An air conditioner refrigerant leakage detection method, device, system and air conditioner

Abstract

The application discloses an air conditioner refrigerant leakage detection method, device and system and an air conditioner. It relates to the field of refrigerant leakage detection. By constructing a state space model combined with refrigerant quality and refrigerant leakage rate, inputting real-time working condition parameters and measurable physical quantities into the model, obtaining the estimated value of refrigerant quality and refrigerant leakage rate and its covariance matrix, and quantifying the uncertainty information of the estimated value, the dynamic threshold value constructed according to the estimated value of refrigerant quality and its covariance matrix can be adaptively adjusted according to working condition fluctuations, and in combination with the criterion that the refrigerant loss is greater than the dynamic threshold value and the estimated value of the leakage rate is greater than zero, the real leakage and working condition disturbance can be effectively distinguished, and the sensitivity and robustness of the leakage detection are improved. At the same time, the target leakage level is divided based on the estimated value of the leakage rate, and a differentiated control strategy is executed, a complete closed loop from accurate perception to graded response is realized, and the safety and reliability of the air conditioning system under all working conditions are significantly improved.

Description

Technical Field

[0001] This application relates to the field of refrigerant leak detection technology, and in particular, to a method, apparatus, system, and air conditioner for detecting refrigerant leaks in air conditioning systems. Background Technology

[0002] Refrigerant leakage is a common fault in air conditioning systems, leading to decreased cooling / heating capacity, deterioration of energy efficiency ratio, and in severe cases, insufficient oil return to the compressor, excessively high exhaust temperature, or even compressor burnout. Existing air conditioning detection solutions rely on sensors to obtain indoor refrigerant concentration to determine whether there is a leak. For example, CN121782691A discloses a refrigerant leak detection device, an air conditioning system, and its control method and device. The refrigerant leak detection device includes: a control main board with multiple valve drive interfaces and at least one refrigerant leak detector interface for connecting multiple solenoid valves and refrigerant leak sensors respectively; the solenoid valves are located on the connecting pipe between the outdoor unit and the indoor unit of the air conditioner; a main power input terminal for receiving main power supply; and an external power module connected in parallel with the main power input terminal to the control main board. The control main board is configured to: based on refrigerant leakage and / or main power abnormality, control the external power module to provide emergency power to the control main board to drive multiple solenoid valves to operate synchronously. This invention addresses the shortcomings of existing refrigerant leak detection devices in terms of multi-valve synchronous drive and power failure emergency response without sacrificing system compactness, thus providing a reliable and efficient safety protection solution for air conditioning systems.

[0003] In addition, traditional refrigerant leak detection methods mainly rely on fixed threshold comparisons or current changes after a certain period of operation. These methods generally have the following drawbacks: (1) Lack of real-time estimation capability of refrigerant quality: It is difficult to quantitatively estimate the refrigerant quality using the sensors (pressure, temperature, current, etc.) already configured in the existing air conditioning system without adding additional hardware sensors; (2) Poor adaptability to operating conditions: Fixed thresholds cannot adapt to changes in ambient temperature, load rate, compressor frequency, etc., often causing false alarms or missed alarms. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, this application provides a method, device, system and air conditioner for detecting refrigerant leakage, so as to solve the problems that the existing refrigerant leakage detection methods require additional hardware sensors and adopt fixed threshold schemes, which are costly and have poor adaptability to operating conditions.

[0005] The technical solution adopted by this application to solve its technical problem is: Firstly, a method for detecting refrigerant leakage in air conditioning systems is provided, including: A state-space model is constructed, which includes state equations describing the evolution of refrigerant quality and refrigerant leakage rate over time, and observation equations describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The real-time operating parameters and the measurable physical quantities are input into the state-space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. A dynamic threshold is constructed based on the estimated refrigerant quality and its covariance matrix; A refrigerant leak is determined when the refrigerant loss exceeds the dynamic threshold and the estimated refrigerant leakage rate is greater than 0.

[0006] As an optional implementation of this application, the construction of the state-space model includes: Constructing discrete state equations:

[0007]

[0008] Where m(k) is the refrigerant mass at time k, m(k-1) is the refrigerant mass at time k-1, r(k) is the refrigerant leakage rate at time k, r(k-1) is the refrigerant leakage rate at time k-1, and Δt is the sampling period. The refrigerant mass process noise at time k-1, The refrigerant leakage rate at time k-1 represents the process noise. Based on the measurable physical quantities and operating parameters, a nonlinear observation equation is constructed:

[0009] in, Let z be the observation vector at time k. P s and P d These are the inhalation pressure and the exhaust pressure, T. s and T d These are the intake temperature and the exhaust temperature, respectively. Let be the compressor current, h be the nonlinear observation function, and u(k) be the input vector u at time k. , For compressor operating frequency, For the opening degree of the electronic expansion valve, Outdoor ambient temperature To measure noise, and , Let be the measurement noise covariance matrix of the system.

[0010] As an optional implementation of this application, the nonlinear observation function is obtained based on a pre-trained model, and the training steps of the model are as follows: Sample data were obtained by collecting measurable parameters under different refrigerant qualities and operating conditions. The input values ​​for each sample data are obtained, which consist of refrigerant quality, compressor operating frequency, electronic expansion valve opening degree, and outdoor ambient temperature. The input values ​​are input into the initial model to obtain the predicted values. The initial model includes an input layer, a hidden layer, and an output layer. The predicted values ​​are composed of intake pressure, exhaust pressure, intake temperature, exhaust temperature, and compressor current. The loss between the predicted and actual values ​​is calculated using mean squared error and L2 regularization. The parameters of the initial model are updated based on the loss until the condition is met.

[0011] As an optional implementation of this application, updating the parameters of the initial model based on the loss until the condition is met includes: Weights of each layer based on loss and bias The impact of the error signal is transmitted backward from the output layer to the input layer, and the parameters are corrected according to a preset step size. For the output layer, linear activation is used, and its error signal... Directly reflects the deviation between the predicted value and the actual value: ; For the hidden layer, the error signal of the next layer is inversely distributed according to the transpose of the weight matrix, and gated modulation is performed by combining the switching characteristics of the ReLU activation function:

[0012] in, This indicates that the elements at each corresponding position of two vectors or matrices are multiplied. For the first Layer activation output; This is a Boolean conditional function; it evaluates to true when the condition inside the parentheses is true. = 1, otherwise 0; For the input layer, the error signal of this layer is compared with the output of the previous layer. The outer product of and calculates the direction of weight adjustment, and an L2 regularization term is added to limit the weight magnitude. Here, for the first layer, is the input vector. The direction of bias adjustment is directly equal to the error signal of this layer, and the parameter update formula is as follows:

[0013]

[0014] in, For learning rate, The regularization coefficient is . Repeat the forward and backward propagation, and evaluate on the validation set after each round; if the validation loss no longer decreases and continues for the preset number of rounds, trigger the early stopping mechanism, save the optimal model parameters on the validation set, and complete the training.

[0015] As an optional implementation of this application, the step of inputting real-time operating parameters and the measurable physical quantity into the state-space model to obtain estimates of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix, includes: During stable operation, the initial value of refrigerant mass is obtained, the initial value of leakage rate is set to zero, and the initial covariance matrix is ​​set to a large diagonal matrix; Using the state estimate and covariance matrix from the previous time step, a one-step prediction is performed based on the state equation to obtain the prior state estimate and its prior covariance matrix at the current time step. The nonlinear observation function is linearized at the current prior estimation point to obtain the observation Jacobian matrix. The Kalman gain is calculated based on the prior covariance matrix and the observed Jacobian matrix; the residual between the real-time measurable physical quantity and the predicted observation value is calculated; the prior state estimate is corrected by weighting the residual with the Kalman gain to obtain the optimal estimate of the refrigerant mass and refrigerant leakage rate at the current moment; at the same time, the covariance matrix is ​​updated, and the estimated variances of the two and their correlation are output. Monitor the change in quality variance in the covariance matrix within the window. When it is continuously less than the threshold, the filter is determined to be converged. Record the current estimated value and the corresponding variance as the estimated value of the refrigerant quality and refrigerant leakage rate, as well as the covariance matrix of the two.

[0016] As an optional implementation of this application, the step of constructing a dynamic threshold based on the estimated value of the refrigerant quality and its covariance matrix includes: Obtain the mean value of refrigerant quality estimates over a period of time as a reference quality, and record the variance of quality estimates for the corresponding period as a reference covariance; At each sampling time, the difference between the reference quality and the current quality estimate is calculated to obtain the real-time quality loss. The reference covariance is added to the current quality estimate variance to approximate the total variance of the quality loss. The total standard deviation is obtained by taking the square root of the total variance, multiplying it by a preset confidence coefficient, and then adding a preset minimum detectable loss to form the adaptive threshold at the current moment.

[0017] As an optional implementation of this application, it also includes: The target leakage level is obtained based on the estimated value of the refrigerant leakage rate. The higher the estimated value of the refrigerant leakage rate, the higher the leakage level. Different control strategies are used for different leakage levels. Control is implemented based on the target control strategy corresponding to the target leakage level.

[0018] As an optional implementation of this application, the control based on the target control strategy corresponding to the target leakage level includes: When the target leakage level is a minor leakage, the target control strategy is to limit the compressor frequency operation to a first operating frequency, reduce the refrigerant circulation speed, and send a first prompt, wherein the first operating frequency = rated operating frequency * first preset coefficient, and the first preset coefficient is a positive number less than 1; When the target leakage level is moderate leakage, the target control strategy is to limit the compressor frequency operation to the upper limit of the second operating frequency, reduce the opening of the electronic expansion valve, and send a second prompt. The second operating frequency is equal to the rated operating frequency multiplied by the second preset coefficient, which is a positive number less than 1 and is less than the first preset coefficient. When the target leakage level is severe leakage, the target control strategy is to control the compressor to stop, close the electronic expansion valve, and send a third prompt.

[0019] Secondly, an air conditioning refrigerant leakage detection device is provided, comprising: The spatial model construction module is used to construct a state space model, which includes state equations describing the evolution of refrigerant quality and refrigerant leakage rate over time, and observation equations describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The output vector estimation module is used to input real-time operating parameters and the measurable physical quantities into the state space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. A dynamic threshold construction module is used to construct a dynamic threshold based on the estimated value of the refrigerant quality and its covariance matrix. The refrigerant leakage determination module is used to determine a refrigerant leak when the refrigerant loss exceeds the dynamic threshold and the estimated refrigerant leakage rate is greater than 0.

[0020] Thirdly, an air conditioning refrigerant leakage detection system is provided, including: At least one processor and at least one memory; The memory stores the executable instructions of the processor; The processor is configured to perform the air conditioning refrigerant leakage detection method described in any of the preceding descriptions.

[0021] Fourthly, an air conditioner is provided, including the aforementioned air conditioning refrigerant leakage detection system.

[0022] Beneficial effects: This application provides a method, device, system, and air conditioner for detecting refrigerant leaks. By constructing a state-space model that combines refrigerant quality and leakage rate, and inputting real-time operating parameters and measurable physical quantities into this model, estimates of refrigerant quality and leakage rate, along with their covariance matrix, are obtained. This quantifies the uncertainty of the estimates, overcoming the limitation of traditional methods that cannot quantify the confidence level of estimates. A dynamic threshold constructed based on the refrigerant quality estimate and its covariance matrix can adaptively adjust with fluctuations in operating conditions. Combined with the criterion that the refrigerant loss is greater than the dynamic threshold and the leakage rate estimate is greater than zero, it effectively distinguishes between actual leaks and operating condition disturbances, improving the sensitivity and robustness of leak detection. Simultaneously, by classifying target leak levels based on the leakage rate estimate and implementing differentiated control strategies, a complete closed loop from precise sensing to graded response is achieved, significantly improving the safety and reliability of the air conditioning system under all operating conditions. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of an air conditioning refrigerant leakage detection method provided in an embodiment of this application; Figure 2 This is a flowchart of a model training process provided in an embodiment of this application; Figure 3 This is a flowchart illustrating an embodiment of the present application that provides an estimate of refrigerant mass and refrigerant leakage rate based on a state-space model. Figure 4 This is a flowchart of a specific air conditioning refrigerant leakage detection method provided in an embodiment of this application; Figure 5 This is a schematic diagram of a three-layer MLP fitting nonlinear observation function provided in an embodiment of this application; Figure 6 This is a flowchart of an EKF algorithm estimation method provided in an embodiment of this application; Figure 7 This is a flowchart of a leakage detection method provided in an embodiment of this application; Figure 8 This is a schematic diagram of a leakage level assessment and graded control response provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an air conditioning refrigerant leakage detection device provided in an embodiment of this application; Figure 10 This is a schematic diagram of an air conditioning refrigerant leakage detection system provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Traditional refrigerant leak detection methods mainly rely on comparisons of fixed thresholds or judgments based on current changes after a certain period of operation. These methods generally have the following drawbacks: (1) Lack of real-time estimation capability of refrigerant quality: It is difficult to quantitatively estimate the refrigerant quality using the sensors (pressure, temperature, current, etc.) already configured in the existing air conditioning system without adding additional hardware sensors; (2) Poor adaptability to operating conditions: Fixed thresholds cannot adapt to changes in ambient temperature, load rate, compressor frequency, etc., often causing false alarms or missed alarms.

[0027] (3) Insufficient use of information: Existing sensors (suction / discharge pressure, evaporation / condensation temperature, compressor current, compressor frequency, electronic expansion valve opening, etc.) are only used for their own independent control loops, lacking a model framework for fusion estimation of refrigerant quality.

[0028] (4) Lack of risk classification: After the alarm, there is no quantitative leakage rate and risk level classification, and it is impossible to take differentiated protection strategies according to the severity of leakage, nor can it form a closed-loop linkage of "sensing-estimation-decision-control" with the air conditioning main control MCU.

[0029] Therefore, this application provides a new method that can estimate the remaining refrigerant mass and leakage rate in real time and quantitatively, and adaptively judge leakage based on state uncertainty. It can also classify the severity of leakage and trigger graded control protection. Moreover, it does not require additional hardware costs and can be implemented simply by upgrading the algorithm.

[0030] Reference Figure 1 This application provides a method for detecting refrigerant leakage in air conditioning systems, including: S11: Construct a state-space model, which includes state equations describing the evolution of refrigerant quality and refrigerant leakage rate over time, and observation equations describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The construction of the state-space model includes: Constructing discrete state equations:

[0031]

[0032] Where m(k) is the refrigerant mass at time k, m(k-1) is the refrigerant mass at time k-1, r(k) is the refrigerant leakage rate at time k, r(k-1) is the refrigerant leakage rate at time k-1, and Δt is the sampling period. The refrigerant mass process noise at time k-1, The refrigerant leakage rate at time k-1 represents the process noise. Based on the measurable physical quantities and operating parameters, a nonlinear observation equation is constructed:

[0033] in, Let z be the observation vector at time k. P s and P d These are the inhalation pressure and the exhaust pressure, T. s and T d These are the intake temperature and the exhaust temperature, respectively. Let be the compressor current, h be the nonlinear observation function, and u(k) be the input vector u at time k. , For compressor operating frequency, For the opening degree of the electronic expansion valve, Outdoor ambient temperature To measure noise, and , Let be the measurement noise covariance matrix of the system.

[0034] In a preferred implementation of this application, the nonlinear observation function is obtained based on a pre-trained model, such as... Figure 2 As shown, the training steps of the model are as follows: Sample data were obtained by collecting measurable parameters under different refrigerant qualities and operating conditions. The operating conditions included compressor operating frequency, electronic expansion valve opening, and outdoor ambient temperature. The measurable parameters included suction pressure, discharge pressure, suction temperature, discharge temperature, and compressor current.

[0035] The input values ​​for each sample data are obtained, which consist of refrigerant quality, compressor operating frequency, electronic expansion valve opening degree, and outdoor ambient temperature. The input values ​​are input into the initial model to obtain the predicted values. The initial model includes an input layer, a hidden layer, and an output layer. The predicted values ​​are composed of intake pressure, exhaust pressure, intake temperature, exhaust temperature, and compressor current. The loss between the predicted and true values ​​is calculated using mean squared error and L2 regularization; where the true value is obtained from the actual value in the sample data.

[0036] The parameters of the initial model are updated based on the loss until the condition is met.

[0037] Further, updating the parameters of the initial model based on the loss until the condition is met includes: Weights of each layer based on loss and bias The impact of the error signal is transmitted backward from the output layer to the input layer, and the parameters are corrected according to a preset step size. For the output layer, linear activation is used, and its error signal... Directly reflects the deviation between the predicted value and the actual value: ; For the hidden layer, the error signal of the next layer is inversely distributed according to the transpose of the weight matrix, and gated modulation is performed by combining the switching characteristics of the ReLU activation function:

[0038] in, This indicates that the elements at each corresponding position of two vectors or matrices are multiplied. For the first Layer activation output; This is a Boolean conditional function; it evaluates to true when the condition inside the parentheses is true. = 1, otherwise 0; For the input layer, the error signal of this layer is compared with the output of the previous layer. The outer product of and calculates the direction of weight adjustment, and an L2 regularization term is added to limit the weight magnitude. Here, for the first layer, is the input vector. The direction of bias adjustment is directly equal to the error signal of this layer, and the parameter update formula is as follows:

[0039]

[0040] in, For learning rate, The regularization coefficient is used. Repeat the forward and backward propagation, and evaluate on the validation set after each round; if the validation loss no longer decreases and continues for the preset number of rounds, trigger the early stopping mechanism, save the optimal model parameters on the validation set, and complete the training.

[0041] S12: Input the real-time operating parameters and the measurable physical quantities into the state-space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. As a preferred implementation of the embodiments of this application, such as Figure 3 As shown, the step of inputting real-time operating parameters and the measurable physical quantities into the state-space model to obtain estimates of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix, includes: During stable operation, the initial value of refrigerant mass is obtained, the initial value of leakage rate is set to zero, and the initial covariance matrix is ​​set to a large diagonal matrix; Using the state estimate and covariance matrix from the previous time step, a one-step prediction is performed based on the state equation to obtain the prior state estimate and its prior covariance matrix for the current time step; for example, the system state... Prior state estimation vector at time step (estimated before update) and prior state estimation covariance matrix for:

[0042]

[0043] in, for The posterior (updated estimate) state estimate vector at time step [time]. for The covariance matrix of the posterior state estimate at time t.

[0044] Linearize the nonlinear observation function at the current prior estimate point to obtain the observation Jacobian matrix; for example, calculate the current state ( The observed Jacobian matrix at the prediction point (time) :

[0045] in, for The prior estimate of refrigerant quality at time t.

[0046] The Kalman gain is calculated based on the prior covariance matrix and the observed Jacobian matrix; the residual between the real-time measurable physical quantity and the predicted observation value is calculated; the prior state estimate is corrected by weighting the residual with the Kalman gain to obtain the optimal estimate of the refrigerant mass and refrigerant leakage rate at the current moment; at the same time, the covariance matrix is ​​updated, and the estimated variances of the two and their correlation are output. Monitor the change in quality variance in the covariance matrix within the window. When it is continuously less than the threshold, the filter is determined to be converged. Record the current estimated value and the corresponding variance as the estimated value of the refrigerant quality and refrigerant leakage rate, as well as the covariance matrix of the two.

[0047] For example, the current time is obtained by the following steps. The optimal state estimate (i.e., the posterior state estimate). ( and They are respectively Estimates of the total refrigerant mass and refrigerant leakage rate at time t(s) and their posterior state estimation covariance matrix. ,in The elements contained therein and These correspond to the covariance of the posterior state estimation for refrigerant quality and leakage rate, respectively. for Kalman gain at time step for The control input vector at time t, It is an identity matrix.

[0048]

[0049]

[0050]

[0051] Posterior state estimation covariance matrix The uncertainty of the estimate is quantified, and one sign of its convergence is that the diagonal elements of the covariance matrix no longer decrease continuously, but instead revolve around a steady-state value. Slight fluctuations. This invention utilizes To determine whether EKF converges, the following changes are used:

[0052] in, To observe the window length, For a very small threshold, when the change is less than the threshold and persists for a period of time, the covariance is considered to have stabilized, and EKF has reached a convergent state. S13: Construct a dynamic threshold based on the estimated value of the refrigerant quality and its covariance matrix; The construction of a dynamic threshold based on the estimated refrigerant quality and its covariance matrix includes: Obtain the mean value of refrigerant quality estimates over a period of time as a reference quality, and record the variance of quality estimates for the corresponding period as a reference covariance; At each sampling time, the difference between the reference quality and the current quality estimate is calculated to obtain the real-time quality loss. The reference covariance is added to the current quality estimate variance to approximate the total variance of the quality loss. The total standard deviation is obtained by taking the square root of the total variance, multiplying it by a preset confidence coefficient, and then adding a preset minimum detectable loss to form the adaptive threshold at the current moment.

[0053] S14: When the refrigerant loss is greater than the dynamic threshold and the estimated refrigerant leakage rate is greater than 0, a refrigerant leak is determined.

[0054] The refrigerant loss is the same as the mass loss mentioned above.

[0055] To ensure accurate judgment, a refrigerant leak is determined when the refrigerant loss exceeds the dynamic threshold and this condition is met continuously for a preset time, and the estimated refrigerant leakage rate is greater than 0.

[0056] Furthermore, the target leakage level is obtained based on the estimated value of the refrigerant leakage rate, wherein the larger the estimated value of the refrigerant leakage rate, the higher the leakage level, and the control strategies for different leakage levels are different; Control is implemented based on the target control strategy corresponding to the target leakage level.

[0057] For example, the control based on the target control strategy corresponding to the target leakage level includes: When the target leakage level is a minor leakage, the target control strategy is to limit the compressor frequency operation to a first operating frequency, reduce the refrigerant circulation speed, and send a first prompt, wherein the first operating frequency = rated operating frequency * first preset coefficient, and the first preset coefficient is a positive number less than 1; When the target leakage level is moderate leakage, the target control strategy is to limit the compressor frequency operation to the upper limit of the second operating frequency, reduce the opening of the electronic expansion valve, and send a second prompt. The second operating frequency is equal to the rated operating frequency multiplied by the second preset coefficient, which is a positive number less than 1 and is less than the first preset coefficient. When the target leakage level is severe leakage, the target control strategy is to control the compressor to stop, close the electronic expansion valve, and send a third prompt.

[0058] The air conditioning refrigerant leakage detection method provided in this application constructs a state-space model containing state equations and observation equations, jointly estimating refrigerant mass and leakage rate as state variables. Based on real-time operating parameters and measurable physical quantities, it obtains the optimal estimates of both and their covariance matrix to quantify the uncertainty information of the estimates, overcoming the deficiency of traditional methods in quantifying the confidence level of estimates. An adaptive dynamic threshold is constructed based on the refrigerant mass estimate and its covariance matrix, automatically adjusting the threshold according to operating condition fluctuations. Combined with the judgment condition that the refrigerant loss is greater than the dynamic threshold and the leakage rate estimate is greater than zero, it effectively distinguishes between actual leakage and operating condition disturbances, improving the sensitivity and robustness of leakage detection. Simultaneously, based on the leakage rate estimate, it classifies target leakage levels and executes differentiated control strategies, achieving a complete closed loop from accurate perception to graded response, significantly improving the safety and reliability of the air conditioning system under all operating conditions.

[0059] It should be noted that any process or method description in the flowchart or otherwise described herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which the embodiments of this application pertain.

[0060] Furthermore, in the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] To more clearly illustrate the proposed solution, a specific example is provided below: A quantitative estimation and control method for refrigerant leakage in air conditioning systems based on extended Kalman filtering. This method is applicable to various air conditioning systems, including residential and commercial systems, and is compatible with commonly used refrigerants such as R410A and R32. Its core implementation process is as follows: Figure 4 As shown, this includes: establishing a state-space model of refrigerant quality, designing an EKF estimation algorithm, designing a leakage criterion function, and assessing leakage levels and implementing graded control responses. Specific implementation details are as follows: I. Establishment of the State-Space Model of Refrigerant Quality Define the measurable observation vector as (It should be noted that these five parameters can be increased or decreased, but this directly affects state observability and estimation accuracy. Adding other parameters essentially increases hardware cost to increase the dimension of the observation vector, providing more fusion information for the EKF, which usually improves estimation accuracy and robustness. However, reducing too many parameters may cause the quality of the EKF and the leakage rate estimation to diverge or become irreversibly coupled, ultimately leading to a continuous widening of the dynamic threshold and loss of micro-leakage detection capability.) and These are the intake pressure and the exhaust pressure, respectively. and These are the intake temperature and the exhaust temperature, respectively. This refers to the compressor current. Furthermore, the control input vector is defined as... (These three parameters are a typical minimal complete set. Adding control input essentially provides more information on operating conditions, enabling the network to more accurately distinguish the observed characteristics under different operating modes, thereby improving the estimation accuracy of the EKF. However, it is not recommended to delete any of them; otherwise, the observation equation h(m, u) will not be uniquely mapped to the measured value, leading to EKF performance degradation or even divergence.) For compressor operating frequency, For the opening degree of the electronic expansion valve, This refers to the outdoor ambient temperature.

[0062] To simultaneously estimate refrigerant mass and leakage rate, a system state vector is defined. :

[0063] in, This refers to the total mass of refrigerant in the system (unit: g). Refrigerant leakage rate (unit: g / min). This indicates that the system is leaking refrigerant. Compared to traditional methods that only detect the refrigerant leakage rate, its advantages are: it elevates leak diagnosis from a relative change perception to an absolute state measurement; the refrigerant quality estimation covariance provides a quantitative reliability measure for leak criteria, enabling the dynamic threshold to adaptively adjust its sensitivity according to the stability of the operating conditions; and based on the current refrigerant quality and leakage rate, the time required to reach each leak level can be estimated.

[0064] The discrete state equation at time k (sampling period Δt) is:

[0065]

[0066] Write it in matrix form:

[0067] Wherein, the state transition matrix is , For the system Time-based process noise, including and The system The process noise of total refrigerant mass and the process noise of refrigerant leakage rate at any given time, and (Indicates that the process noise vector follows a normal distribution), process noise covariance matrix The settings are based on the actual refrigerant drift characteristics, reflecting minute random fluctuations in quality and gradual changes in leakage rate.

[0068] Observation equations Used to describe the nonlinear relationship between refrigerant quality and operating conditions and measurable physical quantities:

[0069] in, To measure noise, and , The measurement noise covariance matrix of the system is determined based on the sensor accuracy.

[0070] In air conditioning systems This is a nonlinear observation function used to describe the relationship between refrigerant mass and measurable physical quantities. It lacks a simple analytical expression and requires offline experiments to build a training dataset and then fitting it using a neural network. Mapping, such as Figure 2 As shown. The specific steps are as follows: Construct a three-layer Multilayer Perceptron (MLP) with the network structure of input layer (four-dimensional) -> hidden layer (three-dimensional) -> output layer (five-dimensional). The training process consists of the following four steps: (1) Data Acquisition and Preprocessing: Steady-state data such as suction / discharge pressure, suction / discharge temperature, and compressor current were collected under different refrigerant charge levels (60%-120% of rated value) and operating conditions (compressor frequency, electronic expansion valve opening, and ambient temperature combination), resulting in a total of 1600 sets of samples. All data were normalized (Z-score) and divided into training set, validation set, and test set at 70%, 15%, and 15%, respectively.

[0071] (2) Forward propagation to calculate the predicted value: input vector The algorithm is computed layer by layer through three hidden layers and activated by the ReLU function. The output layer uses linear activation, ultimately outputting a five-dimensional predicted observation vector. .

[0072] (3) Backpropagation update parameters: The predicted values ​​are calculated using mean squared error (MSE) and L2 regularization term. Compared with the true value The loss. The weights of each layer are adjusted based on the loss. and bias ( The influence of (1, 2, 3, 4) is considered, and the error signal is propagated backward layer by layer from the output layer to the input layer, with parameter correction performed according to a preset step size. The core steps and formulas are as follows: a. Calculate the output layer error signal The output layer (layer 4) uses linear activation, and its error signal... Directly reflects the deviation between the predicted value and the actual value: .

[0073] b. Backtracking hidden layer error signals layer by layer For hidden layers ( = 3, 2, 1), the error signal of the next layer is inversely distributed according to the transpose of the weight matrix, and gated modulation is performed by combining the switching characteristics of the ReLU activation function (1 when the input is greater than 0, 0 otherwise):

[0074] in, This indicates that the elements at each corresponding position of two vectors or matrices are multiplied. For the first The layer's activation output. This is a Boolean conditional function; it evaluates to true when the condition inside the parentheses is true. = 1, otherwise 0.

[0075] c. Calculate parameter adjustments and update. Based on the error signal of this layer and the output of the previous layer ( For the first layer, this is the input vector. The outer product of the two terms is used to calculate the adjustment direction of the weights, and an L2 regularization term is added to limit the weight magnitude. The adjustment direction of the bias is directly equal to the error signal of this layer. The parameter update formula is as follows:

[0076]

[0077] in, For learning rate, This is the regularization coefficient. During training, if the MSE continues to rise and remains high, it may be considered to adjust the regularization coefficient. Increase the regularization strength from 0.001 to 0.01; if the MSE of the training set decreases while the MSE of the validation set begins to increase, then increasing the regularization strength can be considered. Or reduce the network size.

[0078] (4) Iteration and early stopping: Repeat forward and backward propagation, and evaluate on the validation set after each round. If the validation loss no longer decreases and continues for 50 rounds, trigger the early stopping mechanism, save the model parameters with the best performance on the validation set, and complete the training.

[0079] II. Design of EKF Estimation Algorithm Based on the aforementioned state-space model, EKF roughly estimates the initial mass after the system is initially powered on and running stably, using current measurements combined with the inverse MLP model. ,set up (Assuming no initial leakage, set the leakage flag Leak-flag = 0); Initial covariance matrix Take the larger diagonal value to represent the initial uncertainty.

[0080] EKF is subsequently executed in each sampling period. Figure 3 The recursive steps shown are detailed below: (1) Prediction step System status Prior state estimation vector at time step (estimated before update) and prior state estimation covariance matrix for:

[0081]

[0082] in, for The posterior (updated estimate) state estimate vector at time step [time]. for The covariance matrix of the posterior state estimate at time t.

[0083] (2) Linearization Calculate the current state ( The observed Jacobian matrix at the prediction point (time) :

[0084] in, for The prior estimate of refrigerant quality at time t.

[0085] (3) Update step The current time is obtained by the following steps. The optimal state estimate (i.e., the posterior state estimate). ( and They are respectively Estimates of the total refrigerant mass and refrigerant leakage rate at time t(s) and their posterior state estimation covariance matrix. ,in The elements contained therein and These correspond to the covariance of the posterior state estimation for refrigerant quality and leakage rate, respectively. for Kalman gain at time step for The control input vector at time t, It is an identity matrix.

[0086]

[0087]

[0088]

[0089] Posterior state estimation covariance matrix The uncertainty of the estimate is quantified, and one sign of its convergence is that the diagonal elements of the covariance matrix no longer decrease continuously, but instead revolve around a steady-state value. Slight fluctuations. This invention utilizes To determine whether EKF converges, the following changes are used:

[0090] in, To observe the window length, The covariance is considered to be stable when the change is less than the threshold and continues for a period of time, and the EKF reaches a convergent state.

[0091] III. Design of Leakage Criterion Function After EKF convergence, record the total reference mass of the refrigerant in the system. The covariance of the estimated mean over a past period (e.g., 5 minutes) is denoted as . Online construction of leakage detection parameters :

[0092] Calculate its variance approximation :

[0093] Dynamic threshold Designed as follows:

[0094] in, To minimize the detectable refrigerant mass loss, The confidence coefficient C is determined according to the 3σ criterion in this embodiment, with a recommended value of C = 3. This provides a 99.73% confidence level, widely considered a universal baseline for distinguishing between normal random fluctuations and real physical changes in mature fields such as industrial automation. If C is too small, the dynamic threshold will be very narrow, making the system highly sensitive to any small fluctuations. This can lead to frequent false alarms under normal transient conditions such as frequency conversion, defrosting, and outdoor gusts. If C is too large, the dynamic threshold will be abnormally lenient. While this almost eliminates false alarms, it can also make the system sluggish and cause missed alarms.

[0095] when And this condition is met continuously for the confirmation time window. (For example, 2 minutes can be taken) When a refrigerant leak is detected, the leak flag Leak-flag is set to 1, and the procedure is as follows: Figure 4 As shown. Additionally, auxiliary conditions need to be added to require leakage rate estimation. This is to eliminate the possibility of false degradation in refrigerant quality caused by sensor drift.

[0096] This dynamic threshold design mechanism allows the threshold to automatically increase when operating conditions change abruptly or the estimated covariance increases, reducing the probability of false alarms; and when operating smoothly and the covariance decreases, the threshold narrows, enabling sensitive detection of micro-leakage.

[0097] IV. Leakage Level Assessment and Graded Control Response Leakage rate estimation using EKF output Leakage levels are classified according to Table 1.

[0098] Table 1

[0099] Among them, the leakage rate threshold , The system is calibrated based on its total charge and allowable time loss. When the Leak-flag is 0, it indicates that there is currently no refrigerant leak in the air conditioning system, and the system continues to operate normally with current parameters, under continuous monitoring. When the Leak-flag is 1, based on the current refrigerant leak level, it can be determined according to... Figure 5 The illustrated hierarchical strategy controls the air conditioner: (1) Minor leak Status: At this time, the refrigerant leakage is small and has little impact on system performance.

[0100] Measures: Limit compressor frequency operation to approximately 60% to 80% of rated capacity to reduce refrigerant circulation speed; adaptively adjust the opening of the electronic expansion valve to optimize refrigerant supply based on suction superheat and discharge temperature, maintaining stable system pressure; the human-machine interface displays "The system currently has a slight leak and is operating at reduced power." The system is continuously monitored, and the leak level is updated periodically.

[0101] (2) Moderate leakage Status: At this point, the refrigerant leakage is significant, the system's cooling / heating performance is noticeably reduced, and energy consumption increases. Measures: Limit compressor operating frequency to no more than 50% of rated capacity to further slow refrigerant leakage and prevent compressor overheating due to insufficient refrigerant; reduce electronic expansion valve opening by approximately 40% to prevent excessive refrigerant supply and liquid slugging when refrigerant is insufficient; the HMI displays the message "The system currently has a moderate leak and is operating at significantly reduced power. Please check the refrigerant leak." The system is continuously monitored, and the leak level is updated periodically.

[0102] (3) Severe leakage Status: At this time, the refrigerant leakage is large. If it is not dealt with in time, it will cause the compressor to burn dry, overheat and be damaged, and the system will not be able to operate normally.

[0103] Measures: Immediately stop the compressor; close the electronic expansion valve to full shut to prevent abnormal flow of residual refrigerant; turn off the indoor and outdoor fans; the human-machine interface will display "The system is currently experiencing a severe leak and has been shut down. Please contact maintenance personnel immediately," and will issue an audible and visual alarm, locking the air conditioner's operating functions. The air conditioner will need to be unlocked after maintenance personnel resolve the issue.

[0104] Through the above four steps, this system forms a complete closed loop of "perception-estimation-decision-control": Sensing layer: Real-time acquisition of compressor operating frequency, compressor operating current, electronic expansion valve opening, outdoor ambient temperature, suction pressure, discharge pressure, suction temperature, and discharge temperature.

[0105] Estimation layer: EKF recursion gives , and its covariance .

[0106] Decision-making level: Dynamic threshold leakage criteria and level assessment output Leak-flag and leakage level.

[0107] Control layer: The air conditioning controller receives the decision results, automatically switches the control strategy, and continuously monitors the system operating parameters.

[0108] This application embodiment only requires utilizing existing sensor data from the air conditioning system (suction / discharge pressure, evaporation / condensation temperature, compressor current, electronic expansion valve opening, etc.), eliminating the need for additional hardware such as refrigerant mass flow meters and electronic scales, significantly reducing system cost and installation complexity. The leakage criterion is based on a dynamic threshold constructed from the refrigerant mass estimate and its covariance. The threshold adaptively adjusts with estimation uncertainty, automatically widening under high noise or transient conditions to significantly suppress false alarms; it maintains high sensitivity under steady-state low noise conditions, capable of detecting micro-leakages of a few grams per day. By estimating the leakage rate, it classifies leakage into mild, moderate, and severe levels, triggering differentiated control responses such as frequency limiting, load reduction, and forced shutdown, avoiding operational interruptions caused by "one-size-fits-all" protection and preventing compressor damage. The results are fed back to the main control MCU in real time, forming a "sensing-estimation-decision-control" closed loop, upgrading the air conditioning system from passive threshold protection to active fault-tolerant control.

[0109] Based on the same inventive concept, such as Figure 9 As shown in the figure, this application embodiment provides an air conditioning refrigerant leakage detection device 90, including: The spatial model construction module 91 is used to construct a state space model, which includes a state equation describing the evolution of refrigerant quality and refrigerant leakage rate over time, and an observation equation describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The construction of the state-space model includes: Constructing discrete state equations:

[0110]

[0111] Where m(k) is the refrigerant mass at time k, m(k-1) is the refrigerant mass at time k-1, r(k) is the refrigerant leakage rate at time k, r(k-1) is the refrigerant leakage rate at time k-1, and Δt is the sampling period. The refrigerant mass process noise at time k-1, The refrigerant leakage rate at time k-1 represents the process noise. Based on the measurable physical quantities and operating parameters, a nonlinear observation equation is constructed:

[0112] in, Let z be the observation vector at time k. P s and P d These are the inhalation pressure and the exhaust pressure, T. s and T d These are the intake temperature and the exhaust temperature, respectively. Let be the compressor current, h be the nonlinear observation function, and u(k) be the input vector u at time k. , For compressor operating frequency, For the opening degree of the electronic expansion valve, Outdoor ambient temperature To measure noise, and , Let be the measurement noise covariance matrix of the system.

[0113] In a preferred implementation of this application, the nonlinear observation function is obtained based on a pre-trained model, such as... Figure 2 As shown, the training steps of the model are as follows: Sample data were obtained by collecting measurable parameters under different refrigerant qualities and operating conditions. The operating conditions included compressor operating frequency, electronic expansion valve opening, and outdoor ambient temperature. The measurable parameters included suction pressure, discharge pressure, suction temperature, discharge temperature, and compressor current.

[0114] The input values ​​for each sample data are obtained, which consist of refrigerant quality, compressor operating frequency, electronic expansion valve opening degree, and outdoor ambient temperature. The input values ​​are input into the initial model to obtain the predicted values. The initial model includes an input layer, a hidden layer, and an output layer. The predicted values ​​are composed of intake pressure, exhaust pressure, intake temperature, exhaust temperature, and compressor current. The loss between the predicted and true values ​​is calculated using mean squared error and L2 regularization; where the true value is obtained from the actual value in the sample data.

[0115] The parameters of the initial model are updated based on the loss until the condition is met.

[0116] Further, updating the parameters of the initial model based on the loss until the condition is met includes: Weights of each layer based on loss and bias The impact of the error signal is transmitted backward from the output layer to the input layer, and the parameters are corrected according to a preset step size. For the output layer, linear activation is used, and its error signal... Directly reflects the deviation between the predicted value and the actual value: ; For the hidden layer, the error signal of the next layer is inversely distributed according to the transpose of the weight matrix, and gated modulation is performed by combining the switching characteristics of the ReLU activation function:

[0117] in, This indicates that the elements at each corresponding position of two vectors or matrices are multiplied. For the first Layer activation output; This is a Boolean conditional function; it evaluates to true when the condition inside the parentheses is true. = 1, otherwise 0; For the input layer, the error signal of this layer is compared with the output of the previous layer. The outer product of and calculates the direction of weight adjustment, and an L2 regularization term is added to limit the weight magnitude. Here, for the first layer, is the input vector. The direction of bias adjustment is directly equal to the error signal of this layer, and the parameter update formula is as follows:

[0118]

[0119] in, For learning rate, The regularization coefficient is used. Repeat the forward and backward propagation, and evaluate on the validation set after each round; if the validation loss no longer decreases and continues for the preset number of rounds, trigger the early stopping mechanism, save the optimal model parameters on the validation set, and complete the training.

[0120] The output vector estimation module 92 is used to input the real-time operating parameters and the measurable physical quantity into the state space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. As a preferred implementation of the embodiments of this application, such as Figure 3 As shown, the step of inputting real-time operating parameters and the measurable physical quantities into the state-space model to obtain estimates of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix, includes: During stable operation, the initial value of refrigerant mass is obtained, the initial value of leakage rate is set to zero, and the initial covariance matrix is ​​set to a large diagonal matrix; Using the state estimate and covariance matrix from the previous time step, a one-step prediction is performed based on the state equation to obtain the prior state estimate and its prior covariance matrix for the current time step; for example, the system state... Prior state estimation vector at time step (estimated before update) and prior state estimation covariance matrix for:

[0121]

[0122] in, for The posterior (updated estimate) state estimate vector at time step [time]. for The covariance matrix of the posterior state estimate at time t.

[0123] Linearize the nonlinear observation function at the current prior estimate point to obtain the observation Jacobian matrix; for example, calculate the current state ( The observed Jacobian matrix at the prediction point (time) :

[0124] in, for The prior estimate of refrigerant quality at time t.

[0125] The Kalman gain is calculated based on the prior covariance matrix and the observed Jacobian matrix; the residual between the real-time measurable physical quantity and the predicted observation value is calculated; the prior state estimate is corrected by weighting the residual with the Kalman gain to obtain the optimal estimate of the refrigerant mass and refrigerant leakage rate at the current moment; at the same time, the covariance matrix is ​​updated, and the estimated variances of the two and their correlation are output. Monitor the change in quality variance in the covariance matrix within the window. When it is continuously less than the threshold, the filter is determined to be converged. Record the current estimated value and the corresponding variance as the estimated value of the refrigerant quality and refrigerant leakage rate, as well as the covariance matrix of the two.

[0126] For example, the current time is obtained by the following steps. The optimal state estimate (i.e., the posterior state estimate). ( and They are respectively Estimates of the total refrigerant mass and refrigerant leakage rate at time t(s) and their posterior state estimation covariance matrix. ,in The elements contained therein and These correspond to the covariance of the posterior state estimation for refrigerant quality and leakage rate, respectively. for Kalman gain at time step for The control input vector at time t, It is an identity matrix.

[0127]

[0128]

[0129]

[0130] Posterior state estimation covariance matrix The uncertainty of the estimate is quantified, and one sign of its convergence is that the diagonal elements of the covariance matrix no longer decrease continuously, but instead revolve around a steady-state value. Slight fluctuations. This invention utilizes To determine whether EKF converges, the following changes are used:

[0131] in, To observe the window length, For a very small threshold, when the change is less than the threshold and persists for a period of time, the covariance is considered to have stabilized, and EKF has reached a convergent state. The dynamic threshold construction module 93 is used to construct a dynamic threshold based on the estimated value of the refrigerant quality and its covariance matrix. The construction of a dynamic threshold based on the estimated refrigerant quality and its covariance matrix includes: Obtain the mean value of refrigerant quality estimates over a period of time as a reference quality, and record the variance of quality estimates for the corresponding period as a reference covariance; At each sampling time, the difference between the reference quality and the current quality estimate is calculated to obtain the real-time quality loss. The reference covariance is added to the current quality estimate variance to approximate the total variance of the quality loss. The total standard deviation is obtained by taking the square root of the total variance, multiplying it by a preset confidence coefficient, and then adding a preset minimum detectable loss to form the adaptive threshold at the current moment.

[0132] The refrigerant leakage determination module 94 is used to determine a refrigerant leak when the refrigerant loss is greater than the dynamic threshold and the estimated refrigerant leakage rate is greater than 0.

[0133] The refrigerant loss is the same as the mass loss mentioned above.

[0134] To ensure accurate judgment, a refrigerant leak is determined when the refrigerant loss exceeds the dynamic threshold and this condition is met continuously for a preset time, and the estimated refrigerant leakage rate is greater than 0.

[0135] Furthermore, the target leakage level is obtained based on the estimated value of the refrigerant leakage rate, wherein the larger the estimated value of the refrigerant leakage rate, the higher the leakage level, and the control strategies for different leakage levels are different; Control is implemented based on the target control strategy corresponding to the target leakage level.

[0136] For example, the control based on the target control strategy corresponding to the target leakage level includes: When the target leakage level is a minor leakage, the target control strategy is to limit the compressor frequency operation to a first operating frequency, reduce the refrigerant circulation speed, and send a first prompt, wherein the first operating frequency = rated operating frequency * first preset coefficient, and the first preset coefficient is a positive number less than 1; When the target leakage level is moderate leakage, the target control strategy is to limit the compressor frequency operation to the upper limit of the second operating frequency, reduce the opening of the electronic expansion valve, and send a second prompt. The second operating frequency is equal to the rated operating frequency multiplied by the second preset coefficient, which is a positive number less than 1 and is less than the first preset coefficient. When the target leakage level is severe leakage, the target control strategy is to control the compressor to stop, close the electronic expansion valve, and send a third prompt.

[0137] The air conditioning refrigerant leakage detection device provided in this application constructs a state-space model containing state equations and observation equations, jointly estimating refrigerant mass and leakage rate as state variables. Based on real-time operating parameters and measurable physical quantities, it obtains the optimal estimates of both and their covariance matrix to quantify the uncertainty information of the estimates, overcoming the deficiency of traditional methods in quantifying the confidence level of estimates. An adaptive dynamic threshold is constructed based on the refrigerant mass estimate and its covariance matrix, automatically adjusting the threshold according to operating condition fluctuations. Combined with the judgment condition that the refrigerant loss is greater than the dynamic threshold and the leakage rate estimate is greater than zero, it effectively distinguishes between actual leakage and operating condition disturbances, improving the sensitivity and robustness of leakage detection. Simultaneously, based on the leakage rate estimate, it classifies target leakage levels and executes differentiated control strategies, achieving a complete closed loop from accurate perception to graded response, significantly improving the safety and reliability of the air conditioning system under all operating conditions.

[0138] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the air conditioning refrigerant leakage detection method provided in any of the above embodiments.

[0139] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0140] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0141] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0142] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0143] The computer-readable storage medium provided in this application embodiment stores a computer program. When executed by a processor, the computer program implements the steps of the air conditioning refrigerant leakage detection method provided in any of the above embodiments. This constructs a state-space model of refrigerant quality and leakage rate. Based on real-time operating parameters and measurable physical quantities, it obtains the optimal estimates of refrigerant quality and leakage rate, along with their covariance matrix, to quantify uncertain information. An adaptive dynamic threshold is constructed based on the covariance matrix. Leakage is confirmed when the refrigerant loss exceeds the dynamic threshold and the estimated leakage rate is greater than zero. A differentiated control strategy is then executed in conjunction with the leakage level, achieving a complete closed loop of leakage detection and graded control. This effectively improves the accuracy of air conditioning refrigerant leakage detection and the safety of air conditioning operation.

[0144] Based on the same inventive concept, such as Figure 10 As shown, this application also provides an air conditioning refrigerant leak detection system 100, comprising: At least one processor 101 and at least one memory 102; The memory stores the executable instructions of the processor; The processor is configured to perform the air conditioning refrigerant leakage detection method provided in the above embodiments.

[0145] The air conditioning refrigerant leakage detection system provided in this application stores the processor's executable instructions in a memory. When the executable instructions are executed, the processor can construct a state-space model of refrigerant quality and refrigerant leakage rate. Based on real-time operating parameters and measurable physical quantities, it obtains the optimal estimates of refrigerant quality and leakage rate and their covariance matrix to quantify uncertain information. An adaptive dynamic threshold is constructed based on the covariance matrix. When the refrigerant loss is greater than the dynamic threshold and the estimated leakage rate is greater than zero, leakage is confirmed. A differentiated control strategy is executed in conjunction with the leakage level, realizing a complete closed loop of leakage detection and graded control, effectively improving the accuracy of air conditioning refrigerant leakage detection and the safety of air conditioning operation.

[0146] Based on the same inventive concept, this application provides an air conditioner including the refrigerant leak detection system provided in the above embodiments. It can estimate the refrigerant quality and leakage rate and their covariance matrix online in real time, and construct an adaptive dynamic threshold based on the covariance, effectively distinguishing between actual leaks and operational disturbances, thus improving the accuracy and robustness of leak detection. Simultaneously, it executes differentiated control strategies according to the leakage level, realizing closed-loop management from sensing to control, significantly enhancing the safety, reliability, and intelligent maintenance capabilities of the air conditioner under all operating conditions.

[0147] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0148] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0149] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for detecting refrigerant leakage in air conditioning systems, characterized in that, include: A state-space model is constructed, which includes state equations describing the evolution of refrigerant quality and refrigerant leakage rate over time, and observation equations describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The real-time operating parameters and the measurable physical quantities are input into the state-space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. A dynamic threshold is constructed based on the estimated refrigerant quality and its covariance matrix; A refrigerant leak is determined when the refrigerant loss exceeds the dynamic threshold and the estimated refrigerant leakage rate is greater than 0.

2. The method according to claim 1, characterized in that: The construction of the state-space model includes: Constructing discrete state equations: Where m(k) is the refrigerant mass at time k, m(k-1) is the refrigerant mass at time k-1, r(k) is the refrigerant leakage rate at time k, r(k-1) is the refrigerant leakage rate at time k-1, and Δt is the sampling period. The refrigerant mass process noise at time k-1, The refrigerant leakage rate at time k-1 represents the process noise. Based on the measurable physical quantities and operating parameters, a nonlinear observation equation is constructed: in, Let z be the observation vector at time k. P s and P d These are the inhalation pressure and the exhaust pressure, T. s and T d These are the intake temperature and the exhaust temperature, respectively. Let be the compressor current, h be the nonlinear observation function, and u(k) be the input vector u at time k. , For compressor operating frequency, For the opening degree of the electronic expansion valve, Outdoor ambient temperature To measure noise, and , Let be the measurement noise covariance matrix of the system.

3. The method according to claim 2, characterized in that: The nonlinear observation function is obtained based on a pre-trained model, and the training steps of the model are as follows: Sample data were obtained by collecting measurable parameters under different refrigerant qualities and operating conditions. The input values ​​for each sample data are obtained, which consist of refrigerant quality, compressor operating frequency, electronic expansion valve opening degree, and outdoor ambient temperature. The input values ​​are input into the initial model to obtain the predicted values. The initial model includes an input layer, a hidden layer, and an output layer. The predicted values ​​are composed of intake pressure, exhaust pressure, intake temperature, exhaust temperature, and compressor current. The loss between the predicted and actual values ​​is calculated using mean squared error and L2 regularization. The parameters of the initial model are updated based on the loss until the condition is met.

4. The method according to claim 3, characterized in that: The step of updating the parameters of the initial model based on the loss until the condition is met includes: Weights of each layer based on loss and bias The impact of the error signal is transmitted backward from the output layer to the input layer, and the parameters are corrected according to a preset step size. For the output layer, linear activation is used, and its error signal... Directly reflects the deviation between the predicted value and the actual value: ; For the hidden layer, the error signal of the next layer is inversely distributed according to the transpose of the weight matrix, and gated modulation is performed by combining the switching characteristics of the ReLU activation function: in, This indicates that the elements at each corresponding position of two vectors or matrices are multiplied. For the first Layer activation output; This is a Boolean conditional function; it evaluates to true when the condition inside the parentheses is true. = 1, otherwise 0; For the input layer, the error signal of this layer is compared with the output of the previous layer. The outer product of and calculates the direction of weight adjustment, and an L2 regularization term is added to limit the weight magnitude. Here, for the first layer, is the input vector. The direction of bias adjustment is directly equal to the error signal of this layer, and the parameter update formula is as follows: in, For learning rate, The regularization coefficient is . Repeat the forward and backward propagation, and evaluate on the validation set after each round; if the validation loss no longer decreases and continues for the preset number of rounds, trigger the early stopping mechanism, save the optimal model parameters on the validation set, and complete the training.

5. The method according to claim 1, characterized in that: The step of inputting real-time operating parameters and the measurable physical quantities into the state-space model to obtain estimates of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix, includes: During stable operation, the initial value of refrigerant mass is obtained, the initial value of leakage rate is set to zero, and the initial covariance matrix is ​​set to a large diagonal matrix; Using the state estimate and covariance matrix from the previous time step, a one-step prediction is performed based on the state equation to obtain the prior state estimate and its prior covariance matrix at the current time step. The nonlinear observation function is linearized at the current prior estimation point to obtain the observation Jacobian matrix. The Kalman gain is calculated based on the prior covariance matrix and the observed Jacobian matrix; the residual between the real-time measurable physical quantity and the predicted observation value is calculated; the prior state estimate is corrected by weighting the residual with the Kalman gain to obtain the optimal estimate of the refrigerant mass and refrigerant leakage rate at the current moment; at the same time, the covariance matrix is ​​updated, and the estimated variances of the two and their correlation are output. Monitor the change in quality variance in the covariance matrix within the window. When it is continuously less than the threshold, the filter is determined to be converged. Record the current estimated value and the corresponding variance as the estimated value of the refrigerant quality and refrigerant leakage rate, as well as the covariance matrix of the two.

6. The method according to claim 1, characterized in that: The construction of a dynamic threshold based on the estimated refrigerant quality and its covariance matrix includes: Obtain the mean value of refrigerant quality estimates over a period of time as a reference quality, and record the variance of quality estimates for the corresponding period as a reference covariance; At each sampling time, the difference between the reference quality and the current quality estimate is calculated to obtain the real-time quality loss. The reference covariance is added to the current quality estimate variance to approximate the total variance of the quality loss. The total standard deviation is obtained by taking the square root of the total variance, multiplying it by a preset confidence coefficient, and then adding a preset minimum detectable loss to form the adaptive threshold at the current moment.

7. The method according to claim 1, characterized in that, Also includes: The target leakage level is obtained based on the estimated value of the refrigerant leakage rate. The higher the estimated value of the refrigerant leakage rate, the higher the leakage level. Different control strategies are used for different leakage levels. Control is implemented based on the target control strategy corresponding to the target leakage level.

8. The method according to claim 7, characterized in that: The control based on the target control strategy corresponding to the target leakage level includes: When the target leakage level is a minor leakage, the target control strategy is to limit the compressor frequency operation to a first operating frequency, reduce the refrigerant circulation speed, and send a first prompt, wherein the first operating frequency = rated operating frequency * first preset coefficient, and the first preset coefficient is a positive number less than 1; When the target leakage level is moderate leakage, the target control strategy is to limit the compressor frequency operation to the upper limit of the second operating frequency, reduce the opening of the electronic expansion valve, and send a second prompt. The second operating frequency is equal to the rated operating frequency multiplied by the second preset coefficient, which is a positive number less than 1 and is less than the first preset coefficient. When the target leakage level is severe leakage, the target control strategy is to control the compressor to stop, close the electronic expansion valve, and send a third prompt.

9. An air conditioning refrigerant leakage detection device, characterized in that, include: The spatial model construction module is used to construct a state space model, which includes state equations describing the evolution of refrigerant quality and refrigerant leakage rate over time, and observation equations describing the mapping relationship between refrigerant quality and operating parameters and measurable physical quantities. The output vector estimation module is used to input real-time operating parameters and the measurable physical quantities into the state space model to obtain the estimated values ​​of the refrigerant mass and refrigerant leakage rate, as well as their covariance matrix. The covariance matrix is ​​used to represent the uncertainty information of the estimated values. A dynamic threshold construction module is used to construct a dynamic threshold based on the estimated value of the refrigerant quality and its covariance matrix. The refrigerant leakage determination module is used to determine a refrigerant leak when the refrigerant loss exceeds the dynamic threshold and the estimated refrigerant leakage rate is greater than 0.

10. An air conditioning refrigerant leak detection system, characterized in that, include: At least one processor and at least one memory; The memory stores the executable instructions of the processor; The processor is configured to perform the method according to any one of claims 1-8.

11. An air conditioner, characterized in that, Includes the air conditioning refrigerant leakage detection system as described in claim 10.

Images