Energy-saving air source heat pump health status assessment methods, electronic equipment and storage media

By using predictive models and feature extraction methods, the problems of inaccurate prediction and difficulty in fault location in the health status assessment of energy-saving air source heat pumps are solved, and refined health assessment and fault diagnosis of the equipment are realized.

CN123069150APending Publication Date: 2026-11-10LIAONING AIKE SMART HOME CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611250466.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for assessing the health status of energy-saving air source heat pumps suffer from problems such as inaccurate prediction benchmarks, difficulty in extracting degradation signals due to changes in operating conditions, inability to pinpoint the health status to specific components, and high costs associated with building a fault knowledge base.

Method used

A prediction model-based approach is adopted. By acquiring and preprocessing the thermodynamic state variables, performance state variables, control variables, and disturbance variables of the air source heat pump, a difference sequence is generated using the state prediction equation and output equation. The device degradation sequence is then separated by low-pass filtering or least squares regression to extract multidimensional features. Finally, health assessment is performed by combining Mahalanobis distance and a fault knowledge base.

Benefits of technology

It enables accurate extraction of equipment performance degradation signals under different operating conditions, improves the precision of health assessment and fault location capabilities, and provides a comprehensive and interpretable basis for maintenance decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069150A_ABST
    Figure CN123069150A_ABST
Patent Text Reader

Abstract

This disclosure relates to the field of equipment health management and fault diagnosis technology, and in particular to a method, electronic device, and storage medium for assessing the health status of an energy-saving air source heat pump. The method includes: collecting and preprocessing the operating data of the air source heat pump; inputting the preprocessed data into a state prediction equation to obtain predicted values ​​for thermodynamic state variables and performance state variables; then, using an output equation to obtain predicted values ​​for auxiliary observations; calculating the difference between the measured values ​​and the predicted values ​​to construct a difference sequence; distinguishing between steady-state and dynamic operating conditions; using low-pass filtering in steady-state conditions and least-squares regression in dynamic conditions to separate the equipment degradation sequence; extracting multi-dimensional features from the equipment degradation sequence and combining them into a feature vector; calculating the Mahalanobis distance with the health benchmark feature space to obtain a comprehensive health score; matching the feature vector with a fault knowledge base to output the fault type; and generating a health status assessment report.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of equipment health management and fault diagnosis technology, and in particular to an energy-saving air source heat pump health status assessment method, electronic equipment and storage medium. Background Technology

[0002] As a highly efficient and energy-saving heating and cooling device, the economic efficiency and reliability of energy-saving air source heat pumps are highly dependent on the health status of the equipment itself. However, in actual operation, energy-saving air source heat pumps face various performance degradation factors such as frosting, dust accumulation, scaling, refrigerant leakage, and compressor wear. These degradation processes are characterized by nonlinearity, multivariate coupling, and strong time-varying properties. If they are not identified and intervened in a timely manner, they will lead to a significant decrease in system energy efficiency or even equipment failure. Therefore, accurately assessing the health status of air source heat pumps is of great significance for ensuring the safe and efficient operation of the equipment and reducing maintenance costs.

[0003] Existing technologies include several methods for condition monitoring and fault diagnosis of air-source heat pumps or similar refrigeration systems. For example, some solutions use single temperature or pressure thresholds for fault alarms, or employ data-driven methods to classify and identify equipment operating states. However, these existing technologies still have the following shortcomings: First, existing methods typically rely on only a subset of sensor signals for health assessment, lacking a systematic classification and comprehensive utilization of thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. This fails to fully leverage the synergistic value of multi-source sensor information. Furthermore, existing prediction models often employ fixed parameter structures, making it difficult to accurately reflect the impact of equipment performance degradation caused by physical or chemical processes on its thermodynamic output state. This results in inaccurate state prediction benchmarks, consequently affecting the reliability of subsequent health assessments.

[0004] Secondly, existing methods, after generating the difference sequence, typically attribute the deviation between measured and predicted values ​​directly to equipment degradation, without fully considering the impact of changes in operating conditions on the difference. When operating parameters such as compressor operating frequency, outdoor ambient temperature, or water circulation mass flow rate change significantly, the model tracking error itself will generate a large difference. Existing methods struggle to distinguish between the difference fluctuations caused by changes in operating conditions and the difference offsets caused by equipment performance degradation, resulting in distorted extraction of equipment degradation signals under dynamic operating conditions.

[0005] Third, existing health assessment methods are mostly single-level scoring methods, which only provide an overall health index for the equipment and cannot pinpoint the specific components experiencing performance degradation. Although maintenance personnel can be aware of the overall deterioration in the equipment's condition, it is difficult to determine whether the problem lies with the evaporator, condenser, compressor, or fan subsystem, making it inconvenient to develop targeted maintenance strategies.

[0006] Fourth, most existing fault diagnosis methods rely on a large number of actual fault samples for model training. However, air source heat pumps have a low frequency of faults, diverse fault types, and varying degrees of severity in actual operation. Obtaining actual fault samples is costly and time-consuming, making it difficult to build a comprehensive diagnostic knowledge base that covers all fault types. This results in insufficient accuracy and reliability of fault matching. Summary of the Invention

[0007] The purpose of this disclosure is to provide an energy-saving air source heat pump health status assessment method, electronic device, and storage medium to solve the technical problems of inaccurate prediction benchmarks, extraction of degradation signals due to interference from changes in operating conditions, inability to locate health status to specific components, and high cost of building a fault knowledge base.

[0008] To achieve the above objectives, in a first aspect, this disclosure provides a method for assessing the health status of an energy-saving air source heat pump, the method being executed based on a preset prediction model, the prediction model including a state prediction equation and an output equation; The method includes: S1: Data Acquisition and Preprocessing Within the current prediction period of the prediction model, the operating data of the air source heat pump is acquired. The operating data includes: thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. The thermodynamic state variable is a parameter directly measured by a sensor, which characterizes the current thermodynamic output state of the air source heat pump. The performance state variables are parameters obtained in real time through preset initial values ​​or through preset state estimators, which characterize the performance degradation state of the components of the air source heat pump caused by physical or chemical processes. The control variable is an actuator parameter that can be actively adjusted, the disturbance variable is an external environmental parameter that is uncontrollable but measurable, and the auxiliary observation is a parameter that is directly measured by a sensor, other than the thermodynamic state variable, the performance state variable, the control variable, and the disturbance variable. The operating data is preprocessed to obtain the preprocessed thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. The preprocessing includes timestamp alignment and sampling frequency unification, outlier removal and missing value imputation, and normalization. S2: Predictive Model State Prediction The preprocessed thermodynamic state variables, performance state variables, control variables, and disturbance variables are input into the state prediction equation to obtain the predicted values ​​of the thermodynamic state variables and performance state variables for the prediction period output by the state prediction equation. The predicted values ​​of the thermodynamic state variables and the predicted values ​​of the performance state variables are input into the output equation to obtain the auxiliary observation prediction values ​​of the prediction period output by the output equation. S3: Difference Sequence Generation and Decomposition The difference between the thermodynamic state variable and the predicted value of the thermodynamic state variable, and the difference between the auxiliary observation and the predicted value of the auxiliary observation are determined respectively, so as to obtain the difference between the thermodynamic state variable and each variable included in the auxiliary observation within the prediction period; Using the prediction period and the N periods preceding the prediction period as the current time window, for each variable of the thermodynamic state variable and the auxiliary observation, the difference of the variable in each period within the time window is arranged in chronological order to obtain the difference sequence of the variable within the time window; Based on the rate of change of the control variable, the disturbance variable, and the variable used to characterize the operating condition in the auxiliary observation, determine whether the operating condition of the time window belongs to a steady-state condition or a dynamic condition. If the time window is a steady-state condition, then the difference sequence of each variable of the thermodynamic state variable and the auxiliary observation is subjected to low-pass filtering to obtain the equipment degradation sequence of each variable. If the time window is a dynamic operating condition, then the least squares method is used to perform regression analysis on the difference sequence of each variable to obtain the equipment degradation sequence of each variable. S4: Feature Extraction and Health Assessment Features are extracted from the device degradation sequence of each variable, the features including at least one of amplitude features, trend features, correlation features and spectral features, and the features of all variables are combined into a feature vector of the time window; Determine the Mahalanobis distance between the feature vector and the preset health benchmark feature space to obtain the comprehensive health score of the air source heat pump within the time window; For each component of the air source heat pump, according to the preset mapping relationship between the component and the thermodynamic state variables and auxiliary observation quantities, the corresponding variable subset of the component within the time window is obtained, the features of the variable subset are combined into a feature vector subset, and the Mahalanobis distance between the feature vector subset and the preset health benchmark feature space of the component is determined to obtain the component sub-item health score of the component within the time window. The feature vector of the time window is input into a preset fault knowledge base for matching to obtain the fault type output by the fault knowledge base. A health status assessment report is generated based on the comprehensive health score, the health score of each component sub-item, and the fault type, and the health status assessment report is output.

[0009] In a second aspect, this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method described in the first aspect.

[0010] Thirdly, this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects of this disclosure are: By dividing operational data into thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations, and performing state prediction based on a prediction model that includes state prediction equations and output equations, multi-source sensor information can be comprehensively utilized to improve the accuracy of prediction benchmarks and provide a reliable reference for health assessment.

[0012] By calculating the difference between measured and predicted values ​​to construct a difference sequence, and judging steady-state or dynamic operating conditions based on the rate of change of variables characterizing operating conditions, low-pass filtering or least squares regression is then used to separate the equipment degradation sequence. This approach can adapt to different operating conditions, accurately extract equipment performance degradation signals, and effectively avoid interference from changes in operating conditions on degradation assessment.

[0013] By extracting multidimensional features from the equipment degradation sequence and combining them into a feature vector, the Mahalanobis distance between the feature vector and the health benchmark feature space is calculated to obtain a comprehensive health score. Based on the mapping relationship between components and variables, the component-specific health scores are obtained, realizing multi-level health quantification from the whole machine to the components, improving the accuracy of the assessment and the fault location capability.

[0014] By matching feature vectors with a preset fault knowledge base to obtain fault types, and combining comprehensive health scores and component sub-item health scores to generate a health status assessment report, quantitative health status can be combined with qualitative fault diagnosis, providing a comprehensive and interpretable decision-making basis for the operation and maintenance of air source heat pumps. Attached Figure Description

[0015] Figure 1 The overall flowchart of the health status assessment method for energy-saving air source heat pumps provided in this disclosure; Figure 2 This is a schematic diagram of the structure of the prediction model provided in this disclosure; Figure 3 This is a flowchart illustrating the generation and decomposition of the difference column provided in this disclosure; Figure 4 This is a flowchart illustrating the health assessment process provided in this disclosure. Detailed Implementation

[0016] To further illustrate the technical means and effects adopted by this disclosure to achieve the intended purpose of the invention, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, provides a detailed explanation of the specific implementation of an energy-saving air source heat pump health status assessment method, electronic device, and storage medium proposed according to this disclosure. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0018] The present disclosure will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0019] The method disclosed herein can be executed in an embedded controller. The controller is electrically connected to the sensor array, actuator, and communication module of the air source heat pump.

[0020] The sensor group may include: a temperature sensor for measuring the inlet and outlet water temperatures, ambient temperature, compressor suction and discharge temperatures, and condenser outlet temperature; a pressure sensor for measuring the compressor suction and discharge pressure; a flow sensor for measuring the water flow rate; a humidity sensor for measuring indoor and outdoor relative humidity; a power sensor for measuring the total input power of the unit; and a differential pressure sensor for measuring the evaporator pressure drop.

[0021] The controller integrates a pre-defined prediction model. This model is a nonlinear state-space model built upon thermodynamic mechanisms, comprising state prediction equations and output equations. The prediction model takes the thermodynamic state variables, performance state variables, control variables, and disturbance variables of the current prediction period as inputs. It recursively calculates the predicted values ​​of the thermodynamic and performance state variables through the state prediction equations, and then maps these predicted values ​​to predicted values ​​of auxiliary observations through the output equations.

[0022] S1: Data Acquisition and Preprocessing like Figure 1 As shown, the operating data of the air source heat pump is acquired within the current prediction period of the prediction model. The operating data includes: thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations.

[0023] Thermodynamic state variables are parameters directly measured by sensors, characterizing the current thermodynamic output state of the air source heat pump. In this embodiment, the thermodynamic state variables specifically include: Water outlet temperature (°C) is measured by a water outlet temperature sensor. It characterizes the heat or cold output status on the user side and reflects the instantaneous result of the heat or cold power provided by the air source heat pump to the water circuit.

[0024] : Evaporator refrigerant side evaporation temperature (°C), determined by compressor suction pressure The temperature is obtained in real time through the refrigerant property equation, which characterizes the outdoor unit heat exchange temperature corresponding to the heating mode and the indoor unit heat exchange temperature corresponding to the cooling mode. The variables affect the compressor suction specific volume and the air source heat pump frosting rate.

[0025] : Condenser refrigerant side condensing temperature (°C), determined by compressor discharge pressure The heat exchange temperature of the indoor unit is obtained in real time through the refrigerant property equation, which represents the heat exchange temperature of the outdoor unit in the heating mode and the heat exchange temperature of the outdoor unit in the cooling mode. The variables affect the compressor discharge pressure and the heat exchange capacity of the water circuit.

[0026] The compressor suction temperature (°C), measured by a suction port temperature sensor, characterizes the superheated temperature state of the refrigerant at the compressor suction port. This variable is related to the evaporation temperature. Together they determine the intake superheat .

[0027] The refrigerant temperature at the condenser outlet (°C) is measured by a temperature sensor on the condenser outlet pipe and characterizes the temperature state of the refrigerant in the subcooled section at the condenser outlet. This variable is related to the condensing temperature. Together they determine supercooling .

[0028] Performance state variables are parameters obtained in real time through preset initial values ​​or through a preset state estimator, characterizing the performance degradation state of the air source heat pump components due to physical or chemical processes. In this embodiment, the performance state variables specifically include: : Thermal resistance of frost layer (m 2 (·K / W), characterizing the degree to which the frost layer on the evaporator surface hinders heat transfer to the air side under heating conditions; when the frost layer thickens, Increased efficiency leads to decreased heat exchange efficiency, and the temperature is reset to zero after refrigeration or defrosting.

[0029] Heat exchanger overall fouling thermal resistance (m) 2 The K / W value represents the long-term increase in thermal resistance on both the water and refrigerant sides due to the cumulative effects of scaling, dust accumulation, and oil film. This slow-change state reflects the degree of performance degradation of the heat exchanger as the equipment ages (over several months / years).

[0030] The compressor volumetric efficiency benchmark characterizes the decline in volumetric efficiency caused by factors such as long-term wear and tear and increased leakage clearances. This slowly changing state is used to correct the volumetric efficiency in refrigerant mass flow calculations and characterizes the degree of compressor aging.

[0031] In this embodiment, the state estimator employs an extended Kalman filter or an unscented Kalman filter, taking the control variable, disturbance variable, and auxiliary observation as inputs, and recursively corrects the state prediction equation and measurement update equation of the prediction model, outputting the performance state variable at the current moment in real time.

[0032] The control variables are actively adjustable actuator parameters. In this embodiment, the control variables specifically include: The compressor operating frequency (Hz) regulates the refrigerant circulation mass flow rate, thereby controlling the heating or cooling capacity of the air source heat pump.

[0033] The opening degree (steps) of the electronic expansion valve adjusts the refrigerant flow distribution and the degree of throttling.

[0034] Outdoor fan speed (r / min) adjusts the outdoor airflow and affects the evaporation or condensation temperature.

[0035] The disturbance variable is an uncontrollable but measurable external environmental parameter. In this embodiment, the disturbance variable specifically includes: Water return temperature (°C); : Mass flow rate of circulating water in the water circuit (kg / s); Real-time electricity price (RMB / kWh); Indoor ambient temperature (°C); Indoor relative humidity (%) The auxiliary observations are parameters directly measured by sensors, excluding the thermodynamic state variables, performance state variables, control variables, and disturbance variables. In this embodiment, the auxiliary observations specifically include: The total input power (kW) is measured by a power sensor. The compressor discharge temperature (°C) is measured by the discharge port temperature sensor. The compressor suction pressure (MPa) is measured by the suction-side pressure sensor. : Compressor discharge pressure (MPa), measured by the discharge-side pressure sensor; : Evaporator pressure drop (Pa), measured by a differential pressure sensor; Outdoor ambient temperature (°C), measured by an ambient temperature sensor; Outdoor relative humidity (%), measured by a humidity sensor.

[0036] The acquired operational data is preprocessed to obtain preprocessed thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. The preprocessing includes timestamp alignment and sampling frequency unification, outlier removal and missing value imputation, and normalization.

[0037] There may be deviations in the sampling time of each sensor; some sensors may sample at the wrong time. Sampling is completed at +0.02 seconds, while another sensor may be... Sampling is completed at -0.03 seconds, causing the acquisition time of each variable to be inconsistent with the unified standard time grid. At the same time, different sensors may also have different sampling frequencies, resulting in different data densities for each variable at the same moment.

[0038] Therefore, in the preprocessing stage, the timestamps of each sensor are first aligned to the same time base using a unified global clock. Specifically, for each sensor data point, its original timestamp is mapped to a standard time grid under the global clock. This allows all sensors to share the same time coordinate system.

[0039] After timestamp alignment is complete, the data from all sensors are resampled to the preset sampling period. For example, in this embodiment, Second.

[0040] For sensor sampling intervals less than the preset sampling period The data is used to extract sampling points corresponding to the standard time grid. If the grid time is... If there is no exact matching sampling point, then select... The nearest sampling point before and after is used as The value, or the value of The average value of several adjacent sampling points is taken as The value is set to ensure that only one value is retained at any given time for each standard grid.

[0041] For sensor sampling intervals greater than the preset sampling period The data was used to estimate the values ​​at these missing sampling times using linear interpolation.

[0042] By aligning timestamps and unifying sampling frequencies as described above, we ensure that all variables are consistent at each standard sampling time. Each has a corresponding numerical value, forming a multidimensional time-series dataset that is completely synchronized in time, providing a data foundation for subsequent difference calculation and feature extraction.

[0043] During actual operation, sensors may generate invalid data due to reasons such as disconnection, signal fluctuations, or electromagnetic interference. These outliers reflect malfunctions in the measurement system itself, rather than actual changes in the physical state of the equipment. If directly used in subsequent calculations, they will contaminate the difference sequence and affect the accuracy of health assessments.

[0044] Therefore, based on the sensor's range and physical limits, if a measured value at a given moment exceeds the physically possible range of the variable, it is identified as an outlier and removed. Based on the rate of change of measured values ​​between adjacent sampling moments, if the rate of change of a measured value at a given moment compared to the previous moment exceeds the physically achievable maximum rate of change of the variable, it is identified as an outlier and removed.

[0045] For missing data points resulting from outlier removal and missing values ​​in the original data, linear interpolation is used to fill in the missing data. For time intervals... Missing values.

[0046] in, The measurement value is from the previous moment. The measurement value at the next moment. This is the timestamp of the currently missing moment.

[0047] Through the above processing, outliers caused by measurement system malfunctions are eliminated, and the true degradation signals of the equipment are fully preserved, providing an effective data foundation for subsequent difference calculations and health assessments.

[0048] Different physical quantities in the operating data of an air source heat pump have different dimensions and orders of magnitude. For example, the temperature is usually in the range of -20℃ to 50℃, the pressure is in the range of 0.1MPa to 4.0MPa, and the power is in the range of 0kW to 50kW.

[0049] If the original values ​​are used directly for subsequent calculations, variables with larger value ranges will dominate the calculations, leading to distorted evaluation results.

[0050] This embodiment employs the min-max normalization method to uniformly map all variables to the [0,1] interval, eliminating differences between different physical dimensions and orders of magnitude. The normalization formula is: in, For the first The variables at the sampling time The original value, and The first The minimum and maximum values ​​of each variable in historical health operation data. This is the normalized value.

[0051] Through the above normalization process, the numerical range of all variables is uniformly scaled to the [0,1] interval, eliminating the differences between different physical dimensions and orders of magnitude. This ensures that each variable has the same weight scale in subsequent difference calculation, feature extraction, and Mahalanobis distance evaluation, thereby guaranteeing the objectivity and accuracy of the evaluation results.

[0052] S2: Predictive Model State Prediction like Figure 2 As shown, the preprocessed thermodynamic state variables, performance state variables, control variables, and disturbance variables are input into the state prediction equation to obtain the predicted values ​​of the thermodynamic state variables and performance state variables for the current prediction period. Then, the predicted values ​​of the thermodynamic state variables and performance state variables are input into the output equation to obtain the predicted values ​​of the auxiliary observations for the current prediction period.

[0053] In this embodiment, the prediction model is a nonlinear state-space model constructed based on thermodynamics and the law of conservation of mass. To meet the real-time recursive calculation requirements of the embedded controller, the nonlinear model needs to be linearized and discretized.

[0054] First, based on the outdoor ambient temperature Outdoor relative humidity And the thermal resistance of frost layer in the performance state variables The range of values ​​is used to divide the entire operating space of the air source heat pump into multiple operating condition subspaces. Each operating condition subspace corresponds to a set of continuous thermodynamic state variable intervals, performance state variable intervals, control variable intervals, and disturbance variable intervals. Within each operating condition subspace, a characteristic point is selected. The characteristic points are composed of typical thermodynamic state variable values, typical performance state variable values, typical control variable values, and typical disturbance variable values ​​within the subspace. Typical values ​​can be the median, mean, or representative values ​​calibrated experimentally within the subspace.

[0055] At the feature points of each working condition subspace, a first-order Taylor expansion of the nonlinear function is performed, retaining the linear terms, to obtain the continuous linear state prediction model corresponding to the subspace: in, For the parameter set of thermodynamic state variables and performance state variables, To control the variable vector, For the perturbation variable vector, It is a continuous state matrix that describes the dynamic evolution of the state variables themselves. Given a continuous input matrix, describe the influence of control variables on state variables. This is a continuous perturbation matrix that describes the impact of perturbation variables on state variables.

[0056] For the continuous linear state prediction model of each working condition subspace, the zero-order hold method is adopted with a preset sampling period. Discretization yields the discrete state prediction equations for each subspace: in, for Predicted values ​​of periodic thermodynamic state variables and performance state variables. for Periodic thermodynamic state variables and performance state variables, for The periodic control variable vector, for A periodic vector of perturbation variables. Let be a discrete state transition matrix, describing the state variables from... Period End The dynamic evolution of the cycle itself, Given a discrete input matrix, it describes the influence of control variables on state variables. This is a discrete perturbation matrix, describing the impact of perturbation variables on state variables. This is the process noise vector, used to characterize model uncertainty.

[0057] The above The matrix is ​​obtained by interpolating from the pre-stored operating condition grid at the current operating condition point, thereby realizing gain scheduling across the entire operating condition range. That is, it automatically switches to the state space matrix that matches the operating condition under different operating conditions, ensuring that the prediction model has high prediction accuracy across the entire operating condition range.

[0058] State vector The dynamic differential equations corresponding to each component are as follows, and these equations together form the basis of the nonlinear state-space model: Refrigerant mass flow equation: in, This represents the refrigerant mass flow rate (kg / s). For compressor volumetric efficiency, The compressor operating frequency (Hz) The theoretical displacement of the compressor (m) 3 / r), which is an inherent parameter of the device. The specific volume of the compressor suction gas (m³) 3 / kg), derived from the measured inhalation pressure Compressor suction temperature Real-time calculations using refrigerant property equations This is a correction factor for the refrigerant charge amount.

[0059] Compressor volumetric efficiency The calculation formula is: in, This serves as the baseline for compressor volumetric efficiency; the initial value is determined by the factory calibration. This is the pressure ratio correction factor, calibrated by prototype testing. The compressor discharge pressure (MPa) Here is the compressor suction pressure (MPa). According to the formula, the compressor volumetric efficiency varies with the pressure ratio. The increase leads to a decrease. This reflects the decrease in compressor volumetric efficiency benchmark caused by wear and tear from long-term operation.

[0060] Equations for heat exchange between evaporator and condenser: in, Heat absorbed by the refrigerant side of the evaporator (kW). Heat released on the refrigerant side of the condenser (kW). The specific enthalpy of the refrigerant at the compressor suction port (kJ / kg) is determined by the suction pressure. With intake temperature Real-time calculations using refrigerant property equations The specific enthalpy of the refrigerant at the evaporator outlet (kJ / kg) is calculated based on the evaporation pressure and the outlet dryness fraction. The specific enthalpy of the refrigerant at the compressor discharge port (kJ / kg) is determined by the discharge pressure. With exhaust temperature calculate, The specific enthalpy of the refrigerant at the condenser outlet (kJ / kg) is determined by the condensing pressure and subcooling.

[0061] According to the formula, the heat absorbed by the refrigerant from the air side in the evaporator... The heat released by the refrigerant to the water side in the condenser is equal to the refrigerant mass flow rate multiplied by the enthalpy difference between the evaporator inlet and outlet. Similarly.

[0062] Dynamic equation for water outlet temperature: For heating operation: in, The water outlet temperature is (°C). Effective heating capacity (kW) is defined as , For condenser heat exchange efficiency, The specific heat capacity of water at constant pressure. The mass flow rate of the circulating water in the water circuit is (kg / s). The current outlet water temperature, For the return water temperature in the water circuit, The density of water, Water volume in the waterway (m³) 3 ).

[0063] The expression is ,in The heat exchange efficiency is the factory standard. The condenser heat transfer attenuation coefficient is... Due to the inherent thermal resistance of the condenser, To reduce the thermal resistance of dirt on the water path side.

[0064] According to the formula, the dynamic equation for outlet water temperature describes the heat balance on the water path side and the effective heating capacity. The circulating water is heated, and the water flow carries away heat. The difference between the two determines the rate of change of the outlet water temperature.

[0065] For cooling conditions: in, Effective cooling capacity (kW) is defined as follows: , The heat exchange efficiency of the evaporator is expressed as follows: ,in The heat exchange efficiency is the factory standard. This refers to the evaporator fouling factor. Due to the inherent thermal resistance of the evaporator, This is for the thermal resistance of the frost layer.

[0066] Evaporation temperature dynamic equation: in, The evaporation temperature is ℃. Heat (kW) transferred to the evaporator from the air side. The internal volume of the evaporator (m³) 3 ), Density of refrigerant inside the evaporator (kg / m³) 3 ), The specific heat capacity of the refrigerant at constant pressure in the evaporator (kJ / (kg·K)). The total heat capacity of the refrigerant in the evaporator, i.e., the energy storage coefficient required for temperature change.

[0067] Dynamic equation for condensation temperature: in, The condensation temperature is ℃. Heat released on the refrigerant side of the condenser (kW). The heat exchange capacity (kW) on the water circuit side of the condenser is defined as follows: That is, the heat power carried away by the circulating water from the condenser; The internal volume of the condenser (m³) 3 ), The density of the refrigerant inside the condenser (kg / m³) 3 ), The specific heat capacity of the refrigerant at constant pressure in the condenser (kJ / (kg·K)). The total heat capacity of the refrigerant within the condenser determines the sensitivity of the condensing temperature to heat power imbalance.

[0068] Dynamic equation for compressor suction temperature: in, The compressor suction temperature (°C) is the air temperature at which the compressor draws air. This represents the refrigerant mass flow rate (kg / s). The specific enthalpy of the refrigerant at the evaporator outlet (kJ / kg) is determined by the evaporation pressure and the outlet superheat. The actual specific enthalpy at the inlet (kJ / kg) is determined by the current inlet temperature. and inhalation pressure Calculated using physical property equations The combined heat transfer coefficient between the intake pipe and the ambient air (kW / (m²)) 2 ·K), calibrated by prototype testing, The heat exchange area of ​​the outer surface of the intake pipe (m²) 2 ), The outdoor ambient temperature is (°C). It is the total heat capacity (kJ / K) of the metal in the suction line and the refrigerant inside the line. This is the isobaric specific heat capacity of the refrigerant at suction pressure (kJ / (kg·K)).

[0069] Dynamic equation for condenser outlet temperature: in, The refrigerant temperature at the condenser outlet (°C) is given. The isobaric specific heat capacity of the refrigerant at condensation pressure (kJ / (kg·K)). The condensation temperature is ℃. The condenser outlet temperature is (°C). The heat transfer coefficient between the subcooled section piping and the ambient air (kW / (m²)) 2 K), determined by experiment; The heat exchange area of ​​the subcooling section (m²) 2 ), Ambient temperature (°C); This is the total heat capacity (kJ / K) of the subcooled section piping and the refrigerant inside. The specific heat capacity of the refrigerant at constant pressure in the subcooling section (kJ / (kg·K)).

[0070] Dynamic equation for thermal resistance of frost layer: in, For the thermal resistance of the frost layer, This is a frost rate correction factor used to adapt to frost intensity under different environmental conditions. The air-side mass flow rate (kg / s) passing through the evaporator is calculated in real time by an independent soft-sensor model. Moisture content of ambient air (kg / kg dry air), determined by ambient temperature. With relative humidity Calculated using the wet air property equation The saturated moisture content (kg / kg dry air) on the evaporator surface is determined by the evaporator wall temperature. calculate, Frosting efficiency reflects the proportion of water vapor actually deposited as frost, and is corrected by Reynolds number and frost thickness.

[0071] Frosting trigger conditions must be met simultaneously: evaporation temperature Environmental humidity Higher than surface saturation moisture content Evaporator wall temperature Below the ambient air frost point temperature .

[0072] Evaporator wall temperature The calculation formula is: in The refrigerant side of the evaporator absorbs heat. Due to the inherent thermal resistance of the evaporator, This refers to the heat exchange area.

[0073] According to the formula, in the heating frosting branch, the rate of increase in the thermal resistance of the frost layer is proportional to the air-side water vapor migration flux. and frosting efficiency And by correction factor In the adaptive adjustment and defrosting branch, the thermal resistance of the frost layer decreases at a rate proportional to the superheat of the wall, and the thermal resistance of the frost layer remains constant under cooling or frost-free heating conditions.

[0074] The baseline dynamic equation for the combined fouling thermal resistance of the heat exchanger and the volumetric efficiency of the compressor is as follows: in, To reduce the overall fouling thermal resistance of the heat exchanger, As a benchmark for compressor volumetric efficiency, The growth rate coefficient of fouling thermal resistance (unit: m) 2 ·K / (W·s)), The decay rate coefficient of volumetric efficiency (unit: s) -1 Both are positive numbers, calibrated by statistical analysis of long-term equipment operation data, and their numerical range is verified in prototype testing.

[0075] Without external intervention, fouling thermal resistance increases linearly and slowly over time, while volumetric efficiency decreases linearly and slowly over time. Since these changes are negligible within a single control cycle, in practical discretization solutions, their rates of change are typically not integrated in real-time, but rather updated once at relatively long intervals through batch parameter updates, i.e., in the equation... and It is not used as a real-time recursive parameter, but only to describe long-term trends and serve as a basis for offline calibration.

[0076] By combining the dynamic differential equations of the above thermodynamic state variables and performance state variables, a nonlinear state-space model is formed.

[0077] State vector Each component corresponds to a first-order differential equation, where For thermodynamic state variables, These are performance state variables. The right-hand side terms of each equation are nonlinear functions of the thermodynamic state variables, performance state variables, control variables, and disturbance variables. All equations are summarized in the following vector form: in, The function is a nonlinear vector function, and its specific expression is given by arranging all equations in the order of state components. The nonlinear model describes the unified dynamic behavior of the air source heat pump under three operating conditions: heating, cooling, and defrosting.

[0078] During online operation, the controller identifies the current operating condition subspace based on the thermodynamic state variables, performance state variables, control variables, and disturbance variables of the current prediction period, and selects the discrete state prediction model corresponding to that subspace as the prediction model for the current prediction period. Through recursive calculation of the state prediction equations, the predicted values ​​of the thermodynamic state variables and performance state variables for the current prediction period are obtained, denoted as... ,in express Predicted values ​​of thermodynamic state variables and performance state variables for the period.

[0079] After completing the recursive calculation of the state prediction equation, we obtain Predicted values ​​of periodic thermodynamic state variables and performance state variables To obtain the predicted values ​​of the auxiliary observations, the predicted values ​​of the thermodynamic state variables and performance state variables are input into the output equation. The output equation is constructed based on the observation matrix and the physical relationships between each auxiliary observation and the thermodynamic and performance state variables.

[0080] The discrete form of the output equation is: in, for The periodic auxiliary observation predicted value vector, for Predicted values ​​of periodic thermodynamic state variables and performance state variables. The observation matrix is ​​defined by the physical relationships between auxiliary observations and state variables.

[0081] For example, the total input power of the machine The predicted value can be obtained from the predicted values ​​of thermodynamic state variables and performance state variables through a power calculation correlation. (Compressor exhaust temperature) The predicted value can be calculated based on the predicted values ​​of thermodynamic state variables combined with the thermodynamic equations of the compressor compression process, including the evaporator pressure drop. The predicted value can be calculated based on the air-side flow rate and evaporator state parameters.

[0082] Based on the aforementioned state prediction equations and output equations, the prediction model outputs predicted values ​​of thermodynamic state variables, performance state variables, and auxiliary observations in the current prediction period, which are then used for subsequent difference calculations and health assessments.

[0083] S3: Difference Sequence Generation and Decomposition like Figure 3 As shown, the difference between the thermodynamic state variable and the predicted value of the thermodynamic state variable, as well as the difference between the auxiliary observation and the predicted value of the auxiliary observation, are determined respectively, so as to obtain the difference of each variable included in the thermodynamic state variable and the auxiliary observation within the current prediction period.

[0084] For each variable in the thermodynamic state variables and auxiliary observations Calculate variables Sensor measured values With variables Predicted value The difference between them: in, For variables In the current forecast period The difference, For variables Sensor measured values ​​within the current prediction period. For variables Forecast values ​​within the current forecast period.

[0085] The above calculations yield the difference between the thermodynamic state variables and each variable included in the auxiliary observations within the current prediction period. The sign of the difference reflects the direction of deviation of the measured value from the predicted value; a positive difference indicates that the measured value is higher than the predicted value, and a negative difference indicates that the measured value is lower than the predicted value.

[0086] Based on the current forecast period and its predecessor Each period is the current time window. For each variable of the thermodynamic state variable and auxiliary observation, the difference of that variable in each prediction period within the time window is arranged in chronological order to obtain the difference sequence of that variable within the time window.

[0087] For each variable The difference sequence is as follows: in, For variables The difference sequence within the time window, For variables The difference within a single prediction period. The length of the difference sequence is... One cycle.

[0088] Based on the rate of change of the control variables, disturbance variables, and auxiliary observations used to characterize the operating condition, it is determined whether the operating condition within the time window is a steady-state condition or a dynamic condition. In this embodiment, the variables used to characterize the operating conditions among the control variables, disturbance variables, and auxiliary observations include the compressor operating frequency. Outdoor ambient temperature Mass flow rate of circulating water .

[0089] Determine the rate of change of the compressor operating frequency, the outdoor ambient temperature, and the mass flow rate of the circulating water in the water circuit within the time window, respectively. Compressor operating frequency change rate To obtain the difference between the maximum and minimum operating frequencies of the compressor within the time window, divide by the duration of the time window; Outdoor ambient temperature change rate To obtain the difference between the maximum and minimum outdoor ambient temperatures within the time window, divide by the duration of the time window; Water circulation water mass flow rate change rate To obtain the difference between the maximum and minimum mass flow rates of the circulating water within the time window, divide by the duration of the time window.

[0090] The current time window is considered to be in a steady-state condition when the following three conditions are met: Compressor operating frequency change rate Less than the first threshold For example, in this embodiment, ; Outdoor ambient temperature change rate Less than the second threshold For example, in this embodiment, ; Water circulation water mass flow rate change rate Less than the third threshold For example, in this embodiment, .

[0091] When the compressor operating frequency change rate is satisfied Greater than or equal to the first threshold The outdoor ambient temperature change rate Greater than or equal to the second threshold And the rate of change of the mass flow rate of the circulating water in the water circuit. Greater than or equal to the third threshold When at least one of the following conditions is met, the time window is determined to be a dynamic operating condition.

[0092] By determining the rates of change of compressor operating frequency, outdoor ambient temperature, and water circulation mass flow rate within a time window and comparing them with their respective preset thresholds, a steady-state operating condition is determined when all three are less than the threshold; otherwise, a dynamic operating condition is determined. This criterion comprehensively considers the compressor frequency, ambient temperature, and water flow rate of the heat pump, and can comprehensively and accurately reflect the stability of the operating condition. This ensures that the selection of subsequent difference decomposition methods has a clear and reasonable basis, avoiding degradation signal distortion caused by misjudgment of operating conditions.

[0093] If the time window is a steady-state condition, the difference sequence of each variable of the thermodynamic state variable and the auxiliary observation is subjected to low-pass filtering, and the low-frequency trend component is used as the equipment degradation sequence of that variable.

[0094] Under steady-state conditions, the compressor operating frequency Outdoor ambient temperature Mass flow rate of circulating water The parameters characterizing the operating conditions remain essentially constant. Under these conditions, the difference at each moment... Changes over time mainly come from two aspects: The slowly changing part: Irreversible degradation of equipment performance due to scaling, wear, dust accumulation, refrigerant leakage, etc. The time constants of these physical processes are usually on the order of hours, days, or even months, which manifest as low-frequency drift in the difference sequence.

[0095] Rapidly changing factors include sensor measurement noise, model linearization error, and random interference. These factors change rapidly and are highly random, manifesting as high-frequency fluctuations in the difference sequence.

[0096] The physical nature of equipment degradation dictates that its manifestation in the difference sequence is a low-frequency trend. Equipment performance does not change suddenly, but rather deteriorates gradually over time. Therefore, by using low-pass filtering to remove high-frequency fluctuations from the difference sequence and retaining the low-frequency trend component, a signal reflecting the true degradation state of the equipment can be obtained.

[0097] In this embodiment, a first-order low-pass filter is used to filter the difference sequence. The discrete recursive formula for the first-order low-pass filter is: in, For variables In the current forecast period The difference, For variables In the current forecast period The device degradation sequence value, These are the filter coefficients, relative to the cutoff frequency. Sampling period The relationship is .

[0098] Filter coefficients The value of determines the smoothness of the filter. The closer the value is to 1, the smoother the filter output, the lower the cutoff frequency, and the fewer frequency components are retained. For example, in this embodiment, the cutoff frequency of the low-pass filter... The value is 0.001Hz, corresponding to the filter coefficient. This means that only slowly changing components with a change period greater than approximately 16.7 minutes are retained. The time constant of device degradation is typically much greater than 16.7 minutes, so the degradation signal can pass through the low-pass filter completely; while the time constants of sensor noise and model error are typically much less than 16.7 minutes, so they are effectively suppressed during the filtering process.

[0099] Low-frequency trend components of the filter output That is, variables The device degradation sequence represents the long-term performance deviation of the device relative to a health baseline within the current time window.

[0100] The difference at each time step in the difference sequence With the corresponding device degradation sequence value The difference is a variable Model mismatch sequence values: Model mismatch sequence This includes high-frequency components that have been filtered out from the difference sequence, which are used for online correction of biases in subsequent prediction models.

[0101] Under steady-state conditions, the method for obtaining the model mismatch sequence was clarified by using the difference between the difference sequence of each variable and the equipment degradation sequence as the model mismatch sequence for that variable. Since the low-frequency components of the difference sequence under steady-state conditions represent equipment degradation and the high-frequency components represent noise and model errors, this difference operation can effectively separate the high-frequency model mismatch components, providing clean and reliable observation data for subsequent online correction of model coefficients, avoiding contamination of the correction process by equipment degradation signals, and improving the accuracy of model correction.

[0102] If the current time window is a dynamic operating condition, then the variable representing the operating condition is used as the independent variable, and the least squares method is used to perform regression analysis on the difference sequence of each variable, and the regression residual is used as the equipment degradation sequence of that variable.

[0103] Compressor operating frequency under dynamic operating conditions Outdoor ambient temperature Mass flow rate of circulating water The parameters characterizing the operating conditions undergo significant changes. These changes in operating conditions themselves cause tracking errors in the model's predictions. The prediction model switches between different operating condition subspaces through gain scheduling, and the linearization interpolation error during these switching processes is reflected in the differences, which are strongly correlated with the magnitude of the operating condition changes. Therefore, the difference at each time step... The changes mainly come from three aspects: The component strongly correlated with operating conditions: the model's tracking error under different operating conditions. When operating conditions change, this difference changes synchronously, showing a strong correlation with operating condition variables, such as compressor operating frequency, outdoor ambient temperature, and water circulation mass flow rate.

[0104] Components unrelated to operating conditions: performance degradation of the equipment itself. Regardless of changes in operating conditions, deviations caused by equipment degradation always exist and accumulate continuously, and have no direct synchronous relationship with changes in operating conditions.

[0105] Random components: sensor measurement noise, etc.

[0106] Under dynamic operating conditions, if low-pass filtering is directly used to extract equipment degradation signals, large fluctuations in the difference caused by changes in operating conditions will be misjudged as equipment degradation, leading to distorted assessments. Therefore, regression analysis is needed to remove the model tracking error, which is strongly correlated with changes in operating conditions, from the difference, retaining only the components unrelated to changes in operating conditions as equipment degradation signals.

[0107] For each variable Based on the compressor operating frequency Outdoor ambient temperature Mass flow rate of circulating water Establish a multiple linear regression model with as the independent variable: in, For variables In the current forecast period The difference, For the current forecast period The compressor operating frequency, For the current forecast period outdoor ambient temperature, Current forecast period Water circulation water mass flow rate The intercept term represents the fixed component of the difference that is independent of the operating conditions when all independent variables are zero. , , These are regression coefficients, which respectively reflect the variables. The difference in compressor operating frequency Outdoor ambient temperature Mass flow rate of circulating water The degree of sensitivity, The regression residuals represent the portion of the difference that cannot be explained by the operating condition variables.

[0108] The least squares method is used to solve for the regression coefficients of the data within the time window: in, This is a matrix of operating condition variables, including constant 1, , , , For variables The difference sequence vector within the time window, This is the vector of regression coefficients.

[0109] The regression fit value is the estimated value calculated by substituting the operating condition variables at each time point into the regression model: Regression fit value This represents the portion of the difference that changes synchronously with the operating conditions, i.e., the tracking error of the model under different operating conditions, and the regression residual. This represents the portion of the difference that is unrelated to changes in operating conditions, i.e., the fixed deviation caused by the degradation of the equipment's own performance.

[0110] The residual will revert to its original state. As a variable Device degradation sequence value: Regression fitted values As a variable Model mismatch sequence values: Through the above processing, each variable difference sequence The difference at each time step Decomposed into equipment degradation sequence and model mismatch sequence Two components. All moments. Arrange the device degradation sequence that constitutes this variable in chronological order, for all moments. Arranged chronologically, this constitutes the model mismatch sequence for that variable.

[0111] Equipment degradation sequences are used for subsequent health assessments and fault localization, while model mismatch sequences are used to drive online correction of prediction model biases.

[0112] Under dynamic operating conditions, least squares regression is performed on the difference sequences of each variable, using compressor operating frequency, outdoor ambient temperature, and circulating water mass flow rate as independent variables. The regression residuals are used as the equipment degradation sequence, and the regression fit values ​​are used as the model mismatch sequence. This scheme can separate the model tracking error, which is strongly correlated with changes in operating conditions, from the differences, so that the equipment degradation sequence retains only the inherent performance degradation components unrelated to operating conditions. This allows for accurate extraction of equipment degradation signals even under drastic changes in operating conditions, avoiding interference from operating condition fluctuations in health assessment. At the same time, the regression fit values, as the model mismatch sequence, provide a reasonable indication of the source of error for model correction under dynamic operating conditions. When any difference of any variable If the difference in the health value of this variable exceeds a preset multiple of the standard deviation of the health value, for example, in this embodiment, the preset multiple is 3 times, it is determined that there is a potential fault mutation in the current prediction period. At this time, all differences in this variable in the prediction period are included in the equipment degradation sequence value. and model mismatch sequence values Setting it to zero prevents the model mismatch sequence of this prediction period from being used for subsequent parameter updates and model corrections, thus avoiding the contamination of the prediction model by faulty data.

[0113] The standard deviation of the health difference is the standard deviation of the difference sequence of the equipment in a healthy state, which is obtained by statistical analysis of the operating data when the equipment is first put into operation and its health is confirmed. The trigger threshold of this fault protection mechanism can be determined based on the statistical analysis of the historical health operating data of the specific equipment. Those skilled in the art can make adaptive adjustments based on actual operating conditions, which does not constitute a limitation on the scope of protection of this disclosure.

[0114] S4: Feature Extraction and Health Assessment like Figure 4 As shown, features are extracted from the device degradation sequence of each variable, and the features include at least one of amplitude features, trend features, correlation features, and spectral features.

[0115] Equipment degradation sequence This reflects the long-term performance deviation of the device relative to a health baseline within the current time window. To transform the device degradation sequence into a quantifiable health indicator, it is necessary to extract multidimensional features reflecting the degradation state from the device degradation sequence for each variable.

[0116] In this embodiment, four types of features are extracted from the device degradation sequence of each variable: Amplitude features are used to characterize the deviation magnitude and energy level of the device degradation sequence. In this embodiment, amplitude features are quantified by calculating the mean, maximum value, and root mean square value of the device degradation sequence.

[0117] For variables Equipment degradation sequence The formula for calculating the mean is: in, The number of sampling periods within the time window. The first within the time window Each prediction period. The mean reflects the average deviation of the equipment degradation component from zero; the further the mean is from zero, the more severe the overall deviation of the equipment from a healthy state.

[0118] Maximum value The calculation formula is: The maximum value reflects the maximum deviation of the device degradation component within the time window and is used to capture peak anomalies during the degradation process.

[0119] Root mean square value The calculation formula is: The root mean square value reflects the energy magnitude of the equipment degradation component, comprehensively embodying the deviation range and the degree of fluctuation.

[0120] Trend features are used to characterize the direction and rate of change in the equipment degradation sequence. The first-order linear trend of the equipment degradation sequence for each variable is fitted using the least squares method. in, The trend slope reflects the rate and direction of deviation evolution. A positive value indicates that equipment degradation is accelerating. A negative value indicates that the degradation is slowing down or recovering. The intercept reflects the baseline deviation of the degraded sequence at time zero. This is the fitting residual.

[0121] Trend slope Solve using the least squares method: in, This represents the average time of each prediction period within the time window. This represents the mean of the equipment degradation sequence. The steeper the trend slope, the faster the equipment performance degrades.

[0122] Correlation features are used to characterize the coupling relationship between degradation sequences of different variables in equipment, and to distinguish the cooperative change patterns of different faults. Calculate any two variables and variables Pearson correlation coefficient between equipment degradation sequences: in, and Variables and variables The mean of the equipment degradation sequence, This represents the number of sampling periods within the time window. Correlation coefficient. The value range is [−1, 1]. A positive value indicates that the two variables are changing in the same direction, a negative value indicates that they are changing in opposite directions, and the closer the absolute value is to 1, the stronger the correlation.

[0123] Different types of faults can lead to different patterns of coordinated changes in variables. For example, evaporator scaling can cause a simultaneous decrease in evaporation temperature and an increase in evaporator pressure drop, while refrigerant leakage can cause a simultaneous increase in exhaust temperature and a decrease in suction pressure.

[0124] Spectral features are used to identify periodic components in device degradation sequences. For variables... The device degradation sequence is subjected to a Fast Fourier Transform (FFT) to convert the time-domain signal to the frequency domain, obtaining the variables. Device degradation sequences in the frequency domain Spectral energy distribution at: in, For the Fast Fourier Transform operator, For variables Device degradation sequences in the frequency domain The spectral energy distribution at that location.

[0125] The frequency domain complex results obtained from the Fast Fourier Transform Calculate the square of its modulus to obtain the power spectral density. Power spectral density reflects the energy distribution of the device degradation signal across its various frequency components.

[0126] The power spectral density is divided into three frequency bands according to frequency: low frequency band. Characterizing long-term degradation trends, mid-frequency band Characterizing periodic degradation patterns, such as periodic fluctuations caused by wear of rotating parts in compressors, and high-frequency bands. This characterizes random noise. The boundaries of each frequency band can be preset based on the physical characteristics of the equipment degradation signal.

[0127] For each frequency band , Calculate the total energy within the corresponding low-frequency, mid-frequency, and high-frequency bands. Then calculate the total energy of each frequency band. Total energy of all frequency bands Ratio: The set of energy proportions of each frequency band constitutes a variable. Spectral eigenvectors : The energy proportion of each frequency band reflects the distribution characteristics of the degraded signal at different time scales.

[0128] The features of all variables are combined into the feature vector of the time window. : in, Let the mean vector of the device degradation sequence for all variables be denoted as . Let the vector be the maximum value of the device degradation sequence for all variables. Let the vector be the root mean square value of the device degradation sequence for all variables. Let the trend slope vector be the vector of all variables. This is the upper triangular element vector of the correlation coefficient matrix between all pairs of variables. This is the spectral eigenvector of all variables.

[0129] The Mahalanobis distance between the feature vector and the preset health benchmark feature space is determined to obtain the comprehensive health score of the air source heat pump within the time window.

[0130] The health baseline feature space is pre-constructed in the following way: Feature vector samples of the air source heat pump were collected for multiple time windows during its healthy state. The healthy state refers to the steady-state operation period during which the equipment is first put into service and confirmed to be fault-free, during which no significant performance degradation occurs.

[0131] The mean vector is obtained by determining the average value of all feature vector samples in each feature dimension. : in, For the number of healthy samples, For the first The feature vector of a healthy sample.

[0132] Determine the covariance between every two feature dimensions in all feature vector samples to obtain the covariance matrix. : in, For the first The difference vector between the feature vector and the mean vector of a healthy sample, and this vector and its transpose. Multiplying them together yields a square matrix. The square matrix is ​​of the [missing value]. Line number Column elements are ,when When is the covariance between two different feature dimensions, when The variance of that feature dimension is calculated by summing the variances of all healthy samples and dividing by 0. This yields the covariance matrix. The diagonal elements represent the variance of each feature dimension, reflecting the degree of fluctuation of each feature under healthy conditions, while the off-diagonal elements represent the covariance between different feature dimensions, reflecting the correlation between features under healthy conditions.

[0133] Mean vector Covariance Matrix Together, they form the health baseline feature space. This health baseline feature space describes the statistical distribution characteristics that the feature vectors should possess under a healthy state. The mean vector defines the central location under healthy conditions, and the covariance matrix defines the allowable range and direction of variation of the feature vectors under healthy conditions.

[0134] Determine the Mahalanobis distance between the feature vector and the preset health benchmark feature space to obtain the comprehensive health score of the air source heat pump within the time window, including: The feature vector of the current time window With the health mean vector Subtracting them yields the difference vector. The difference vector reflects the direction and magnitude of the deviation of the current feature vector from the health center in each feature dimension.

[0135] Using the inverse of the covariance matrix For the difference vector Perform the transformation to obtain the transformed difference vector. The transformation is used to eliminate the dimensional differences and correlations between the feature dimensions, and the covariance matrix enables the comparison of different features on a uniform scale.

[0136] Transformed difference vector The square root of the sum of the squares of all elements in the matrix is ​​used to obtain the Mahalanobis distance. Mahalanobis distance measures the degree to which the current feature vector deviates from the healthy center; a larger distance indicates that the device is further away from a healthy state and the degradation is more severe. Unlike Euclidean distance, Mahalanobis distance considers the correlation between features and the variance differences between features, and can more accurately reflect the overall deviation of multidimensional feature vectors.

[0137] Mahalanobis distance Input a preset exponential function to obtain a comprehensive health score: in, A preset attenuation coefficient is used to control the rate at which the health score decreases with increasing Mahalanobis distance. For example, in this embodiment, The exponential function maps Mahalanobis distance to a value between 0 and 1. When the Mahalanobis distance is 0, the feature vector completely coincides with the health mean, and the health score is 1. As the Mahalanobis distance increases, the health score decreases exponentially, approaching 0.

[0138] The Mahalanobis distance is negatively correlated with the health score. The smaller the Mahalanobis distance, the closer the health score is to 1, indicating a better health status of the air source heat pump. The larger the Mahalanobis distance, the closer the health score is to 0, indicating a worse health status of the air source heat pump.

[0139] For each component of the air source heat pump, a preset mapping relationship is established between the component and the thermodynamic state variables and auxiliary observation quantities.

[0140] For example, in this embodiment, the mapping relationship between each component and the thermodynamic state variables and auxiliary observation quantities is as follows: Compressor subsystem: Compressor discharge temperature Compressor suction pressure Compressor discharge pressure Total input power .

[0141] Evaporator subsystem: Evaporation temperature evaporator pressure drop .

[0142] Condenser subsystem: Condensation temperature Outlet water temperature .

[0143] Fan subsystem: Evaporator pressure drop Outdoor fan speed .

[0144] Obtain the corresponding subset of variables for the component within the time window, and combine the features of the subset of variables into a feature vector subset, specifically: For each component Obtaining components The corresponding variable subset's device degradation sequence is used to extract multidimensional features from the device degradation sequence of each variable and combine them into a component. Corresponding feature vector subset .

[0145] Determine the subset of feature vectors The Mahalanobis distance between the component and the preset health reference feature space. Then, the Mahalanobis distance method, the same as the comprehensive health score, was used to determine the components. Corresponding feature vector subset With corresponding components Mahalanobis distance in the health baseline feature space ,part The health benchmark feature space consists of components The mean vector and covariance matrix of the feature vectors of the healthy samples corresponding to the variable subset are calculated and constructed. The components are then obtained through an exponential function mapping. Sub-item health score: in, The same preset attenuation coefficient as the overall health score.

[0146] Component health rating Similarly, values ​​between 0 and 1 indicate the healthier the component, while values ​​closer to 1 indicate more severe degradation.

[0147] The feature vector of the time window is input into a preset fault knowledge base for matching to obtain the fault type output by the fault knowledge base. A health status assessment report is generated based on the comprehensive health score, the health score of each component sub-item, and the fault type, and the health status assessment report is output.

[0148] By collecting feature vector samples from multiple time windows under healthy conditions, a health benchmark feature space is constructed by calculating the mean vector and covariance matrix. The covariance matrix characterizes the correlation and fluctuation degree among features under healthy conditions. During evaluation, the current feature vector is subtracted from the mean vector to obtain the difference vector. The difference vector is then calculated using the inverse of the covariance matrix. This distance not only considers the deviation of each feature from the health center but also eliminates correlations and dimensional differences between features, making the overall health score more objective and comprehensive. Finally, an exponential function maps the Mahalanobis distance to a continuous score between 0 and 1. A higher score indicates better health, and a lower score indicates more severe degradation, making it intuitive and convenient for threshold setting and trend monitoring in engineering applications.

[0149] In a preferred embodiment, the construction and matching process of the fault knowledge base is as follows.

[0150] The fault knowledge base is pre-built using the following methods: The prediction model under healthy conditions is used as the baseline model. All physical parameters in the baseline model are at their calibrated values ​​under healthy conditions; for example, the evaporator fouling factor... =0, condenser heat exchange efficiency attenuation coefficient The refrigerant charge correction factor is 0. The compressor volumetric efficiency benchmark is 1.0. It is 0.95.

[0151] For various preset fault types, such as evaporator ash accumulation, condenser scaling, refrigerant leakage, expansion valve jamming, compressor wear, fan failure, and insufficient water flow, the model coefficient offset corresponding to each fault type is input into the baseline model to perform fault simulation.

[0152] Specifically, model coefficient offset refers to the increment or decrement applied to the physical parameters characterizing component performance in the baseline model. The model coefficient offsets for each fault type are as follows: For evaporator ash accumulation faults, the evaporator fouling coefficient in the benchmark model is... Apply a positive offset, for example, in this embodiment, the evaporator fouling factor. The efficiency increases from 0 to 0.05–0.3. Evaporator heat exchange efficiency. The calculation formula is: in The heat exchange efficiency is the factory standard. This refers to the evaporator fouling factor. Due to the inherent thermal resistance of the evaporator, For the thermal resistance of the frost layer, To mitigate the thermal resistance of fouling. When the efficiency is increased, the evaporator heat exchange efficiency The decrease simulates the reduction in air-side heat exchange capacity of the evaporator due to dust accumulation.

[0153] For condenser scaling faults, the condenser heat transfer efficiency attenuation coefficient in the benchmark model is affected. Applying a positive offset, for example, in this embodiment, the condenser heat transfer efficiency attenuation coefficient. The efficiency increases from 0 to 0.05–0.3. (Condenser heat exchange efficiency) The calculation formula is: in The heat exchange efficiency is the factory standard. The condenser heat transfer attenuation coefficient is... Due to the inherent thermal resistance of the condenser, To mitigate the thermal resistance of fouling. When the efficiency is increased, the condenser heat exchange efficiency The decrease simulates the reduction in heat exchange capacity of a condenser due to scaling on the water side.

[0154] For refrigerant leakage faults, the refrigerant charge correction factor in the benchmark model is adjusted. Apply a negative offset; for example, in this embodiment, the refrigerant charge correction factor. Decrease from 1.0 to 0.5–0.8. Refrigerant mass flow rate. The calculation formula is: in, This is a correction factor for refrigerant charge. When reduced, refrigerant mass flow rate The decrease simulates insufficient refrigerant circulation due to refrigerant leakage.

[0155] For compressor wear failure, the compressor volumetric efficiency benchmark in the aforementioned benchmark model... Apply a negative offset; for example, in this embodiment, the compressor volumetric efficiency benchmark is... The volumetric efficiency of the compressor decreased from 0.95 to 0.6–0.8. The calculation formula is: in, When reduced, the compressor volumetric efficiency The decrease simulates the reduction in gas delivery efficiency caused by wear and tear from long-term operation of the compressor.

[0156] For wind turbine failures, the outdoor wind turbine speed in the benchmark model is... Apply a constraint offset; for example, in this embodiment, the outdoor fan speed... Limiting the speed to 20% to 30% of the rated speed simulates insufficient airflow caused by a fan malfunction.

[0157] For the fault of insufficient water flow, the mass flow rate of the circulating water in the reference model is... Apply a negative offset; for example, in this embodiment, the mass flow rate of the circulating water in the water circuit. Reducing the flow rate from the rated flow rate to 30%–50% simulates the decrease in heat exchange capacity caused by insufficient water-side flow.

[0158] The baseline model after applying the above model coefficient offset is used as the fault simulation model. The same operating conditions as the healthy state are input, such as the same outdoor ambient temperature, relative humidity, and water return temperature, to perform simulation calculations and generate simulation operation data corresponding to each fault type of the air source heat pump.

[0159] For each fault type, the simulation data is processed sequentially, including data acquisition and preprocessing, prediction model state prediction, difference sequence generation and decomposition, and feature extraction and combination. This process determines the feature vector of the simulation data, which is then used as the feature vector corresponding to that fault type. For each fault type, simulations are performed at different severity levels. The feature vectors of different severity levels under the same fault type are clustered to obtain the standard feature vector and its distribution range for that fault type.

[0160] The fault knowledge base is constructed based on the mapping relationship between each fault type and its corresponding feature vector. Each knowledge base record may include: fault type, corresponding feature vector, feature distribution range, fault mechanism description, and maintenance and handling suggestions.

[0161] By using a predictive model under healthy conditions as a baseline model, and inputting the model coefficient offsets corresponding to each of the preset fault types for fault simulation, simulation data is generated. The same data processing and feature extraction process is then followed to determine the feature vector corresponding to each fault type, thereby constructing a fault knowledge base. This approach does not require a large number of actual fault samples; it generates simulation data covering multiple fault types and varying degrees of severity simply by applying physical parameter offsets to the mechanistic model. This significantly reduces the cost and time required to build the knowledge base. Furthermore, the simulation data and online assessment use the same predictive model and feature extraction process, ensuring consistency between the knowledge base features and online features. This improves the accuracy and reliability of fault matching, providing a solid data foundation for rapid and automated fault diagnosis.

[0162] In a preferred embodiment, the state prediction equation includes updatable physical performance degradation parameters, which are parameters characterizing the performance degradation of air source heat pump components due to physical processes, and include fast-changing parameters and slow-changing parameters.

[0163] The rapid change parameter includes a frosting rate correction factor. This is used to correct the rate of increase in thermal resistance of the frost layer during the frosting process. The slowly varying parameters include the evaporator fouling coefficient and the condenser heat transfer efficiency decay coefficient. The evaporator fouling coefficient is used to correct the evaporator heat transfer efficiency in the state prediction equation, and the condenser heat transfer efficiency decay coefficient is used to correct the condenser heat transfer efficiency in the state prediction equation.

[0164] In this embodiment, the equipment degradation sequence includes: a sequence corresponding to evaporation temperature, a sequence corresponding to evaporator pressure drop, and a sequence corresponding to outlet water temperature.

[0165] The rapid parameter update process is as follows: Within each first update cycle, obtain the start time of the first update cycle. and end time The start time and end time The values ​​corresponding to the evaporation temperature sequence and the values ​​corresponding to the evaporator pressure drop sequence are respectively input into the preset frost thermal resistance soft measurement model to obtain the frost thermal resistance soft measurement value at the initial time. And the soft measurement value of the thermal resistance of the frost layer at the end time .

[0166] In this embodiment, the expression for the soft measurement model of frost thermal resistance is: in, The thermal resistance of the frost layer is measured in soft-measured values ​​(m). 2 (·K / W) For evaporator pressure drop, The outdoor ambient temperature is (°C). The evaporation temperature is ℃. Outdoor relative humidity (%) , , The fitting coefficients are calibrated through wind tunnel frosting tests.

[0167] The soft measurement value of the frost thermal resistance at the end time. Compared with the soft measurement value of frost thermal resistance at the initial moment The difference is taken as the first change in thermal resistance of the frost layer output by the soft measurement model of the frost layer thermal resistance. : Obtain the state prediction equation at the initial time. Predicted thermal resistance of frost layer output and end time Predicted thermal resistance of frost layer output The predicted value of the frost thermal resistance at the end time. Predicted thermal resistance of frost layer at the start time The difference is used as the change in thermal resistance of the second frost layer predicted by the state prediction equation. : Determine the difference between the change in thermal resistance of the first frost layer and the change in thermal resistance of the second frost layer: Difference Reflects the current frost rate correction factor The estimation error, if This indicates that the actual frost layer grew faster than the model predicted. Too small; if This indicates that the actual frost layer growth was slower than the model expected. Too large.

[0168] With the difference The observed value is corrected by the correction factor for the frost rate. Using the variable as the independent variable, an observation equation is constructed. The recursive least squares method with a forgetting factor is used to solve the observation equation, yielding the correction amount of the frost rate correction coefficient within the current first update cycle. .

[0169] Updated Frost Rate Correction Factor The dynamic equation for frost thermal resistance, substituted into the state prediction equation, is used to calculate the frost thermal resistance in subsequent prediction cycles. Through the above recursive process, It can quickly track real-time changes in frost intensity, ensuring that the prediction model's forecast of frost thermal resistance remains accurate.

[0170] The slow-varying parameter update process is as follows: Within each second update cycle, it is first determined whether the current operating conditions meet the steady-state operating conditions. If the update conditions are met within the current second update cycle, the sequence corresponding to the evaporation temperature is extracted. Sequence corresponding to evaporator pressure drop and the sequence corresponding to the outlet water temperature As an observation vector, it includes: Sequence corresponding to evaporation temperature ; Sequence corresponding to evaporator pressure drop ; Sequence corresponding to outlet water temperature .

[0171] in, This represents the number of sampling points within the current second update cycle. The three sequences mentioned above are processed at the same time... The values ​​are combined into an observation vector. : Assume the slowly changing parameter vector to be updated is The current slowly varying parameter value at the start of the second update cycle is The correction amount for the slowly varying parameters to be solved in this update is: Observation vector With slow-varying parameter correction amount The linearized observation equation between them is: in, The current sampling time The sensitivity matrix, whose elements reflect the sensitivity of each component in the observation vector to each slowly varying parameter, can be obtained in real time through numerical perturbation or automatic differentiation.

[0172] With observation vector As the dependent variable, the correction amount of the slowly varying parameter. Using the independent variable, construct the following parameter optimization objective function: in, It is the square of the second norm of the vector. This is the fitting error term, used to measure the overall deviation between the model's predicted values ​​and the observed values ​​under the current slowly varying parameter values. The second term... This is a regularization term used to constrain the magnitude of parameter adjustments, preventing drastic changes in parameters during a single update.

[0173] The correction amount for slowly varying parameters is solved using the batch least squares method: in, The correction amount for the slowly varying parameter to be solved. To and The identity matrix of degree matching, This is the regularization coefficient.

[0174] The updated slow-varying parameters are: For example, in this embodiment, the updated slowly varying parameters satisfy the following physical constraint boundaries: , If the update exceeds the boundary, clamp to the boundary value and log the exception.

[0175] Updated slow-varying parameters and That is, the evaporator heat transfer efficiency is included in the state prediction equation. and condenser heat exchange efficiency The calculation formula enables the prediction model to continuously track the performance degradation of equipment caused by long-term processes such as dust accumulation and scaling.

[0176] It should be noted that since the change in frost thermal resistance is mainly affected by ambient humidity and temperature, its time constant is typically on the order of minutes, requiring a rapid response; while the decay of dirt thermal resistance and compressor volumetric efficiency baselines stems from long-term cumulative effects, with time constants typically on the order of months or years, and does not require high-frequency updates. Therefore, in this embodiment, the first update cycle is shorter than the second update cycle to adapt to the different evolution rates of fast-changing degradation parameters and slow-changing degradation parameters, ensuring the accuracy of frost formation prediction while avoiding unnecessary numerical disturbances to slow-changing parameters due to high-frequency updates.

[0177] By classifying the physical performance degradation parameters in the state prediction equation into fast-changing and slow-changing parameters, and setting a first update cycle and a second update cycle respectively, with the first update cycle being shorter than the second update cycle, the frost rate correction coefficient is updated in real time using the difference between the first change in frost thermal resistance output by the frost thermal resistance soft measurement model and the second change in frost thermal resistance predicted by the state prediction equation. This allows the model to quickly track the real-time changes in frost intensity and ensure the prediction accuracy of key state quantities such as evaporation temperature and evaporator pressure drop during the frost process. For slow-changing parameters, under steady-state conditions, the equipment degradation sequence corresponding to evaporation temperature, evaporator pressure drop, and outlet water temperature is used as the observation value. Batch least squares is used for low-frequency updates, effectively adapting to the slow evolution rate of long-term cumulative effects such as fouling thermal resistance and compressor aging, and avoiding unnecessary numerical disturbances introduced by high-frequency updates. This scheme, through a multi-timescale frequency division update strategy, takes into account both the response speed of fast-changing physical processes and the stability of slow-changing physical processes, enabling the prediction model to maintain high-fidelity tracking of the actual physical state throughout the entire operating conditions and life cycle.

[0178] In a preferred embodiment, in each third update cycle, the model mismatch sequence is used as the observation value, and the correction amount of the model coefficients is used as the independent variable to construct the observation equation.

[0179] The model coefficients are the correlation coefficients in the state prediction equation, calibrated by prototype testing, and used to describe the thermo-coupling relationship between components. For example, in this embodiment, the model coefficients include: pressure ratio correction coefficients. The combined heat transfer coefficient between the intake duct and the ambient air The heat transfer coefficient between the subcooled section piping and the ambient air Empirical coefficients in the frosting efficiency formula wait.

[0180] The above coefficients were obtained through data fitting calibration in the prototype test. However, due to individual differences between the prototype and the mass production machine, as well as performance changes after long-term operation of the equipment, these coefficients may have deviations and need to be continuously calibrated through online correction.

[0181] The observation equation is: in, for Periodic model mismatch sequence value vector, for The sensitivity matrix for the periodicity of the model is composed of the partial derivatives of each model mismatch sequence value with respect to each model coefficient correction. It describes the degree of influence of the model coefficient correction on the model mismatch sequence. This is the correction vector of the model coefficients in the current third update cycle, which is the independent variable to be solved by the recursive least squares method. This is the residual term.

[0182] The observation equations are solved using the recursive least squares method to obtain the correction amounts for each model coefficient within the current correction period. The recursive formula is as follows: Kalman gain calculation: in, The covariance matrix of the previous third correction period, This is the sensitivity matrix for the current period. It is a forgetting factor.

[0183] New interest calculation: in, This is the model mismatch sequence value vector for the current period. ) is the model coefficient correction vector for the previous third correction cycle.

[0184] Parameter correction update: Covariance update: The correction amounts for each of the aforementioned model coefficients are respectively added to the current model coefficients: Complete the online correction of the model coefficients in the state prediction equation.

[0185] Through the above recursive process, the correlation coefficients in the state prediction equation calibrated by the prototype test can be continuously corrected online according to the model mismatch sequence, so that the prediction model's description of the thermo-mechanical coupling relationship between heat pump components can continuously approach the real physical behavior, thereby maintaining prediction accuracy throughout the entire life cycle.

[0186] By acquiring the model mismatch sequence for each variable and constructing an observation equation in each third update cycle using the model mismatch sequence as the observation value and the correction amount of the model coefficients as the independent variable, the correction amount of each model coefficient is obtained by recursive least squares method and superimposed on the current model coefficients, thus realizing online continuous correction of the model coefficients in the state prediction equation. This scheme can compensate for individual differences between prototypes and mass-produced machines, as well as the deviation of model coefficients caused by performance changes after long-term operation of the equipment. This allows the prediction model to continuously approach the actual physical behavior in describing the thermodynamic coupling relationship between heat pump components, thereby maintaining prediction accuracy throughout the entire life cycle and avoiding the drift of evaluation results caused by a fixed model.

[0187] This embodiment also provides an electronic device, including: at least one processor, and a memory communicatively connected to the at least one processor, the memory storing a computer program executable by the at least one processor, wherein when the computer program is executed by the at least one processor, the at least one processor implements the above-described energy-saving air source heat pump health status assessment method.

[0188] This embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described energy-saving air source heat pump health status assessment method.

[0189] The above description is merely a preferred embodiment of this disclosure and is not intended to limit the scope of protection of this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for assessing the health status of an energy-saving air source heat pump, characterized in that, The method is executed based on a preset prediction model, which includes a state prediction equation and an output equation. The method includes: S1: Data Acquisition and Preprocessing Within the current prediction period of the prediction model, the operating data of the air source heat pump is acquired. The operating data includes: thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. The thermodynamic state variable is a parameter directly measured by a sensor, which characterizes the current thermodynamic output state of the air source heat pump. The performance state variables are parameters obtained in real time through preset initial values ​​or through preset state estimators, which characterize the performance degradation state of the components of the air source heat pump caused by physical or chemical processes. The control variable is an actuator parameter that can be actively adjusted, the disturbance variable is an external environmental parameter that is uncontrollable but measurable, and the auxiliary observation is a parameter that is directly measured by a sensor, other than the thermodynamic state variable, the performance state variable, the control variable, and the disturbance variable. The operating data is preprocessed to obtain the preprocessed thermodynamic state variables, performance state variables, control variables, disturbance variables, and auxiliary observations. The preprocessing includes timestamp alignment and sampling frequency unification, outlier removal and missing value imputation, and normalization. S2: Predictive Model State Prediction The preprocessed thermodynamic state variables, performance state variables, control variables, and disturbance variables are input into the state prediction equation to obtain the predicted values ​​of the thermodynamic state variables and performance state variables for the prediction period output by the state prediction equation. The predicted values ​​of the thermodynamic state variables and the predicted values ​​of the performance state variables are input into the output equation to obtain the auxiliary observation prediction values ​​of the prediction period output by the output equation. S3: Difference Sequence Generation and Decomposition The difference between the thermodynamic state variable and the predicted value of the thermodynamic state variable, and the difference between the auxiliary observation and the predicted value of the auxiliary observation are determined respectively, so as to obtain the difference between the thermodynamic state variable and each variable included in the auxiliary observation within the prediction period; Using the prediction period and the N periods preceding the prediction period as the current time window, for each variable of the thermodynamic state variable and the auxiliary observation, the difference of the variable in each period within the time window is arranged in chronological order to obtain the difference sequence of the variable within the time window; Based on the rate of change of the control variable, the disturbance variable, and the variable used to characterize the operating condition in the auxiliary observation, determine whether the operating condition of the time window belongs to a steady-state condition or a dynamic condition. If the time window is a steady-state condition, then the difference sequence of each variable of the thermodynamic state variable and the auxiliary observation is subjected to low-pass filtering to obtain the equipment degradation sequence of each variable. If the time window is a dynamic operating condition, then the least squares method is used to perform regression analysis on the difference sequence of each variable to obtain the equipment degradation sequence of each variable. S4: Feature Extraction and Health Assessment Features are extracted from the device degradation sequence of each variable, the features including at least one of amplitude features, trend features, correlation features and spectral features, and the features of all variables are combined into a feature vector of the time window; Determine the Mahalanobis distance between the feature vector and the preset health benchmark feature space to obtain the comprehensive health score of the air source heat pump within the time window; For each component of the air source heat pump, according to the preset mapping relationship between the component and the thermodynamic state variables and auxiliary observation quantities, the corresponding variable subset of the component within the time window is obtained, the features of the variable subset are combined into a feature vector subset, and the Mahalanobis distance between the feature vector subset and the preset health benchmark feature space of the component is determined to obtain the component sub-item health score of the component within the time window. The feature vector of the time window is input into a preset fault knowledge base for matching to obtain the fault type output by the fault knowledge base. A health status assessment report is generated based on the comprehensive health score, the health score of each component sub-item, and the fault type, and the health status assessment report is output.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the model mismatch sequence for each of the variables; The state prediction equation includes model coefficients that can be corrected online; In each third update cycle, the model mismatch sequence is used as the observation value, and the correction amount of the model coefficients is used as the independent variable to construct an observation equation. The observation equation is used to describe the mapping relationship between the correction amount of the model coefficients and the model mismatch sequence. The observation equations are solved using the recursive least squares method to obtain the correction amounts for each model coefficient within the current correction period; The correction values ​​of each of the model coefficients are respectively superimposed onto the current model coefficients to complete the online correction of the model coefficients in the state prediction equation.

3. The method according to claim 2, characterized in that, The step of obtaining the model mismatch sequence for each variable includes: If the time window is a steady-state condition, for each variable, the difference between the difference sequence of the variable and the equipment degradation sequence of the variable is taken as the model mismatch sequence of the variable.

4. The method according to claim 2, characterized in that, The control variables, disturbance variables, and auxiliary observations used to characterize the operating conditions include: compressor operating frequency, outdoor ambient temperature, and water circulation water mass flow rate. The least squares method is used to perform regression analysis on the difference sequence of each variable to obtain the equipment degradation sequence of each variable, including: For each of the variables, the compressor operating frequency, the outdoor ambient temperature, and the water circulation mass flow rate are used as independent variables. The least squares method is used to perform regression analysis on the difference sequence of the variables to obtain the regression residuals. The regression residuals are used as the equipment degradation sequence of the variables. The step of obtaining the model mismatch sequence for each variable includes: For each variable, the regression fit value of that variable is used as the model mismatch sequence of that variable.

5. The method according to claim 1, characterized in that, The control variables, disturbance variables, and auxiliary observations used to characterize the operating conditions include: compressor operating frequency, outdoor ambient temperature, and water circulation mass flow rate; determining whether the operating conditions within the time window are steady-state or dynamic includes: Determine the rate of change of the compressor operating frequency, the outdoor ambient temperature, and the mass flow rate of the circulating water in the water circuit within the time window, respectively. When the rate of change of the compressor operating frequency is less than a first threshold, the rate of change of the outdoor ambient temperature is less than a second threshold, and the rate of change of the mass flow rate of the circulating water is less than a third threshold, the time window is determined to be a steady-state condition. The time window is determined to be a dynamic operating condition when at least one of the following conditions is met: the rate of change of the compressor operating frequency is greater than or equal to a first threshold; the rate of change of the outdoor ambient temperature is greater than or equal to a second threshold; and the rate of change of the mass flow rate of the circulating water is greater than or equal to a third threshold.

6. The method according to claim 1, characterized in that, The state prediction equation includes updatable physical performance degradation parameters, which are parameters characterizing the performance degradation of air source heat pump components due to physical processes. The physical performance degradation parameters include fast-changing parameters and slow-changing parameters. The rapidly changing parameters include a frosting rate correction coefficient, which is used to correct the rate of increase of the thermal resistance of the frost layer during the frosting process. The slowly changing parameters include an evaporator fouling coefficient and a condenser heat exchange efficiency decay coefficient. The evaporator fouling coefficient is used to correct the evaporator heat exchange efficiency in the state prediction equation, and the condenser heat exchange efficiency decay coefficient is used to correct the condenser heat exchange efficiency in the state prediction equation. The equipment degradation sequence includes: a sequence corresponding to evaporation temperature, a sequence corresponding to evaporator pressure drop, and a sequence corresponding to outlet water temperature; the performance state variables include frost thermal resistance. The method further includes: Within each first update cycle, obtain the start time and end time of the first update cycle; The values ​​of the sequence corresponding to the evaporation temperature and the sequence corresponding to the evaporator pressure drop at the start and end times are respectively input into the preset frost thermal resistance soft measurement model to obtain the frost thermal resistance value at the start and end times respectively. The difference between the frost thermal resistance value at the end time and the frost thermal resistance value at the start time is taken as the first frost thermal resistance change output by the frost thermal resistance soft measurement model. Obtain the predicted value of frost thermal resistance output by the state prediction equation at the start time and the predicted value of frost thermal resistance output at the end time. Use the difference between the predicted value of frost thermal resistance at the end time and the predicted value of frost thermal resistance at the start time as the second change in frost thermal resistance predicted by the state prediction equation. Determine the difference between the change in thermal resistance of the first frost layer and the change in thermal resistance of the second frost layer; Using the difference as the observed value and the correction amount of the frost rate correction coefficient as the independent variable, an observation equation is constructed. The observation equation is solved using the recursive least squares method with a forgetting factor to obtain the correction amount of the frost rate correction coefficient in the current first update cycle. In each second update cycle, and when steady-state conditions are met, the sequence corresponding to evaporation temperature, the sequence corresponding to evaporator pressure drop, and the sequence corresponding to outlet water temperature are used as observation values, and the batch least squares method is used to update the evaporator fouling coefficient and the condenser heat exchange efficiency decay coefficient. The first update cycle is shorter than the second update cycle.

7. The method according to claim 1, characterized in that, The preset health benchmark feature space includes: The feature vector samples of the air source heat pump in a healthy state are collected from multiple time windows. The average value of all feature vector samples in each feature dimension is determined to obtain the mean vector. The covariance between every two feature dimensions in all feature vector samples is determined to obtain the covariance matrix. The mean vector and the covariance matrix are used as the healthy baseline feature space. The process of determining the Mahalanobis distance between the feature vector and the preset health benchmark feature space to obtain the comprehensive health score of the air source heat pump within the time window includes: The feature vector of the time window is subtracted from the mean vector to obtain the difference vector. The difference vector is then transformed using the inverse of the covariance matrix to obtain the transformed difference vector. The square root of the sum of the squares of each element in the transformed difference vector is then taken to obtain the Mahalanobis distance. The Mahalanobis distance is input into a preset exponential function to obtain a value between 0 and 1 output by the exponential function. This value is used as the comprehensive health score, wherein the Mahalanobis distance is negatively correlated with the health score.

8. The method according to claim 1, characterized in that, The preset fault knowledge base is constructed in the following way: The prediction model under a healthy state is used as the baseline model; For multiple preset fault types, the offset of the model coefficient corresponding to each preset fault type is input into the benchmark model to perform fault simulation and generate simulation operation data corresponding to each fault type of the air source heat pump. For each type of fault, a feature vector of the simulation running data is determined based on the simulation running data, and the feature vector is used as the feature vector corresponding to the fault type. The fault knowledge base is constructed based on the mapping relationship between each fault type and the corresponding feature vector.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the energy-saving air source heat pump health status assessment method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the energy-saving air source heat pump health status assessment method as described in any one of claims 1 to 8.