Early fault detection method for heating and ventilation equipment based on abnormal response of control loop

Abstract

The present application relates to the technical field of heating and ventilation equipment, and particularly relates to a heating and ventilation equipment early fault detection method based on control loop response abnormality, comprising the following steps: S1: collecting field signals of the control loop, wherein the field signals at least include valve position control instructions, valve position feedback original values and controlled process quantities, and the valve position feedback original values are normalized; the present application performs intelligent cleaning and credible fusion on the valve position feedback signals to eliminate abnormal data points; an adaptive time-varying time delay alignment technology is adopted to establish an accurate dynamic response model online to accurately extract system parameters; finally, layered decision is implemented by evaluating data quality and excitation sufficiency. The present application effectively distinguishes feedback link abnormality and equipment loop fault, and significantly improves the accuracy and reliability of early fault detection of valve sticking, performance degradation and the like under signal interference.

Description

Technical Field

[0001] This invention relates to the field of HVAC equipment technology, and more specifically to an early fault detection method for HVAC equipment based on abnormal control loop response. Background Technology

[0002] HVAC systems, as a crucial component of building electromechanical systems, are widely used in office buildings, hospitals, data centers, and industrial plants. Their operational status directly impacts indoor thermal and humidity environments, energy consumption levels, and equipment reliability. With the increasing scale and complexity of control strategies, chillers, air handling units, terminal coils, and various water and air valves operate under variable loads and frequent adjustments for extended periods. Equipment performance degradation and failures often exhibit characteristics such as high concealment, slow development, and weak early warning signs. Traditional maintenance methods rely heavily on periodic inspections or post-failure repairs, which can easily lead to delayed maintenance, incorrect maintenance, or persistent abnormal energy consumption. Therefore, there is an urgent need for a technology capable of online identification and early warning in the early stages of failure.

[0003] In modern HVAC control systems, process variables such as temperature, humidity, pressure, flow rate, and valve opening are typically collected by sensors and input to an automatic controller. The controller outputs a control quantity based on the deviation between the setpoint and the feedback value, driving the actuator to change the valve or damper opening, thereby regulating the controlled object. Particularly in the acquisition of valve position feedback signals from chilled water regulating valve actuators, the purpose is to obtain a direct representation of the actual valve opening. This is used to cross-reference the opening command output by the controller with changes in the controlled quantity, thereby determining whether the valve is acting according to the command, whether there is jamming or hysteresis, and whether early signs of deterioration such as slowing down the loop dynamics are present. This feedback signal plays a crucial bridging role in early fault detection of HVAC equipment, connecting control commands and process responses. It enables algorithms to establish a correspondence between the actual valve execution state and changes in the system's thermal response, thus distinguishing abnormal loop responses from complex operating condition disturbances.

[0004] However, in high-humidity environments or under conditions where the surface temperature near the valve body is low, condensation may occur inside the actuator and seep into the potentiometer-type valve position feedback mechanism. This causes transient instability in the contact resistance at the potentiometer contact interface due to the influence of liquid film and contamination. The sliding arm may experience intermittent contact within a certain angle range, resulting in short-term jumps and spikes in the feedback signal, while the valve's mechanical position does not undergo a corresponding abrupt change. This directly undermines the accuracy of existing fault detection methods in judging the control loop response. On one hand, spikes may be misinterpreted as rapid valve action or loop oscillation, leading to false alarms and contaminating the response characteristics. On the other hand, spikes may dominate residual and characteristic quantity changes, masking the true slow degradation trend and causing missed detections.

[0005] Therefore, a method for early fault detection of HVAC equipment based on abnormal control loop response is proposed to address the aforementioned problems. Summary of the Invention

[0006] Technical problems to be solved:

[0007] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an early fault detection method for HVAC equipment based on abnormal control loop response. It can effectively solve the problem of false alarms and missed alarms caused by burrs in valve position feedback signals of HVAC equipment due to moisture in the existing technology, and achieve the technical objective of accurately and reliably detecting early faults in the control loop even when the feedback signal is contaminated.

[0008] Technical solution:

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] This invention provides a method for early fault detection of HVAC equipment based on abnormal control loop response, comprising the following steps:

[0011] S1: Acquire field signals from the control loop, the field signals including at least valve position control commands, valve position feedback raw values ​​and controlled process quantities, and normalize the valve position feedback raw values.

[0012] S2: The acquired field signals are smoothed and filtered to suppress noise and obtain a smooth signal sequence that represents the trend of physical change.

[0013] S3: Calculate the difference between the smoothed signal sequence in adjacent sampling periods to characterize the dynamic characteristics of the control action and the object response;

[0014] S4: Based on the smoothed value of the valve position feedback, perform at least the following logical checks: amplitude range check, rate of change speed limit check, and direction consistency check, to generate a single-point usability flag characterizing the reliability of the valve position feedback data; wherein, the rate of change speed limit check is used to determine whether the change in the smoothed value of the valve position feedback within a single sampling period exceeds the upper limit of the actuator speed determined by the maximum physical speed of the actuator; the direction consistency check is used to determine whether the direction of change of the smoothed value of the valve position feedback is consistent with the direction of change of the smoothed value of the valve position control command within a preset fault tolerance range;

[0015] S5: Based on the smoothed value of the valve position control command and the physical speed constraint of the actuator, construct and update the physically realizable predicted valve position;

[0016] S6: Based on the single-point availability flag, perform reliable fusion and smooth switching between the smooth value of the valve position feedback and the predicted valve position, output a continuous and physically reliable clean valve position signal, and calculate the feedback link anomaly index based on the single-point availability flag.

[0017] S7: Time-varying delay alignment is performed on the cleaning valve position signal and the controlled process quantity, and a response model describing the dynamic relationship between the cleaning valve position and the controlled process quantity is established online based on the aligned signal; wherein, time-varying delay alignment refers to performing mean-removal processing on the cleaning valve position signal and the controlled process quantity to focus on dynamic changes, then estimating the optimal integer delay by calculating the normalized cross-correlation score within a sliding window, and further obtaining the fractional delay through parabolic interpolation to achieve subsampling-level accurate alignment; then, the first-order ARX model parameters are updated online using this aligned signal; this is used to avoid misjudging delay drift as abnormal system parameters;

[0018] S8: Extract monitoring indicators for characterizing the health status of the loop from the online identification results of the response model. The monitoring indicators include at least the equivalent static gain, the equivalent time constant, and the model residual energy.

[0019] S9: Based on the feedback link anomaly index and the excitation sufficiency of the cleaning valve position signal, generate a data quality availability flag and an excitation availability flag, and control the update of the response model and the output logic of the monitoring index accordingly; wherein, the excitation sufficiency is used to quantify whether the valve position change within the most recent window is sufficient;

[0020] S10: Output the feedback link anomaly index and the monitoring index together, and determine the early fault of the HVAC equipment control loop based on the continuous change trend of the monitoring index.

[0021] Further, step S1 involves normalizing the original valve position feedback value, including: In the formula, k is the discrete-time index; This represents the original valve position feedback value collected at time k; This represents the normalized valve opening at time k; This is the lower limit of the calibration when the valve corresponding to the feedback channel is fully closed; This is the upper limit of the calibration when the valve corresponding to the feedback channel is fully open; This is a clipping function.

[0022] Furthermore, the smoothing filtering process in step S2 employs a first-order low-pass filter, including: In the formula, The original observation value at the current time k; Let k be the filtered signal value at time k; The signal value after filtering by k-1 at the previous time step; 'a' is the weighting coefficient; 'x' is the input quantity. y represents the valve opening command; y represents the controlled process variable. This is the normalized valve position opening.

[0023] Furthermore, the calculation formula for the rate of change limit verification in step S4 is as follows: In the formula, This indicates the result of the validity determination of the rate of change at time k; This indicates the upper limit of the maximum achievable rate of change of the actuator at the normalized valve position scale; The time interval between adjacent sampling points; This represents the change in the valve position feedback smoothing value in the kth sampling period relative to the previous sampling period.

[0024] Furthermore, the construction and updating of the predicted valve position in step S5 includes the following sub-steps:

[0025] S501: Calculate the following error between the smoothed value of the valve position control command and the predicted valve position at the previous moment;

[0026] S502: Calculate the expected increment based on the following error and a following coefficient;

[0027] S503: Based on the actuator's maximum speed limit and the sampling time interval, calculate the maximum allowable change in valve position within a single cycle;

[0028] S504: Limit the desired increment by the maximum permissible change in the valve position to obtain the achievable valve position increment;

[0029] S505: Utilize the method to update the predicted valve position at the current moment using the valve position increment.

[0030] Furthermore, the formula for updating the predicted valve position in step S505 is as follows:

[0031] In the formula, This represents the prediction threshold for the k-th sampling period; This represents the valve position increment that can be achieved within the k-th sampling period.

[0032] Further, step S6, outputting the cleaning valve position signal, includes:

[0033] According to the fusion state and transition weight Calculate the cleaning valve position signal ;

[0034] When the fusion status indicator uses feedback:

[0035] ;

[0036] When the fusion state indicator uses prediction:

[0037] ;

[0038] In the formula, This represents the valve position feedback smoothing value at time k; This is the valve position after deviation compensation.

[0039] Further, the time-varying delay alignment in step S7 includes estimating the time-varying delay by calculating the normalized cross-correlation score between the cleaning valve position signal and the controlled process quantity within a sliding window. The formula for calculating the normalized cross-correlation score is as follows:

[0040] ;

[0041] In the formula, This represents the normalized cross-correlation score between the mean-demeaned component of the clean valve position and the mean-demeaned component of the controlled process quantity within the current window when the candidate lag is d; N is the statistical sliding window length; d is the pure lag step number. To remove the mean value component from the clean valve position; is the mean-removed component of the controlled process variable; k is the discrete-time index.

[0042] Furthermore, the formulas for calculating the equivalent static gain and equivalent time constant extracted in step S8 are as follows:

[0043] In the formula, This represents the equivalent static gain at time k; Represents the equivalent time constant at time k; The time interval between adjacent sampling points; Let be the autoregressive coefficient at time k; where This represents the input gain coefficient at time k.

[0044] Furthermore, the formula for calculating the feedback link anomaly index in step S6 is as follows:

[0045] ;

[0046] In the formula, This represents the average percentage of suspected glitch / intermittent contact feedback that is unavailable within the most recent time window of the kth sampling period; is a single-point availability flag; N is the statistical sliding window length.

[0047] Beneficial effects:

[0048] The technical solution provided by this invention has the following advantages compared with the prior art:

[0049] This invention initially suppresses noise through smoothing filtering, then performs three types of logic checks on the valve position feedback signal: amplitude range, rate of change limit, and directional consistency. A single-point usability flag is generated to identify reliable data points. Simultaneously, a predicted valve position trajectory is dynamically generated based on control commands and actuator physical constraints. Finally, by introducing hysteresis confirmation, slow-varying deviation compensation, and smooth transition weights, a seamless and continuous reliable fusion is achieved between the feedback signal and the predicted value, outputting a clean valve position signal. This effectively identifies and eliminates feedback link anomalies such as poor potentiometer contact and signal glitches caused by high humidity condensation, preventing these measurement artifacts from contaminating subsequent loop dynamic analysis. This fundamentally reduces the risk of false alarms and missed alarms caused by signal quality issues, improving the input reliability of the detection system.

[0050] Meanwhile, this scheme abandons the fixed delay assumption. First, it performs mean-reduction processing on the cleaning valve position and the controlled process quantity to focus on dynamic changes. Then, within a sliding window, it estimates the optimal integer time delay by calculating the normalized cross-correlation score, and further obtains the fractional time delay through parabolic interpolation to achieve subsampling-level accurate alignment. Using this aligned signal, the parameters of the first-order ARX model are updated online recursively. It can automatically adapt to the transmission delay drift caused by changes in system flow rate and velocity, solving the input-output misalignment problem caused by fixed time delay under varying operating conditions. This makes the established response model more realistically reflect the current dynamics of the system, avoiding misjudging delay drift as abnormal system parameters, thereby improving the trend accuracy of key health indicators such as equivalent time constant and static gain, and the sensitivity of early fault detection.

[0051] Furthermore, data quality is quantified by calculating feedback link anomaly indicators, and the identifiability of operating conditions is assessed by calculating the excitation intensity of the cleaning valve position. Based on this, data quality availability flags and excitation availability flags are generated. These flags are used for updating and freezing the gating online model, as well as the output logic of the final conclusion. The beneficial effects are: it enables the initial differentiation of fault sources; when feedback link anomaly indicators are high, measurement problems are prioritized; and model-based loop response anomaly conclusions are only output when the data is reliable and the excitation is sufficient. This avoids misjudging signal acquisition failures or steady-state lack of excitation as early equipment degradation, making the output conclusions more targeted and interpretable, effectively supporting subsequent accurate maintenance decisions. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0053] Figure 1 This is a flowchart illustrating the early fault detection method for HVAC equipment in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0055] The present invention will be further described below with reference to embodiments.

[0056] Example 1:

[0057] See appendix Figure 1 This case proposes an early fault detection method for HVAC equipment based on abnormal control loop response, including the following steps:

[0058] S1: Acquire the raw field signals related to the control loop under the same time reference, and complete the unit / range calibration and normalization to provide a directly calculable data stream for subsequent processing. The acquired raw field signals include:

[0059] Valve opening command This indicates the target opening command from the controller to the actuator of the two-way control valve. It is in percentage form and reflects the actual output command from the controller. It is used to determine whether the actuator responds according to the command.

[0060] Valve position feedback raw value This represents the raw analog reading of the actuator position feedback, commonly in the form of voltage or current. It serves as the basis for reconstructing the actual valve position and supports the judgment of feedback quality and the output of fault indicators, reflecting the health status of the position measurement link and the actuator feedback mechanism. The raw valve position feedback value is then converted into a dimensionless normalized valve opening for subsequent processing. In the formula, k is the discrete-time index; This represents the original valve position feedback value collected at time k; This represents the normalized valve position opening at time k, with a value range of [0, 1]. This is the lower limit of the calibration when the corresponding valve of the feedback channel is fully closed; This is the upper limit of the calibration when the valve corresponding to this feedback channel is fully open; This is a clipping function used to force the result to be limited to [0, 1]; that is... Where x is the input quantity.

[0061] Controlled process quantity It provides the actual response results of the controlled object; it is used to establish a dynamic response model from valve position to process quantity, calculate residuals and detect response anomalies, and reflect the control effect, dynamic characteristics of the object and its deterioration trend.

[0062] S2: The original field signals collected are smoothed and denoised to obtain a stable sequence that better represents the real physical change trend; In the formula, The original observation value at the current time k; Let k be the filtered signal value at time k; is the signal value after filtering k-1 at the previous time step; 'a' is the weighting coefficient, set based on experience.

[0063] S3: Perform adjacent sampling difference processing on the filtered signal values ​​to obtain the dynamic characteristics and direction relationship used to characterize the control action and the object response, i.e.:

[0064] In the formula, This represents the valve position feedback smoothing value at time k; This represents the valve position feedback smoothing value at time k-1; This represents the change in the valve position feedback smoothing value in the kth sampling period relative to the previous sampling period, used to characterize the actual degree of change on the valve feedback side within a sampling period; The smoothed value of the valve position control command at time k; This represents the smoothed value of the valve position control command at time k-1; It represents the change in the smoothed value of the valve position control command in the kth sampling period relative to the previous sampling period. It is used to characterize how much the controller requires the valve to change within a sampling period and to make a consistency judgment with the smoothed value of the valve position feedback, distinguishing between command-driven changes and abnormal feedback changes. This represents the smoothed value of the controlled process variable at time k; This represents the smoothed value of the controlled process variable at time k-1; It represents the change in the smoothed value of the controlled process variable in the kth sampling period relative to the previous sampling period. It is used to characterize the short-term response intensity and direction of the object's output, help determine whether the control action has triggered a reasonable response, and provide dynamic characteristics for subsequent processing and analysis.

[0065] S4: Based on the difference between the smoothed value of the valve position feedback and the smoothed value of the valve position control command, three types of logic checks are performed: amplitude range, rate of change limit, and direction consistency. These checks are used to identify and eliminate glitch / unreliable sampling points in the valve position feedback. The check results are then synthesized into a feedback availability flag, which is used to automatically select reliable data or switch to predictive substitution during subsequent fusion and modeling, thereby avoiding false alarms and improving fault detection reliability. Specifically:

[0066] S401: Based on the valve position feedback smoothing value, a value legitimacy is defined to quickly determine whether its amplitude falls within the normalized physical range, thereby marking out-of-bounds spikes / open-circuit anomalies and providing a basis for subsequent data removal or fusion, namely:

[0067] In the formula, The result of the amplitude validity judgment at time k is a binary flag used to mark whether the valve position feedback at the sampling point is reliable. When its value is 1, it means that the amplitude at the point is reliable and usable. When its value is 0, it means that the amplitude at the point is unreliable and needs to be removed / reduced in weight. It is used as one of the inputs for the subsequent comprehensive availability flag.

[0068] S402: Based on the change in valve position feedback smoothing value, a rate of change validity is defined to determine whether an impossible jump occurs within a single sampling period according to the actuator's maximum speed constraint, thereby marking glitch / intermittent contact anomalies and providing a basis for subsequent availability fusion.

[0069] In the formula, The result of the legality judgment of the rate of change at time k is a binary flag used to mark whether the feedback change at this sampling point meets the upper limit of the actuator speed; when its value is 1, it means that the change amplitude is within the achievable range and the change at this point is reliable; when its value is 0, it means that the speed limit is exceeded and the point is suspected of being a glitch / abnormal. It represents the upper limit of the maximum achievable change rate of the actuator at the normalized valve position scale. It is used to provide a physical constraint benchmark to limit how much the valve position can change per second and prevent electrical signal spikes from being misinterpreted as real motion. This represents the time interval between adjacent sampling points.

[0070] S403: Based on the product of the smoothed change in valve position feedback value and the smoothed change in valve position control command value, a criterion is defined that is consistent with the sign of the command change. This criterion is used to verify whether the direction of feedback change is consistent with the direction of command change, thereby identifying glitches / discontinuous contact anomalies such as feedback reversal jumps when the command has not changed significantly, and marking the availability of that point; that is:

[0071] In the formula, The result of the consistency between the instruction at time k and the feedback direction is represented by a binary flag. When its value is 1, it means that the directional relationship is within the fault tolerance range and can be considered consistent. When its value is 0, it means that there is obvious inconsistency in the reverse direction, which is suspected to be a glitch / anomaly. This represents the tolerance margin for determining directional consistency; its value is greater than or equal to 0 and is determined based on fitting experimental data.

[0072] S404: Based on the validity judgment results of the amplitude at time k, the validity judgment results of the rate of change, and the validity judgment results of the consistency between the instruction and the feedback direction, a single-point usability flag is generated, namely:

[0073] In the formula, The single-point availability flag at time k is used to fuse the results of multiple criteria and serve as the final switching quantity for determining whether the feedback at that sampling point is reliable and whether it should participate in subsequent processing. This represents the logical AND operation, which means that the AND operation is performed only if... , , When all values ​​are 1, The value is 1 only if it is 1, otherwise it is 0.

[0074] S5: Based on command and actuator constraints, this system enables valve position prediction. It ensures a physically reasonable and achievable valve position trajectory even when valve position feedback is unavailable or unreliable, such as due to glitches, jumps, or intermittent contact. Specifically:

[0075] S501: Obtain the smooth value of the valve position control command at the current moment. and the predicted valve position at the previous moment. Based on both, the following error is calculated to quantify the deviation between the current valve position command and the predicted valve position state. This deviation serves as the driving force for the next step of calculating the expected increment, thereby determining the direction and strength of the predicted valve position change. In the formula, The following error represents the magnitude and direction of the deviation between the commanded valve position at time k and the predicted valve position at the previous time.

[0076] S502: Calculate the expected increment based on the following error. The deviation between the current valve position command and the predicted valve position state is converted into the expected step size for the predicted valve position adjustment during the cost sampling period, i.e.: In the formula, This represents the expected increment of the predicted valve position that should ideally change during the kth sampling period, used to drive the predicted valve position closer to the command. The following coefficient is used to adjust the speed and smoothness of the predicted valve position following the command, and is determined based on fitting experimental data.

[0077] S503: Calculate the maximum permissible change in valve position based on the actuator's maximum valve position change rate and the time interval of the sampling period. This value is used as a limiting threshold to constrain the expected increment, preventing impossible jumps in the predicted valve position; that is: In the formula, This indicates the maximum permissible change in valve position within a sampling period.

[0078] S504: Based on the maximum permissible change in valve position, the expected increment is symmetrically limited, and the expected increment is transformed into a physically achievable valve position increment by the actuator within the current sampling period through speed limiting constraints, thus avoiding the prediction of impossible valve position jumps; that is:

[0079] In the formula, This represents the valve position increment that can be achieved within the k-th sampling period, i.e., the actual increment after speed limiting.

[0080] S505: Predicts the valve position based on achievable valve position increment updates, ensuring that the predicted changes conform to physical constraints, i.e.: ;in This represents the prediction threshold for the k-th sampling period.

[0081] S6: Reliably fuses feedback and prediction to output a clean valve position with stable switching and seamless splicing, obtaining a continuous and reliable valve position to characterize and synchronously reflect the health status of the feedback side. Specifically, when the valve position feedback smoothing value may be contaminated by spikes / discontinuous contact, a single-point availability flag is used to select between the true feedback and physically realizable prediction, generating a continuous and reliable valve position signal, avoiding bringing measurement artifacts into subsequent loop analysis; that is:

[0082] In the formula, This represents the cleaning valve position after feedback availability determination and replacement in the k-th sampling period; when When the time is right, it means that feedback is available. Use actual valve position measurements whenever possible; when When this happens, it indicates that feedback is unavailable. Predictive valve position is used instead of glitch feedback.

[0083] S602: Calculate statistical indicators based on single-point availability flags for early exposure of feedback link-related problems, namely: In the formula, This represents the average percentage of suspected glitches / intermittent contacts that are unavailable in the most recent time window during the kth sampling period. It is used to quantify the frequency of feedback link anomalies, serving as an early fault metric and for alarming / locating feedback link problems. N is the length of the statistical sliding window, representing the number of sampling points included in the sliding window, ranging from k-N+1 to k, for a total of N points. This is a single-point availability flag, indicating whether valve position feedback is available during the i-th sampling period in the sliding window.

[0084] However, when hard-switching and fusing the valve position feedback smooth value and the predicted valve position, the valve position sequence is prone to becoming discontinuous and jittery, especially at the moment when the feedback recovers from unavailable to available, resulting in obvious splicing jumps. The root cause is that the single-point availability indicator is a binary signal obtained from threshold discrimination, which frequently flips near the threshold boundary. Once it flips, it switches back and forth between the valve position feedback smooth value and the predicted valve position, and even a small deviation between the two will create a step error at the switching point. In addition, the predicted valve position is generated by commands and realizable constraints, and naturally does not include slow-varying errors such as zero-point offset and range drift on the feedback side. This can lead to a visible difference accumulating between the feedback recovery and the predicted value, further amplifying the splicing jumps. Consequently, the artificial step is injected into subsequent processing as a real excitation, causing or amplifying loop anomalies and false alarms; it also distorts the response characteristics, causing unrealistic fluctuations in the time constant and gain near the switching window, reducing the stability and interpretability of early fault trends.

[0085] To address this, by first performing hysteresis confirmation on the single-point availability flag, the easily jittered binary gating is transformed into a stable fusion state, reducing frequent switching at the source. Subsequently, during periods of reliable and stable feedback, the slowly varying deviation between prediction and feedback is estimated and tracked to compensate for the predicted value, reducing the connection drop during signal recovery. When the fusion state switches, a finite-length smooth transition is introduced, allowing the output to gradually transition from the current source to the target source within the transition window, avoiding the step-like splicing caused by hard switching. Finally, the statistical caliber of the feedback problem frequency index is changed to a sliding window statistical method based on the stable fusion state, making the index more robust and consistent with the actual source of use. This ensures that the valve position fusion output remains continuous, low-jitter, and interpretable, thereby reducing the interference of switching spurious excitations on subsequent dynamic identification and loop anomaly detection, and more clearly and independently quantifying the glitches or intermittent contact problems in the feedback link. More specifically:

[0086] S603: Based on the single-point availability flag, two consecutive counters are defined to count the duration of consecutive unavailability / consecutive availability feedback, thereby achieving hysteresis confirmation, suppressing jitter near the threshold, and avoiding frequent switching of fusion state; that is:

[0087] In the formula, A continuously unavailable counter indicates that it has appeared consecutively up to time k. The cumulative number of times is used to determine whether unavailability has been maintained continuously for a long enough time, thereby triggering the fusion state to switch to the use prediction and suppressing false switching caused by single-point spikes; A continuously available counter indicates that the number of consecutive occurrences up to time k is... The cumulative number of times is used to determine whether the available state has been continuously restored for a long enough time, thereby triggering the fusion state to switch back to use feedback and avoiding flipping back and forth near the threshold.

[0088] S604: Defines the fusion state based on a continuous counter, used to stably determine whether to use feedback or prediction with continuous confirmation and hysteresis mechanism that maintains the previous state, suppressing jitter and avoiding frequent switching, i.e.: In the formula, This represents the fusion state in the k-th sampling period, and its value set is... This indicates that it can only switch between two states, which serve as the final decision for subsequent fusion, determining whether the current output should use feedback or prediction, and providing a stable state signal to suppress jitter switching; where I and V are the state values, I indicates that feedback is invalid and the current signal should be the prediction signal, which is used to instruct the fusion logic to select the prediction channel; V indicates that feedback is valid and the current signal should be the feedback signal, which is used to instruct the fusion logic to select the feedback channel. This represents the threshold of the number of consecutive unavailable confirmations required to switch from using feedback to using prediction; This represents the threshold of the number of consecutive available confirmations required to switch from usage prediction to usage feedback.

[0089] S604: Based on the valve position feedback smoothing value and the predicted valve position definition observation residual, it quantifies the deviation between the feedback valve position and the model predicted valve position at the same time, used to identify systematic differences between the two, such as zero drift / range offset, and to provide error signals for subsequent deviation compensation and smoothing fusion; that is: In the formula, This represents the observation residual between the feedback valve position and the predicted valve position in the k-th sampling period.

[0090] S605: An update enable based on the fusion state definition, allowing updates to slowly varying deviations only when the feedback is reliable and the command / feedback is approximately stationary, to avoid treating real actions as deviations and to suppress deviation estimates from being contaminated by noise or transients; that is:

[0091] In the formula, This indicates the update enable for the k-th sampling period, used to determine whether to update the slow-varying bias at that moment; The change in the smooth value of the valve position control command is used to set a threshold with sufficiently small changes on the command side, and the control deviation update is only performed in the command smooth area; This represents the threshold for judging the stability of the change in the valve position feedback smooth value. It is used to set a threshold with sufficiently small changes on the feedback side, and the control deviation update is only performed in the feedback stable region.

[0092] S606: Robust limiting of observed residuals, used to truncate abnormally large residuals, such as occasional spikes / peaks. Within this framework, we must prevent these outliers from skewing subsequent bias estimates or introducing abrupt changes. In the formula, The truncated robust residuals are used to suppress the contamination of the estimation results by abnormally large residuals; This is the amplitude limiting threshold, used to set the strength of anti-peak.

[0093] S607: When update enable is met, asymptotic estimation of slow-varying deviation is performed based on robust residuals. This is used to estimate and track the system's slow-varying deviation / drift in each sampling period, and is subsequently used for deviation compensation of the valve position to reduce long-term systematic errors; that is:

[0094] In the formula, This represents the slowly varying deviation in the k-th sampling period, which is used as the current deviation compensation amount. The step size for bias updates is determined based on fitting experimental data.

[0095] S608: Calculates the valve position after deviation compensation based on slow-varying deviation and predicted valve position, which is used as the final output / execution valve position command; that is: In the formula, The valve position after deviation compensation has a value range of [0, 1].

[0096] S609: Defines edge events based on the fusion state to detect when the state switches from one mode to another in adjacent time intervals and triggers the corresponding switching control and fusion update logic, avoiding repeated execution in the steady state; that is: In the formula, This indicates that a rising edge event is detected at time k, meaning the state changes from I to V. This is used to trigger the logic for entering the transition interval and to start / refresh the transition start point memory. The falling edge event is detected at time k, that is, the state changes from V to I, which is used to trigger the transition process. Regardless of the switching direction, the switching is guaranteed to be completed in a smooth transition manner.

[0097] S610: Based on the edge event record, the most recent switching starting point is recorded. If any edge occurs, then... This is used to reopen a new transition window each time a switch occurs, allowing the splicing / merging process to complete gradually and avoiding abrupt jumps in the output; and to record the length L of the transition window; where This indicates the index of the start time of the most recent detected state transition.

[0098] S611: Define linear transition weights based on the start time index of the state transition and the transition window length, which are used as weighting coefficients for subsequent fusion; that is: In the formula, This represents the transition weight for the k-th sampling period, used to smoothly transition from the old mode to the new mode.

[0099] S612: Update the cleaning valve position based on the valve position after deviation compensation and the transition weight:

[0100] When the current target state is to use feedback, that is Then, the transition from prediction to feedback is smooth; that is: ;

[0101] When the current target state is to use prediction, i.e. Then, the transition from feedback to prediction is smooth; that is: .

[0102] S613: Based on the percentage of prediction usage within the most recent time window, this statistical method reflects whether the prediction usage is too high and whether the status is stable, providing a basis for subsequent fusion weight, threshold determination, or strategy switching; that is: In the formula, This represents the percentage of time k within the nearest window that is in state I. The larger the value, the more frequently the feedback becomes unreliable / needs to be replaced within that window.

[0103] S7: Time-delay alignment is performed between the cleaning valve position and the controlled process variable, and an online response model is established to characterize the true dynamic relationship from control input to process output. Specifically:

[0104] S701: Utilizes a first-order discrete ARX structure to describe the local dynamics of the controlled process quantity from the clean valve position, used to separate the normal, interpretable response from the uninterpretable portion caused by anomalies / faults; that is: In the formula, This represents the inertia of the controlled process quantity itself; where The autoregressive coefficient at time k is used to characterize the dynamic velocity and decay characteristics of the system. This indicates the strength of the effect of the cleaning valve position on the controlled process variable; where This represents the input gain coefficient at time k; The clean valve position is indicated at time k−d; d represents the pure time delay step in the sense of discrete sampling, that is, the number of sampling points between the valve position change and the occurrence of the process response, which is used to achieve time alignment of input and output. This indicates the baseline bias that is not explicitly modeled within the current operating window, such as environmental load or the average effect of unmodeled disturbances. This represents the model error at time k.

[0105] However, in the process of hydraulic and thermal transmission, the changes in the valves at the input end are not transmitted to the output end, such as temperature, after a fixed time. Instead, they drift significantly forward or backward depending on the operating conditions. When a constant pure time delay step d is used to align the input and output for online identification, this alignment error will be treated as an anomaly or parameter change in the system itself, resulting in a continuous systematic bias in the identification results. This is because the system operates under variable flow conditions. Changes in the overall network operating conditions will alter the pressure difference across the valves and the branch flow, thereby changing the flow velocity in the coil. At the same time, the sensor is not located at the instantaneous position of energy exchange, but rather across a certain transport distance and mixing volume. As the flow velocity changes, the transport time required for heat and water to reach the measuring point also changes, causing the equivalent pure time delay to continuously change under different operating conditions. Even if the valve position signal has been processed to be continuous, the subsequent alignment with a fixed delay will still cause misalignment when the operating conditions change. This misalignment will directly increase the prediction error and cause the error level to rise periodically or in a condition-related manner. It will also force online identification to use incorrect time alignment to interpret the observation data, making the model parameters appear to drift unrealistically with the operating conditions. This will dilute and confuse the trend signal that should be used for early fault identification, reducing identifiability and repeatability.

[0106] To address this, by imposing continuity and stability constraints on the switching process between valve position feedback and prediction, and adaptively compensating for the slow-varying differences between the two, a smooth and consistent cleaning valve position sequence is generated. This results in a continuous, low-jitter, and physically interpretable cleaning valve position signal, used to avoid pseudo-step contamination during switching and subsequent dynamic identification, allowing residuals and parameter changes to more accurately reflect the loop state and early abnormal trends. More specifically:

[0107] S7011: Since time delay inversion depends more on change than absolute value, a moving average is applied to the input and output to obtain the mean-removed excitation component. This is used to suppress the effects of slow drift / baseline changes, allowing subsequent correlation calculations in time delay inversion to focus more on real dynamic changes, thus estimating time delay more stably and accurately; that is: In the formula, This represents the sliding mean of the cleaning valve position with k as the end point of the time window and a length of N. It is used to estimate and remove the slow-varying component / bias of the cleaning valve position, so that subsequent processing pays more attention to the amount of change. This represents the moving average of the controlled process variable with time window k as the end point and length N. It is used to estimate and eliminate the slow-changing trend of the controlled process variable, such as slow load changes and environmental drift, to avoid them dominating the time delay judgment. To remove the mean value component from the clean valve position; The mean-removed component of the controlled process quantity.

[0108] S7012: Defines time delay inversion enable to avoid misalignment when noise is dominant or operating conditions are almost constant. In the formula, Enables time delay inversion at time k, which determines whether to perform time delay estimation update at this time and suppresses misalignment and parameter drift caused by unreliable data or no excitation; For time k, the data quality index reflects a comprehensive measure of measurement reliability, anomaly level, and missing data level. It is used to determine whether the data is reliable enough to allow for updates and to prevent noise, glitches, and sensor anomalies from dominating time delay inversion. This is a data quality threshold used to separate usable and unusable data segments; when Updates should be disabled during this period to reduce false positives; The excitation intensity index at time k reflects whether the input has enough change within the current window for identification and correlation analysis, and is used to determine whether there is enough information to estimate the time delay; it avoids instability of the correlation spectrum and random jumps in time delay estimation when the operating conditions are almost unchanged; The excitation strength threshold is used to ensure that the delay is updated only when the excitation is sufficient, thereby improving the stability and repeatability of delay inversion.

[0109] S7013: If Then the time delay is frozen, maintaining the estimate from the previous time step, i.e.:

[0110] In the formula, This represents the estimated time delay at time k;

[0111] like Within the current sliding window, a set of possible pure time delay steps d is scanned to find the time misalignment that most strongly couples the demeaned component of the clean valve position of the input with the demeaned component of the controlled process variable of the output. This yields the initial integer time delay value used for subsequent fractional time delay refinement and input-output alignment. More specifically:

[0112] S70131: Define the candidate integer lag set In the formula and These are the minimum and maximum allowed integer lags, respectively: ;in This represents the possible time delay range.

[0113] S70132: Based on the candidate integer lag set, a normalized cross-correlation score is defined to quantify the coupling strength between the mean-de-mean component of the input clean valve position and the mean-de-mean component of the output controlled process quantity under different pure lag steps, and to eliminate the influence of amplitude scaling.

[0114] In the formula, This represents the normalized cross-correlation score of the demeaned component of the clean valve position and the demeaned component of the controlled process quantity within the current window when the candidate lag is d. It is used to compare which alignment is best under different pure lag steps d.

[0115] S70133: Select the lag corresponding to the peak value based on the normalized cross-correlation score, i.e.:

[0116] In the formula, Let k be the optimal integer initial delay value.

[0117] S70134: However, relying solely on integer lag can lead to alignment skips. Therefore, parabolic interpolation is performed on the peak neighborhood to obtain fractional correction, i.e.:

[0118] In the formula, It represents the fractional offset of the peak position at time k relative to the initial integer time delay value. It is used to refine the integer lag into a fractional lag to obtain a more accurate time delay.

[0119] S70135: Calculate the original fractional delay of continuous time quantities based on fractional offsets, i.e.:

[0120] In the formula, Let k be the original fractional time delay at time k.

[0121] S70136: To avoid time delay jitter due to correlated spectral noise, a continuity constraint is introduced to limit the update rate based on the original fractional time delay, i.e.:

[0122] In the formula, p is the maximum allowable time delay variation for each sample.

[0123] S7014: Converts the time delay estimate into fractional sampling delay, and decomposes it into integer and fractional delays, so as to achieve actual discrete signal alignment and compensation using integer shift and fractional delay interpolation; that is: In the formula, The fractional sampling delay at time k is equal to the time delay estimate converted into the number of sampling points, and is used as the target delay amount for discrete fractional delay filtering; n is the integer delay, used to implement integer shifting, and the main delay is completed by moving the sampling points. The fractional delay is used to interpolate between two adjacent sampling points to achieve subsampling level alignment, and its value range is [0, 1).

[0124] S7015: Calculates the fractional delay filter for alignment to the output time based on integer and fractional delays, i.e.: In the formula, This represents the aligned signal value obtained at output time k after aligning the input cleaning valve position according to the fractional sampling delay.

[0125] S7016: Replace the fixed delay term with the alignment signal value, i.e.:

[0126]

[0127] S702: Define the regression vector and parameter vector to construct comparable indicators of whether the loop response is abnormal; that is: In the formula, The regression vector at time k is a set of features composed of observable data; The parameter vector representing time k is the set of model coefficients to be identified online.

[0128] S703: Based on the regression vector and parameter vector, calculate the predicted values ​​and residuals of process quantities to construct comparable indicators of loop response anomalies and monitor model mismatch; that is: In the formula, This represents the predicted value of the process quantity at time k; This represents the process residual at time k; This represents the transpose of the regression vector.

[0129] S704: Calculate the gain vector based on the regression vector, which is used to adaptively determine the weight of the current parameter update according to the information content of the current data and the parameter uncertainty, mapping the process variable residual to the parameter correction quantity and updating it; that is: In the formula, The RLS gain vector at time k is used to convert the process residual into a parameter correction value; the larger the value, the greater the impact of the current data on the parameter update. Let represent the parameter estimation covariance matrix at time k−1; This represents the forgetting factor, used to control the rate at which historical data is forgotten.

[0130] S705: Update the parameter vector and covariance matrix based on the gain vector, i.e.;

[0131] In the formula, Let represent the parameter estimation covariance matrix at time k.

[0132] S8: Extract the equivalent static gain, equivalent time constant, and residual energy from the parameter vector to compress the online identification results into three types of measurable health indicators. These indicators quantify the loop's effect strength, response speed, and degree of unexplained anomalies, respectively, for early fault trend tracking and anomaly detection; that is: In the formula, It represents the equivalent static gain at time k, i.e. the strength of the steady-state influence of the input on the output. It is used to monitor whether the loop effect is weakening or strengthening, for example, whether the change in process quantity caused by the same valve position change is reduced, and to support the attenuation judgment of the ability to support early faults. It represents the equivalent time constant at time k, describes how quickly the output responds to changes in the input, and is used to monitor whether the loop is slowing down. It is a core speed-type indicator for early fault trend detection. This represents the mean square energy of the predicted residual within a time window of length N, ending at time k. .

[0133] S9: By logically gating the valve position data quality and excitation sufficiency, and accordingly selecting branches for processing such as freezing, weighting, output loop anomalies, or feedback link problems, reliable loop response anomaly conclusions are given when the data is trustworthy and identifiable, avoiding misjudging acquisition anomalies or insufficient operating conditions as early equipment failures. Specifically:

[0134] S901: Calculates a data quality score based on the average percentage of unavailable feedback, used to quantify the reliability of the current valve position data, thereby determining whether subsequent loop modeling and anomaly detection should continue, be frozen, or prioritize outputting feedback link issues to avoid misjudgments; that is:

[0135] In the formula, This represents the data quality score.

[0136] S902: Defines a quality judgment threshold used to convert continuous data quality scores into an executable trustworthiness decision switch. A quality availability flag is generated based on the quality judgment threshold and data quality score. This flag is used to explicitly trigger subsequent branch strategies in the engineering implementation. In the formula, The data quality availability flag at time k is used to determine whether subsequent processing allows the use of the current window data.

[0137] like This indicates that the data quality is substandard, and no strong conclusions should be drawn about abnormal loop responses. Although the model input / alignment / identification can be calculated, its statistical significance is insufficient, and it is easy to mistake the acquisition problem for a loop problem. The conclusion that the risk of the feedback acquisition link has increased should be given priority in the output.

[0138] like This indicates that the data quality meets the standards. Continue to calculate the incentive strength and generate an incentive availability flag to determine whether the incentive is sufficient.

[0139] S903: Calculates the excitation intensity based on the clean valve position to quantify whether the valve position change within the most recent window is sufficient; that is: In the formula, This represents the cumulative change intensity of the cleaning valve position within a window of length n, ending at time k. It is used to measure whether there is sufficient excitation in the input within this window to determine whether subsequent model identification and loop anomaly determination are allowed. A larger value usually indicates more sufficient excitation.

[0140] S904: Defines a threshold for determining whether a continuous stimulus intensity is sufficient, used to convert it into a definite stimulus intensity. Based on the intensity judgment threshold and the excitation intensity, an excitation availability flag is generated to determine whether subsequent modules consider the current window data as identifiable and valid data; that is:

[0141] In the formula, The excitation available at a given time can be flagned to control whether to update model parameters, output abnormal loop conclusions, or enter an insufficient excitation branch.

[0142] like This indicates insufficient stimulus, freezes model updates, and maintains the parameter vector and parameter estimation covariance matrix from the previous time step; it also freezes or reduces the weight of the output loop index, continuing to display historical stable values ​​or marking it as not updatable.

[0143] like This indicates that the stimulus is sufficient, and the system enters the normal identification and loop anomaly determination path, processing the results output by S7 and S8 as valid.

[0144] S10: Combines output feedback to collect problem indicators and loop response anomaly indicators, used to differentiate between measurement link problems and early equipment / loop faults, and to support subsequent diagnostic decisions. Includes quantitative results of intermittent valve position feedback contact. and abnormal loop response results based on cleaning valve position .

[0145] And use persistent logic to give early fault detection conclusions, for example: when Continue to rise and When the drift exceeds the baseline allowable range, the circuit response is deemed abnormal, such as dynamic changes caused by increased valve mechanical friction, decreased heat exchange capacity, or bypass leakage, and this conclusion is no longer contaminated by the burrs of the valve position feedback original value.

[0146] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early fault detection of HVAC equipment based on abnormal control loop response, characterized in that, Includes the following steps: S1: Acquire field signals from the control loop, the field signals including at least valve position control commands, valve position feedback raw values ​​and controlled process quantities, and normalize the valve position feedback raw values. S2: The acquired field signals are smoothed and filtered to suppress noise and obtain a smooth signal sequence that represents the trend of physical change. S3: Calculate the difference between the smoothed signal sequence in adjacent sampling periods to characterize the dynamic characteristics of the control action and the object response; S4: Based on the smoothed value of the valve position feedback, perform at least the following logical checks: amplitude range check, rate of change speed limit check, and direction consistency check, to generate a single-point usability flag characterizing the reliability of the valve position feedback data; wherein, the rate of change speed limit check is used to determine whether the change in the smoothed value of the valve position feedback within a single sampling period exceeds the upper limit of the actuator speed determined by the maximum physical speed of the actuator; the direction consistency check is used to determine whether the direction of change of the smoothed value of the valve position feedback is consistent with the direction of change of the smoothed value of the valve position control command within a preset fault tolerance range; S5: Based on the smoothed value of the valve position control command and the physical speed constraint of the actuator, construct and update the physically realizable predicted valve position; S6: Based on the single-point availability flag, perform reliable fusion and smooth switching between the smooth value of the valve position feedback and the predicted valve position, output a continuous and physically reliable clean valve position signal, and calculate the feedback link anomaly index based on the single-point availability flag. S7: Time-varying delay alignment is performed on the cleaning valve position signal and the controlled process quantity, and a response model describing the dynamic relationship between the cleaning valve position and the controlled process quantity is established online based on the aligned signal; wherein, time-varying delay alignment refers to performing mean-removal processing on the cleaning valve position signal and the controlled process quantity to focus on dynamic changes, then estimating the optimal integer delay by calculating the normalized cross-correlation score within a sliding window, and further obtaining the fractional delay through parabolic interpolation to achieve subsampling-level accurate alignment; then, the first-order ARX model parameters are updated online using this aligned signal; this is used to avoid misjudging delay drift as abnormal system parameters; S8: Extract monitoring indicators for characterizing the health status of the loop from the online identification results of the response model. The monitoring indicators include at least the equivalent static gain, the equivalent time constant, and the model residual energy. S9: Based on the feedback link anomaly index and the excitation sufficiency of the cleaning valve position signal, generate a data quality availability flag and an excitation availability flag, and control the update of the response model and the output logic of the monitoring index accordingly; wherein, the excitation sufficiency is used to quantify whether the valve position change within the most recent window is sufficient; S10: Output the feedback link anomaly index and the monitoring index together, and determine the early fault of the HVAC equipment control loop based on the continuous change trend of the monitoring index.

2. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, Step S1 involves normalizing the original valve position feedback value, including: In the formula, k is the discrete-time index; This represents the original valve position feedback value collected at time k; This represents the normalized valve opening at time k; This is the lower limit of calibration when the valve corresponding to the feedback channel is fully closed; This is the upper limit of the calibration when the valve corresponding to the feedback channel is fully open; This is a clipping function.

3. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, The smoothing filtering process in step S2 uses a first-order low-pass filter, including: In the formula, The original observation value at the current time k; Let k be the filtered signal value at time k; The signal value after filtering by k-1 at the previous time step; 'a' is the weighting coefficient; 'x' is the input quantity; y represents the valve opening command; y represents the controlled process variable. This is the normalized valve position opening.

4. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, The calculation formula for the rate of change limit verification in step S4 is as follows: In the formula, This indicates the result of the validity determination of the rate of change at time k; This indicates the upper limit of the maximum achievable rate of change of the actuator at the normalized valve position scale; The time interval between adjacent sampling points; This represents the change in the valve position feedback smoothing value in the kth sampling period relative to the previous sampling period.

5. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, Step S5, which involves constructing and updating the predicted valve position, includes the following sub-steps: S501: Calculate the following error between the smoothed value of the valve position control command and the predicted valve position at the previous moment; S502: Calculate the expected increment based on the following error and a following coefficient; S503: Based on the maximum speed limit of the actuator and the sampling time interval, calculate the maximum allowable change in valve position within a single cycle; S504: Limit the desired increment by the maximum permissible change in the valve position to obtain the achievable valve position increment; S505: The predicted valve position at the current moment is updated using the valve position increment.

6. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 5, characterized in that, The formula for updating the predicted valve position in step S505 is as follows: In the formula, This represents the prediction threshold for the k-th sampling period; This represents the valve position increment that can be achieved within the k-th sampling period.

7. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, Step S6 outputs the cleaning valve position signal, including: According to the fusion state and transition weight Calculate the cleaning valve position signal ; When the fusion status indicator uses feedback: ； When the fusion state indicator uses prediction: ； In the formula, This represents the valve position feedback smoothing value at time k; This is the valve position after deviation compensation.

8. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, The time-varying delay alignment in step S7 includes estimating the time-varying delay by calculating the normalized cross-correlation score between the cleaning valve position signal and the controlled process variable within a sliding window. The formula for calculating the normalized cross-correlation score is as follows: ； In the formula, This represents the normalized cross-correlation score between the mean-demeaned component of the clean valve position and the mean-demeaned component of the controlled process quantity within the current window when the candidate lag is d; N is the statistical sliding window length; d is the pure lag step number. To remove the mean value component from the clean valve position; is the mean-removed component of the controlled process variable; k is the discrete-time index.

9. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, The formulas for calculating the equivalent static gain and equivalent time constant extracted in step S8 are as follows: In the formula, This represents the equivalent static gain at time k; Represents the equivalent time constant at time k; The time interval between adjacent sampling points; Let be the autoregressive coefficient at time k; where This represents the input gain coefficient at time k.

10. The method for early fault detection of HVAC equipment based on abnormal control loop response according to claim 1, characterized in that, The formula for calculating the feedback link anomaly index in step S6 is as follows: ； In the formula, This represents the average percentage of suspected glitch / intermittent contact feedback that is unavailable within the most recent time window of the kth sampling period; is a single-point availability flag; N is the statistical sliding window length.

Images