[0051] Based on deep learning methods, a power failure prediction model is established. By analyzing the changing trends of historical failure data and current operating data, the probability of power failure of equipment is inferred, providing accurate fault warning information for equipment maintenance personnel.

[0052] When the fault prediction model identifies a power failure risk that reaches a preset threshold, it automatically issues a warning signal through a linkage response mechanism, notifies the operator through a preset notification mechanism, and triggers the equipment's adaptive adjustment mechanism to maintain normal equipment operation when a fault risk occurs.

[0053] Combination Figure 1 The specific implementation methods and effects of steps S1–S4 are described respectively, including the specific steps of installing sensing devices to collect air conditioning data as follows:

[0054] The specific steps for installing sensor equipment to collect air conditioning data are as follows:

[0055] A power monitoring module is deployed on the power input side of the air conditioning equipment. The power monitoring module is connected in parallel with the power input terminal of the air conditioning equipment. The power monitoring module includes a voltage sensor, a current sensor, a power sensor, and a frequency sensor. Status detection modules are installed at key functional components of the air conditioning equipment, such as the compressor and indoor fan motor. The status detection modules include temperature sensors, pressure sensors, speed sensors, and flow sensors.

[0056] The power monitoring module and the status detection module are electrically connected to the main control board of the air conditioning equipment. The collected power data and operating data are periodically sent to the main control board in digital form through the CAN bus interface. According to the preset sampling period, such as 500ms, the current voltage value, current value, active power, fan speed, heat exchanger temperature, ambient temperature, system pressure and operating mode flag are read from the power monitoring module and the status detection module in sequence. Each set of data is appended with a timestamp and equipment identification information.

[0057] The specific steps for integrating fault data and air conditioning data are as follows:

[0058] Real-time operating data, including key parameters such as voltage, current, power, fan speed, temperature, and pressure, is read from the power monitoring module and status detection module of the air conditioning equipment. At the same time, historical fault data is combined with real-time data, such as the time, frequency, type, and affected equipment components of power faults, and merged into a comprehensive dataset through data fusion technology.

[0059] Specifically, the data fusion technology is used to merge the data into a comprehensive dataset. First, the real-time operation data and historical power failure data of the air conditioning equipment are aligned and marked. Using the equipment identifier and timestamp as a joint index, the operation data of the same equipment on the same time axis are matched with its corresponding power failure record. The data sample corresponding to the time of the power failure is marked. According to the preset time window, the sampling period before the power failure occurs and the sampling period after the failure ends are marked as the pre-failure stage, the failure stage, and the failure recovery stage. The remaining time period is marked as the normal stage. The operation status of different failure stages is explicitly distinguished in the same dataset. The operation data within the continuous sampling period is merged into a time window sample. Statistical characteristics and change characteristics of each parameter within the window are calculated, such as average, standard deviation, slope, fluctuation amplitude, number of steps, etc. The features are combined into a multi-dimensional feature vector according to a fixed field order and the corresponding failure stage label and power failure type label are attached to form a comprehensive data sample for power failure analysis.

[0060] The continuous features in the comprehensive data sample are discretized. Numerical parameters such as voltage, current, power, temperature, pressure, and rotational speed are divided into multiple interval levels according to their statistical distribution, such as low voltage, normal voltage, and high voltage. Each feature vector is converted into a symbolic set of feature terms, where each feature term consists of a parameter name and an interval level. The fault stage label of the sample is also retained, forming a transaction dataset suitable for association rule mining. An association rule mining algorithm is executed on the transaction dataset. For example, this invention preferably uses the Apriori association rule mining algorithm. The Apriori association rule mining algorithm is an algorithm that utilizes the prior property that a subset of a frequent itemset is necessarily also a frequent itemset, expanding layer by layer and filtering based on support and confidence to automatically mine the association relationships between feature terms from the sample data. Using the power fault stage label as a consequent candidate and the multi-parameter combination state as an antecedent candidate, minimum support and minimum confidence thresholds are set to mine association rules that meet the conditions. The support formula is:

[0061] ,

[0062] Where X represents the antecedent candidate, Y represents the consequent candidate, t represents the transaction, and D represents the aggregated dataset;

[0063] The confidence formula is: ;

[0064] The specific steps for establishing a predictive model to infer the probability of failure are as follows:

[0065] Training samples are constructed based on a comprehensive dataset. Air conditioning equipment operation data within a continuous sampling period is selected as input, using time windows as units. A power failure prediction model is established using deep learning methods. Preferably, the power failure prediction model adopts a temporal neural network structure based on a Long Short-Term Memory (LSTM) network. The model input receives a multi-dimensional feature sequence arranged in chronological order. Internally, a multi-layer LSTM network is used to extract features from the joint changing trends of power supply parameters and equipment performance parameters. An attention mechanism layer is added to strengthen the focus on key time segments before the failure. The model output layer uses a Sigmoid activation function to output the probability value of power failure risk. During the training phase, labeled samples in the comprehensive dataset are divided into training and validation sets. Power failure labels are used as supervision signals. Cross-entropy loss function and gradient descent optimization algorithms are used to iteratively update the network parameters until the loss on the validation set converges. The loss function formula is:

[0066] ,

[0067] Where i represents the sample label and N represents the total number of samples. This represents the label of the i-th sample; a value of 1 indicates a fault, and a value of 0 indicates normal operation. This represents the failure probability of the i-th sample;

[0068] The trained power failure prediction model is embedded in the main control board of the air conditioning equipment. During the actual operation of the air conditioning equipment, the main control board constructs the feature sequence of the current time period in real time according to the same time window and feature organization method as the training phase and inputs it into the power failure prediction model. The model outputs the power failure probability value corresponding to the current time window.

[0069] The specific steps for issuing an early warning through the linkage response mechanism are as follows:

[0070] When the power failure prediction model outputs the power failure risk probability value p corresponding to the current time window in each prediction period, according to the pre-set multi-level risk threshold table, the first risk threshold is set to 0.3, the second risk threshold is set to 0.6 and the third risk threshold is set to 0.8, which correspond to three risk levels: attention level, warning level and high risk level, respectively. After each new risk value is obtained, it is compared with the thresholds at each level and combined with the risk level at the previous moment. The current risk level is determined by state machine logic with hysteresis interval to avoid frequent switching caused by the probability value fluctuating around the threshold.

[0071] Based on the current risk level, a linkage event object is generated. This event object includes fields such as device identifier, timestamp, power failure risk probability value, risk level, and trigger source. For events reaching the warning level or above, the event object is submitted to a preset notification mechanism for distribution. This preset notification mechanism includes a human-machine interface display module, a remote monitoring platform interface, and a message push interface. Power failure risk information is sent to operators by displaying alarm messages on the local screen and sending alarm messages to the operation and maintenance platform. Simultaneously with the warning notification, an adaptive adjustment mechanism is invoked according to the risk level to adjust the operating parameters of the air conditioning equipment in a tiered manner, prioritizing optimal settings. The adaptive adjustment mechanism pre-sets a control strategy table corresponding to the risk level. The control strategy includes adjustment rules for compressor operating frequency, outdoor fan speed, indoor fan speed, target temperature setpoint, and compressor start-stop mode. For example, at the concern level, the upper limit of compressor frequency is appropriately lowered, and frequent start-stop in a short period of time is restricted. The above control strategy is converted into control commands that can be recognized by the air conditioner main control board and sent to the compressor drive circuit, fan drive circuit, and actuator control circuit through the internal control bus. Without relying on traditional overcurrent, undervoltage, and other hardware protection actions, the equipment operating status is adjusted in advance based on the predicted power failure risk.

[0072] Unlike existing protection methods that directly trip and shut down the circuit breaker when voltage or current exceeds the limit, the linkage response mechanism of this embodiment uses the risk probability output by the power supply fault prediction model as the core decision quantity. It links the notification mechanism and the adaptive adjustment mechanism in the same decision logic, realizing an integrated response from prediction to early warning to hierarchical control. Thus, while maintaining basic operation, it performs active avoidance control in conjunction with the prediction results, forming a power supply fault linkage response feature that is different from the existing single over-limit protection action.

[0073] like Figure 2 The graph shows the effect comparison of a power failure detection method for air conditioning equipment. The horizontal axis lists the key performance indicators, and the vertical axis represents the exemplified performance scores, ranging from 0 to 100. The higher the value, the better the performance. The purpose is to intuitively demonstrate the expected improvement of the present invention in key capabilities compared to typical prior art.

[0074] Example 2:

[0075] Based on the above embodiment 1, a power supply fault detection method for air conditioning equipment in a data center power supply fault detection scenario is as follows:

[0076] Step 1: Install a power monitoring device in the distribution box that supplies power to the air conditioning equipment. This device is coupled to the power supply of each air conditioning circuit via a current transformer. It integrates multiple voltage sensors, Hall current sensors, and energy metering chips to synchronously collect power-side parameters such as phase voltage, current, active power, reactive power, power factor, and grid frequency. On the equipment side, temperature sensors, pressure sensors, speed sensors, and flow sensors are respectively made into status detection sub-nodes and fixed to key parts such as the compressor housing, indoor fan motor end, and heat exchanger inlet and outlet pipes. The collected operating data is sent to the edge acquisition gateway via short-range wireless means. The data is collected periodically according to a preset sampling period, such as 200ms. Simultaneously, sampling is triggered when events such as voltage drop or current change are detected, and the sorted structured data frame is sent to the upper main control board to realize the synchronous acquisition and centralized management of air conditioning power data and operating status data.

[0077] Step 2: Based on the operating conditions of the air conditioning equipment, the continuous operation process is divided into operating condition segments according to events such as compressor start-up and shutdown, operation mode switching, and target temperature changes. Each operating condition segment corresponds to a relatively complete cooling or heating process. The equipment identification and the start and end time of the operating condition segment are matched with historical power failure records to determine whether the operating condition segment spans the period of power failure. Based on this, each operating condition segment is assigned three types of labels: normal operating condition, pre-fault operating condition, and fault operating condition. At the granularity of the operating condition segment, segment-level features are calculated for parameters such as voltage, current, power, temperature, pressure, and speed, including average value, extreme value, number of fluctuations, number of voltage drops, and overcurrent duration. The segment-level features, along with the corresponding operating condition labels and power failure types, are stored in the comprehensive dataset.

[0078] Step 3: Based on the comprehensive dataset, training samples are constructed in units of time windows. The power supply-side sequence features such as voltage, current, and power within each time window are arranged chronologically to form the first input sequence. The segment-level statistical features within the same time window are arranged to form the second input vector, creating a dual-channel input structure. Preferably, a temporal deep learning network based on a Transformer encoder is used as the power supply fault prediction model. The first input sequence is positionally encoded and then input into a multi-layer self-attention encoder to learn the long-term dependencies and abrupt change patterns of power supply-side parameters in the time dimension. The second input vector undergoes nonlinear mapping through a multi-layer fully connected network to extract device performance features. The two feature paths are then concatenated in the feature fusion layer and fed into subsequent fully connected layers. The output layer uses a Sigmoid activation function to provide the power supply fault risk probability for each time window. During the training phase, power supply fault labels are used as supervision signals. Samples are divided into training and validation sets. A binary cross-entropy loss function and a gradient descent-based optimization algorithm are used to update the network parameters until the validation set converges.

[0079] Step 4: While the power failure prediction model outputs the power failure risk probability value corresponding to the current time window in each prediction period, the linkage response mechanism will calculate the average risk and rate of change in the most recent prediction periods, generate a comprehensive risk index, and adaptively generate a risk threshold range based on the statistical results within the sliding time window. The comprehensive risk index is divided into three levels: attention level, early warning level, and high risk level according to the quantile method. The risk classification boundary is dynamically adjusted according to the air conditioning model, seasonal load, and power grid fluctuations. After obtaining the current risk level, the edge control node uniformly generates linkage event objects. For events that reach the early warning level or above, grouped alarms are generated on the monitoring interface according to the loop dimension. Based on the number of high-risk devices and the comprehensive load in each loop, a graded peak shaving strategy is selected from the strategy library, and different operating constraints are issued to the air conditioning equipment in different loops. For example, multiple air conditioners in a high-risk loop are operated in a staggered and frequency-reduced manner.

[0080] The embodiments of the present invention described above are subject to modification and change of method by those skilled in the art without departing from the embodiments and broader aspects of the present invention. The appended claims are intended to include all such modifications and changes of method that do not depart from the present invention.

Claims

1. A method for detecting power supply faults in air conditioning equipment, characterized in that, include: By deploying power monitoring and status detection modules in air conditioning equipment, power and operation data of the air conditioning equipment can be collected in real time, and the working status of the air conditioning equipment can be comprehensively monitored, providing data support for fault diagnosis and prediction. The collected real-time operating data of air conditioning equipment and historical fault data are fused together using data fusion technology to form a comprehensive dataset. Then, an association rule mining algorithm is used to reveal the correlation between equipment performance and power failure. Based on deep learning methods, a power failure prediction model is established. By analyzing the changing trends of historical failure data and current operating data, the probability of power failure of the equipment is inferred, and fault warning information is provided to equipment maintenance personnel. When the fault prediction model identifies a power failure risk that reaches a preset threshold, it automatically issues a warning signal through a linkage response mechanism, notifies the operator through a preset notification mechanism, and triggers the equipment's adaptive adjustment mechanism to maintain equipment operation when a fault risk occurs.

2. The power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for real-time acquisition of power and operating data of the air conditioning equipment are as follows: A power monitoring module is deployed on the power input side of the air conditioning equipment. The power monitoring module is connected in parallel with the power input terminal of the air conditioning equipment. The power monitoring module includes a voltage sensor, a current sensor, a power sensor, and a frequency sensor. A status detection module is installed at the functional components of the air conditioning equipment. The status detection module includes a temperature sensor, a pressure sensor, a speed sensor, and a flow sensor.

3. The power supply fault detection method for an air conditioning device according to claim 2, characterized in that, The specific steps for real-time acquisition of power and operating data of air conditioning equipment also include: The power monitoring module and the status detection module are electrically connected to the main control board of the air conditioning equipment. The collected power data and operating data are periodically sent to the main control board in digital form through the CAN bus interface. Parameters are read from the power monitoring module and the status detection module in sequence according to the preset sampling period, and a timestamp and device identification information are added to each set of data.

4. The power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for data fusion using the aforementioned data fusion technology are as follows: First, real-time operating data and historical power failure data of air conditioning equipment are aligned and labeled. Using equipment identification and timestamp as a joint index, the operating data of the same equipment on the same time axis is matched with its corresponding power failure record. The data sample corresponding to the time of power failure is marked. According to the preset time window, the sampling period before the power failure occurs and the sampling period after the failure ends are marked as the pre-failure stage, the failure stage, and the failure recovery stage. The remaining time period is marked as the normal stage. The operating status of different failure stages is explicitly distinguished in the same dataset. The operating data within the continuous sampling period is merged into a time window sample. Statistical features and change features of each parameter within the window are calculated. The features are combined into a multi-dimensional feature vector according to a fixed field order and the corresponding failure stage label and power failure type label are attached to form a comprehensive data sample for power failure analysis.

5. The power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for using association rule mining algorithms to reveal the correlation between equipment performance and power failure are as follows: The continuous features in the comprehensive data sample are discretized. For numerical parameters, they are divided into multiple interval levels according to statistical distribution. Each feature vector is converted into a symbolic set of feature terms, where each feature term consists of a parameter name plus an interval level. At the same time, the fault stage label of the sample is retained, forming a transaction dataset suitable for association rule mining.

6. The power supply fault detection method for an air conditioning device according to claim 5, characterized in that, The specific steps for using association rule mining algorithms to reveal the correlation between equipment performance and power failures also include: An association rule mining algorithm is executed on the transaction dataset, using the power failure stage label as a consequent candidate and the multi-parameter combination state as an antecedent candidate. Minimum support and minimum confidence thresholds are set to mine association rules that meet the conditions. The support formula is: ， Where X represents the antecedent candidate, Y represents the consequent candidate, t represents the transaction, and D represents the aggregated dataset; The confidence formula is: .

7. The power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for establishing the power failure prediction model are as follows: Training samples are constructed based on a comprehensive dataset. Air conditioning equipment operation data within a continuous sampling period is selected as input, using time windows as units. A power failure prediction model is established using deep learning methods. The model input receives a multi-dimensional feature sequence arranged in chronological order. Internally, a multi-layer LSTM network is used to extract features from the joint changing trends of power supply parameters and equipment performance parameters. An attention mechanism layer is added to strengthen the focus on key time segments before a failure. The model output layer uses a Sigmoid activation function to output the probability value of power failure risk. During the training phase, labeled samples in the comprehensive dataset are divided into training and validation sets. Power failure labels are used as supervision signals. Cross-entropy loss function and gradient descent optimization algorithms are used to iteratively update the network parameters until the loss on the validation set converges. The loss function formula is: ， Where i represents the sample label and N represents the total number of samples. This represents the label of the i-th sample; a value of 1 indicates a fault, and a value of 0 indicates normal operation. Let represent the failure probability of the i-th sample.

8. The power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for analyzing the trends in historical fault data and current operating data to infer the probability of a power supply failure in the equipment are as follows: The trained power failure prediction model is embedded in the main control board of the air conditioning equipment. During the actual operation of the air conditioning equipment, the main control board constructs the feature sequence of the current time period in real time according to the same time window and feature organization method as in the training phase and inputs it into the power failure prediction model. The model outputs the probability value of power failure occurrence corresponding to the current time window.

9. A power supply fault detection method for an air conditioning device according to claim 1, characterized in that, The specific steps for automatically issuing early warning signals through the linkage response mechanism are as follows: When the power failure prediction model outputs the power failure risk probability value p corresponding to the current time window in each prediction period, according to the pre-set multi-level risk threshold table, the first risk threshold is set to 0.3, the second risk threshold is set to 0.6, and the third risk threshold is set to 0.8, corresponding to three risk levels: attention level, warning level, and high risk level, respectively. After obtaining a new risk value each time, it is compared with the thresholds at each level and combined with the risk level at the previous moment. The current risk level is determined by state machine logic with hysteresis interval to avoid frequent switching caused by the probability value fluctuating around the threshold.

10. A method for detecting power supply faults in an air conditioning device according to claim 1, characterized in that, The specific steps for informing operators through a preset notification mechanism and simultaneously triggering the device's adaptive adjustment mechanism are as follows: Based on the current risk level, a linkage event object is generated. For events reaching the warning level or above, the event object is submitted to the preset notification mechanism for distribution. The power failure risk information is sent to the operator by popping up alarm information on the local display screen and sending alarm messages to the operation and maintenance platform. At the same time as completing the warning notification, the adaptive adjustment mechanism is invoked according to the risk level to adjust the operating parameters of the air conditioning equipment in stages. The adaptive adjustment mechanism has a pre-set control strategy table corresponding to the risk level. The control strategy is converted into control commands that can be recognized by the air conditioning main control board and sent to the compressor drive circuit, fan drive circuit and actuator control circuit through the internal control bus. Without relying on hardware protection actions, the equipment operating status is adjusted in advance based on the predicted power failure risk.

Images