A heat pump system fault prediction method based on time series chain diagram

Abstract

The present invention relates to the technical field of heat pump fault detection and diagnosis, and discloses a heat pump system fault prediction method based on a time series chain graph, comprising: configuring and deploying a sensor group, collecting monitoring variable data of the heat pump system from normal operation to the time when a fault occurs, recording the heat pump system fault type at the same time, and constructing a heat pump system degradation time series data set; constructing a time series chain graph degradation model of the heat pump system from normal operation to the time when a fault occurs; constructing a hierarchical graph model of monitoring variables, remaining life, and fault type; constructing a competing risk time series chain graph degradation model, and calculating a competing risk cumulative occurrence function; obtaining time series data monitored during the operation of the heat pump system, and predicting the remaining life and fault type of the heat pump system based on the competing risk cumulative occurrence function; the method improves the accuracy and efficiency of heat pump fault detection and diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of heat pump fault detection and diagnosis, and in particular to a heat pump system fault prediction method based on a time series chain diagram. Background Art

[0002] Heat pumps provide a low-carbon energy solution for district heating and industrial production. Failure to promptly identify heat pump failures during operation can lead to system downtime, disrupting the continuity of heating or industrial processes, increasing energy consumption and uneven heating, and increasing maintenance and repair costs. These failures can negatively impact the availability, performance, and operating costs of heat pump systems. Research indicates that heat pumps can exhibit 48 types of failures, encompassing over 120 different types. The causes of these failures vary and are often correlated. For example, a defective shaft seal can be a cause of refrigerant leakage, but this type of failure can also be caused by excessive vibration due to a screw compressor slide valve failure. Given that failures can interconnect and propagate between components, predicting the type and timing of failure is a complex task.

[0003] Currently, commonly used heat pump fault detection and diagnosis methods include causal analysis based on prior knowledge, expert systems, and physical models, as well as data-driven statistical analysis, machine learning, and fuzzy logic. The coefficient of performance of a heat pump system is highly dependent on its operating conditions, which encounter a variety of operating states and uncertain environmental conditions during actual operation. Heat pump fault detection and diagnosis methods based on prior knowledge offer good predictive accuracy, but they require a comprehensive understanding of the heat pump system's equipment, components, processes, and operating conditions. This leads to long modeling times and high costs, limiting their practical application. In contrast, data-driven heat pump fault detection and diagnosis methods are more efficient. However, existing methods typically treat the heat pump system as a "black box," making it difficult to define the probability distribution functions for heat pump fault types and occurrence times required in practical applications. Furthermore, they fail to characterize the potential changes in the degradation mechanisms and patterns of the heat pump throughout its lifecycle, resulting in low detection accuracy. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a heat pump system fault prediction method based on a time series chain diagram to solve the problems of low efficiency, high cost and low accuracy in the prior art of heat pump system fault prediction.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A heat pump system fault prediction method based on a time series chain diagram comprises the following steps:

[0007] S1. Configure and deploy a sensor group to collect monitoring variable data of the heat pump system from normal operation to the moment of failure, record the failure type of the heat pump system, and construct a time series dataset of heat pump system degradation;

[0008] Among them, the monitoring variable data includes operating status data and operating environment data;

[0009] S2. Based on the heat pump system degradation time series dataset, a time series chain graph degradation model of the heat pump system from normal operation to the moment of failure is constructed;

[0010] S3. Based on the time series chain graph degradation model, a hierarchical graph model of monitoring variables, remaining life, and fault types is constructed;

[0011] S4. Based on the hierarchical graph model and the time series chain graph degradation model, a competing risk time series chain graph degradation model is constructed, and the competing risk cumulative occurrence function is calculated;

[0012] S5. Obtain the time series data monitored during the operation of the heat pump system, and predict the remaining life and fault type of the heat pump system based on the competing risk cumulative occurrence function.

[0013] The present invention has the following beneficial effects:

[0014] 1. This invention proposes a heat pump system fault prediction method based on a time series chain graph. Based on the time series chain graph, a competing risk time series chain graph degradation model is constructed. This method has flexible structural properties and can accurately characterize dynamics. This method improves the accuracy and precision of heat pump fault detection and diagnosis, while reducing the instability of prediction capabilities caused by the complexity and diversity of heat pump system fault types, the interconnectedness and propagation of faults, and the high correlation between faults and dynamically changing operating conditions.

[0015] 2. Modeling based on time series chain graphs reduces modeling costs and improves modeling efficiency. At the same time, the constructed competing risk time series chain graph degradation model is used to present the degradation mechanism and regular changes of the heat pump system throughout its life cycle, which can effectively predict the remaining life and fault type of the heat pump system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a heat pump system fault prediction method based on a time series chain diagram proposed by the present invention;

[0017] Figure 2 It is a schematic diagram of the hierarchical graph model;

[0018] Figure 3 Schematic diagram of the competing risk time series chain graph degradation model. DETAILED DESCRIPTION

[0019] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0020] like Figure 1 As shown, a heat pump system fault prediction method based on a time series chain diagram includes the following steps S1-S5:

[0021] S1. Configure and deploy a sensor group to collect monitoring variable data of the heat pump system from normal operation to the moment of failure. Simultaneously, record the failure type of the heat pump system and construct a heat pump system degradation time series dataset. Monitoring variable data includes operating status data and operating environment data.

[0022] In this embodiment, based on the user's monitoring needs for the heat pump system, with the goal of maximizing observability, detectability, and separability, and taking into account sensor failure and cost, temperature, pressure, flow, vibration, and other sensors are configured and deployed on the main equipment, components, and parts of the heat pump system to collect the operating status data of the heat pump system (such as the inlet and outlet water temperature of the heat pump, the inlet and outlet water temperature of the buried pipe, and the system pressure). Temperature and humidity sensors are configured and deployed in the operating environment of the heat pump system to collect the operating environment data of the heat pump system (such as soil temperature, soil humidity, indoor and outdoor temperature, etc.). The operating environment data and operating status data of the heat pump system are sampled through sensors to monitor the process from normal operation to failure of the heat pump, record the failure type, and thus generate a heat pump system degradation time series dataset, so that the time series chain diagram can be subsequently introduced to predict the failure type and remaining life of the heat pump system.

[0023] Specifically, step S1 includes S11-S12:

[0024] S11. Configure and deploy a first sensor group on the main equipment, components, and parts of the heat pump system, and configure and deploy a second sensor group in the operating environment of the heat pump system; the first sensor group is used to monitor operating status data composed of several monitoring variables including temperature, pressure, flow, and vibration; the second sensor group is used to monitor operating environment data composed of several monitoring variables including temperature and humidity.

[0025] S12. Set the sampling frequency and number of sampling times, use the first sensor group and the second sensor group to collect monitoring variable data of the heat pump system from normal operation to the time of failure, and record the failure type of the heat pump system to construct a heat pump system degradation time series data set, that is:

[0026]

[0027] in, represents the heat pump system degradation time series data set, N represents the total number of sampling times, i represents the number of sampling times, X i (τ) represents the monitoring variable data sampled at the i-th time, Z i represents the heat pump system fault type of the i-th sampling, τ represents the sampling time point, A i (τ) represents the operating environment data of the ith sampling at the sampling time point τ, B i (τ) represents the operating status data of the ith sampling at the sampling time point τ, k represents the heat pump system fault type, K represents the total number of heat pump system fault types, A i1 (τ) represents the data of the first operating environment monitoring variable at the sampling time point τ during the i-th sampling, A i2 (τ) represents the data of the second operating environment monitoring variable at the sampling time point τ during the i-th sampling, A ip (τ) represents the data of the pth operating environment monitoring variable at the τ sampling time point during the i-th sampling, B i1 (τ) represents the data of the first operating status monitoring variable at the sampling time point τ during the i-th sampling, B i2 (τ) represents the data of the second operating status monitoring variable at the sampling time point τ during the i-th sampling, B iq (τ) represents the data of the qth operating status monitoring variable at the τ sampling time point during the i-th sampling.

[0028] S2. Based on the heat pump system degradation time series dataset, a time series chain graph degradation model of the heat pump system from normal operation to the moment of failure is constructed.

[0029] In this embodiment, a probabilistic graphical model structure learning method is used to construct an undirected graphical model of each monitoring variable, that is, an undirected graph of variable correlation, based on the degradation time series data set. According to the process from normal operation to failure of the heat pump and the characteristics of the time series chain graph, the undirected graphical model is converted into a time series chain graph degradation model. That is, the historical operating environment data and operating status data of the heat pump system are used to determine the dynamic degradation process of the operating status of the heat pump system and the dynamic impact of the operating environment on the degradation process, so that in the subsequent steps, a competing risk time series chain graph degradation model that associates the time series chain graph degradation model with the heat pump failure type, remaining life and probability distribution parameters is established, thereby estimating the parameters of the competing risk time series chain graph degradation model.

[0030] Specifically, step S2 includes S21-S22:

[0031] S21. Based on the heat pump system degradation time series dataset, a probabilistic graphical model structure learning method is used to establish an undirected graphical model of each monitoring variable; wherein the undirected graphical model of each monitoring variable is composed of all variable nodes and edges connecting the variable nodes, and the variable nodes are nodes of the monitoring variables.

[0032] S22. Based on the undirected graph model of each monitoring variable, a time series chain graph degradation model of the heat pump system from normal operation to the moment of failure is constructed, which specifically includes S221-S223:

[0033] S221. Convert undirected edges connecting variable nodes at different sampling time points in the undirected graph model into directed edges pointing from an earlier sampling time point to a later sampling time point.

[0034] S222 . Convert the undirected edges connecting the variable nodes of different fault types at the same sampling time point in the undirected graph model into directed edges pointing from the operating environment data to the operating status data.

[0035] S223. The undirected edges connecting the variable nodes of the same fault type at the same sampling time point in the undirected graph model are retained unchanged, and a time series chain graph degradation model of the heat pump system from normal operation to the moment of fault occurrence is obtained, namely:

[0036]

[0037] in, Represents the time series chain graph degradation model, V TS The variable node set representing the time series chain graph degradation model, E TS represents the set of directed and undirected edges in the time series chain graph degradation model, X(τ) represents the monitoring variable data at the sampling time point τ, A(τ) represents the operating environment data at the sampling time point τ, B(τ) represents the operating status data at the sampling time point τ, A1(τ) represents the data of the first operating environment monitoring variable at the sampling time point τ, A2(τ) represents the data of the second operating environment monitoring variable at the sampling time point τ, A p (τ) represents the data of the pth operating environment monitoring variable at the sampling time point τ, B1(τ) represents the data of the first operating state monitoring variable at the sampling time point τ, B2(τ) represents the data of the second operating state monitoring variable at the sampling time point τ, and B q (τ) represents the data of the qth operating status monitoring variable at the τ sampling time point.

[0038] S3. Based on the time series chain graph degradation model, a hierarchical graph model of monitoring variables, remaining life and fault types is constructed.

[0039] In this embodiment, the established hierarchical graph model is as follows Figure 2 As shown, Figure 2 The diagram shows the correlation between monitoring variables and the remaining life and fault type of the heat pump system at each sampling time point in the hierarchical graph model. The construction process is as follows:

[0040] Specifically, step S3 includes S31-S33:

[0041] S31. Based on the time series chain graph degradation model, for each time sampling point τ=-T,…,-2,-1,0, add node π τ , β τ , σ τ ,λ τ and {RUL,Z}(τ);

[0042] Among them, π τ ={π 1,τ ,π 2,τ ,…,π K,τ}、β τ ={β 1,τ ,β 2,τ ,…,β K,τ}、σ τ ={σ 1,τ ,σ 2,τ ,…,σ K,τ} and λ τ ={λ 1,τ ,λ 2,τ ,…,λ K,τ},π τ β represents the parameter of the probability distribution of the heat pump system failure type at the sampling time point τ, τ , σ τ ,λ τ They represent the scale, shape and location parameters of the probability distribution of the remaining life of the heat pump system for a given fault type at the sampling time point τ, π 1,τ , π 2,τ , π K,τ They represent the probability of failure type 1, 2 and K if the heat pump system fails at the sampling time point τ, β 1,τ , β 2,τ , β K,τ They represent the scale parameters of the probability distribution of the remaining life of the heat pump system when the fault types are 1, 2 and K at the sampling time point τ, σ 1,τ , σ 2,τ , σ K,τ They represent the shape parameters of the probability distribution of the remaining life of the heat pump system when the fault types are 1, 2 and K at the sampling time point τ, respectively, and λ 1,τ ,λ 2,τ ,λ K,τThey represent the location parameters of the probability distribution of the remaining life of the heat pump system when the fault types are 1, 2 and K at the sampling time point τ.

[0043] Among them, {RUL,Z}(τ) represents the remaining life RUL of the heat pump system and the fault type Z at the sampling time point τ, and the fault type obeys the parameter π 1,τ , π 2,τ , π K,τ Classification distribution of

[0044] At the same time, according to the correlation between the remaining life of the heat pump system and the fault type, at the sampling time point τ, if the fault type Z = k is given, the remaining life of the heat pump system RUL|Z = k is subject to the scale parameter β k,τ , shape parameter is σ k,τ and the position parameter is λ k,τ The generalized gamma distribution of

[0045] In this embodiment, the generalized gamma distribution is used to simulate common probability distributions for fault remaining life prediction, such as Weibull distribution, gamma distribution, and lognormal distribution, thereby increasing the versatility of the model.

[0046] S32, add the following from each variable node {A1(τ), A2(τ), ..., A p (τ),B1(τ),B2(τ),…,B q (τ)} point to the node π τ , β τ , σ τ ,λ τ The directed edge of .

[0047] S33. Add slave node π τ , β τ , σ τ ,λ τ The directed edges pointing to the node {RUL, Z}(τ) respectively obtain the hierarchical graph model of monitoring variables, remaining life and fault types.

[0048] S4. Based on the hierarchical graph model and the time series chain graph degradation model, a competing risk time series chain graph degradation model is constructed, and the competing risk cumulative occurrence function is calculated.

[0049] In this embodiment, the historical operating environment data and operating status data of the heat pump system are used to quantify the dynamic relationship between the remaining life of the heat pump system, the fault type, the operating environment and the system status. The steps of constructing the competing risk time series chain graph degradation model are as follows: Figure 3As shown in the figure, it specifically shows how to use the historical operating environment data and operating status data of a given heat pump system to construct a competing risk time series chain graph degradation model, as follows:

[0050] Specifically, step S4 includes S41-S45:

[0051] S41. Based on the hierarchical graph model and the time series chain graph degradation model, a competing risk time series chain graph degradation model is constructed, namely:

[0052]

[0053] in, represents the competing risk time series chain graph degradation model, and π, β, σ, and λ are all latent variables that associate the monitoring variables with the remaining life of the heat pump system under different fault types.

[0054] In this embodiment, the purpose of constructing a competing risk time series chain graph degradation model is to define the correlation between the remaining life of the heat pump system under different fault types and latent variables, latent variables and monitoring variables, and the temporal changes between latent variables. That is, the historical operating environment data and operating status data of the heat pump system are used to determine the dynamic correlation between the fault type of the heat pump system and the time of fault occurrence, operating environment, and degradation process.

[0055] S42. Calculate the model parameters of the competing risk time series chain graph degradation model, and obtain the joint probability distribution function of all monitoring variables of the competing risk time series chain graph degradation model in each time period.

[0056] In this embodiment, the function implemented is to quantify the dynamic correlation between the remaining life of the heat pump system and the operating environment and the equipment degradation process.

[0057] Specifically, step S42 includes S421-S426:

[0058] S421, remove the time series chain graph degradation model All directed edges in the graph are removed, and the graph without directed edges is obtained. Among them, Figure Each maximal connected subgraph connected by undirected edges in Each chain component From the picture The variable node v in Represents a graph with directed edges removed The collection of all chain components in .

[0059] S422, if the chain components There is no parent node, that is Then the joint probability distribution function of the variable nodes contained in the chain component is:

[0060]

[0061] in, Represents a chain component The parent node set of represents the empty set, δ1 represents the chain component The parameter set of the joint probability distribution function of the variable nodes included, Represents a chain component The joint probability distribution function of the included variable nodes.

[0062] S423, if the chain components There is a parent node, then for the variable node v∈V with a parent node TS and Then the conditional probability distribution function of the variable node with a parent node is:

[0063]

[0064] Among them, pa(v) represents the graph The parent node set of the variable node, δ2 represents the parameter set of the conditional probability distribution function of the variable node v with a parent node, Represents the conditional probability distribution function of a variable node v with a parent node.

[0065] S424. Using the Bayesian parameter estimation method, calculate the parameter set of the joint probability distribution function of the variable nodes included in the chain component and the parameter set of the conditional probability distribution function of the variable nodes with parent nodes, that is:

[0066] δ={δ1,δ2}

[0067]

[0068] Where δ represents the chain component The parameter set of the joint probability distribution function of the contained variable nodes or the parameter set of the conditional probability distribution function of the variable node v with a parent node, Represents a heat pump system degradation time series dataset The probability distribution function of the parameter set δ is, Indicates the generation of a heat pump system degradation time series dataset under the condition of parameter set δ The probability of represents the probability distribution of the empirically initialized parameter set δ, Indicates that the heat pump system degradation time series dataset is collected probability.

[0069] S425, time series chain graph degradation model Decompose it into a subgraph consisting of each variable node and its parent node, and calculate the posterior distribution and probability distribution of the time series chain graph degradation model parameters, namely:

[0070]

[0071] Among them, δ v|pa(v) Represents the subgraph parameter consisting of the variable node v and the parent node of the variable node v, Represents a heat pump system degradation time series dataset The values of the monitoring variables corresponding to the variable node v and the parent node set pa(v) of the variable node v, Represents the posterior distribution of the subgraph parameters consisting of the variable node v and the parent node of the variable node v, Represents the probability distribution of the subgraph parameters consisting of the variable node v and the parent node of the variable node v.

[0072] S426, according to the conditional probability distribution function of the variable node with a parent node Joint probability distribution function of the variable nodes contained in the chain component Computational time series chain graph degradation model The joint probability distribution function of all monitored variables in each time period is:

[0073]

[0074] Among them, X(-T) represents the monitoring variable data at the sampling time point of -T, X(-1) represents the monitoring variable data at the sampling time point of -1, and X(0) represents the monitoring variable data at the sampling time point of 0. Represents the time series chain graph degradation model The joint probability distribution function of all monitoring variables (X(-T),…,X(-1),X(0)) in each time period, Represents a chain component The joint probability distribution function of .

[0075] S43. Establish a multivariate random forest regression model of latent variables for the competing risk time series chain graph degradation model, and calculate the probability distribution function of the remaining life and failure type of the heat pump system.

[0076] In this embodiment, the function implemented is to quantify the dynamic correlation between the remaining life of the heat pump system under various fault types and the operating environment and equipment degradation process.

[0077] Specifically, step S43 includes S431-S435:

[0078] S431. Initialize the initial points and initial covariance matrix of latent variables π, β, σ, and λ, and perform several iterations using a multivariate random forest regression model.

[0079] S432, at the nth iteration, according to the current parameter sample π (n) , β (n) , σ (n) ,λ (n) With the current covariance matrix Σ (n) , generate candidate parameter samples π′, β′, σ′, λ′, and use the Metropolis-Hastings method to calculate the probability of receiving the candidate parameter samples, that is:

[0080]

[0081] Among them, π (n) , β (n) , σ (n) ,λ (n) They represent the latent variables of the nth iteration, π′, β′, σ′, and λ′ represent the candidate latent variables generated by the nth iteration, α represents the probability of receiving the candidate parameter sample, and min represents the minimum value. Represents a time series dataset based on the degradation of the heat pump system The probability of the inferred candidate parameter samples π′, β′, σ′, λ′, Represents a time series dataset based on the degradation of the heat pump system Infer the current parameter sample π (n) , β (n) , σ (n) ,λ (n) probability.

[0082] S433: If the probability of receiving the candidate parameter sample is greater than the set critical value, the candidate parameter samples π′, β′, σ′, λ′ are generated when the nth iteration is accepted, and the parameter sample of the n+1th iteration is updated to π (n+1) =π ′ ,β (n+1) =β ′ ,σ (n+1) =σ ′ ,λ (n+1) =λ′, otherwise, keep the parameter sample of the n+1th iteration unchanged, which is π (n+1) =π (n) ,β (n+1) =β (n) ,σ (n+1) =σ (n) ,λ (n+1) =λ (n) .

[0083] Among them, π(n+1) , β (n+1) , σ (n+1) ,λ (n+1) represent the latent variables of the n+1th iteration respectively.

[0084] S434, based on the parameter samples generated in the first n+1 iterations, update the covariance matrix Σ of the n+1th iteration (n+1) ;

[0085] Among them, the parameter samples generated by the first n+1 iterations are π (1) ,β (1) ,σ (1) ,λ (1) ,…,π (n) ,β (n) ,σ (n) ,λ (n) ,π (n+1) ,β (n+1) ,σ (n+1) ,λ (n+1) .

[0086] Among them, π (1) , β (1) , σ (1) ,λ (1) represent the latent variables of the first iteration respectively.

[0087] S435, repeat the iterative process of steps S432-S434, when the difference between the parameter samples of two adjacent iterations is less than the set minimum critical value, By calculating the latent variables π, β, σ, and λ, we can obtain the probability distribution function of the remaining life of the heat pump system and the failure type, namely:

[0088]

[0089] in, Represents a time series dataset based on the degradation of the heat pump system The probability distribution of the latent variables is inferred to be π, β, σ, and λ, Indicates that the heat pump system degradation time series dataset is generated under the condition that the latent variables are π, β, σ, and λ The probability of Respectively represent the probability distribution of initializing latent variables to π, β, σ, and λ based on experience, represents the conditional probability distribution of the remaining life of the heat pump system when the fault type is Z at the sampling time point τ, represents the generalized gamma distribution function with latent variables β, σ, and λ, represents the probability distribution of the heat pump system failure type Z, Represents the classification distribution function with latent variable π.

[0090] S44. Based on the probability distribution function of the remaining life of the heat pump system and the failure type, and the joint probability distribution function of all monitoring variables in the competing risk time series chain graph degradation model in each time period, calculate the joint probability distribution function of the monitoring variables, the remaining life of the heat pump system, and the failure type, namely:

[0091]

[0092] in, represents the joint probability distribution function of the monitoring variables {X(τ),τ=-T,…,-2,-1,0} and the remaining life of the heat pump system is RUL and the fault type is Z at the sampling time point τ, s represents the specific sampling time point, represents the joint probability distribution of the remaining life RUL of the heat pump system and the fault type Z at the sampling time point τ, represents the conditional probability distribution of the remaining life RUL and fault type Z of the heat pump system at the sampling time point τ when the hidden variables are given as π, β, σ, and λ, represents the joint probability distribution when the latent variables are σ and λ, represents the conditional probability distribution of latent variables π and β when the monitoring variable group X(τ) is observed, represents the probability distribution of observing the monitoring variable group X(s), Indicates that the jth monitoring variable X is observed j The probability distribution of (τ).

[0093] In this embodiment, the function implemented is to quantify the dynamic correlation between the remaining life of the heat pump system, the fault type, the operating environment, and the equipment degradation process.

[0094] S45. Calculate the cumulative occurrence function of competitive risk based on the joint probability distribution function of the monitored variables, the remaining life of the heat pump system, and the failure type, namely:

[0095]

[0096] in,

[0097] Where t represents time, x represents the time series data value monitored during the operation of the heat pump system, and F(t,k|x) represents the cumulative probability distribution function of a set of time series data values x, remaining life t, and fault type k monitored during the operation of the heat pump system. It represents the conditional probability that the remaining life of the heat pump system is less than t and the fault type is k, given the observed time series data value x of the monitoring variable during the operation of the heat pump system. represents the joint probability that the remaining life of the heat pump system is less than t, the fault type is k, and the time series data of the monitoring variables during the operation of the heat pump system are observed to take the value x, represents the probability of observing the time series data of the monitoring variable taking the value x during the operation of the heat pump system, Represents the joint probability distribution of time series data, remaining life, and fault type monitored during the operation of the heat pump system.

[0098] In this embodiment, the function implemented is: based on a set of collected operating environment and operating status data, the remaining life of the heat pump system and the probability of failure type are calculated.

[0099] S5. Obtain the time series data monitored during the operation of the heat pump system, and predict the remaining life and fault type of the heat pump system based on the competing risk cumulative occurrence function.

[0100] In this embodiment, the cumulative probability of observing the time series data to be predicted under each fault type is calculated by the calculation model, and for each fault type, the empirical distribution function of the time from the heat pump to the occurrence of the fault is calculated, and the time period and fault type that achieve the maximum value are found, so as to obtain the predicted fault type and the remaining life under the fault type.

[0101] Specifically, step S5 includes S51-S57:

[0102] S51. Obtain the time series data {x0(t), z0} monitored during the operation of the heat pump system, where x0(t) represents the time series data of the monitoring variables during the operation of the heat pump system observed at the t-th moment, and z0 represents the type of fault when a fault occurs in the observed heat pump system.

[0103] If the time length of the monitoring variable time series data x0(t) is l, and the actual start time of the monitoring variable time series data x0(t) in the degradation stage is S, when S = -T, then the model fragment of the competing risk time series chain graph degradation model is:

[0104]

[0105] When S=-T, the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is: -T+l-1.

[0106] S52. Based on the competition risk cumulative occurrence function, calculate the competition risk cumulative probability of each fault type when the model time slice E is -T+1-1, that is:

[0107] F(-T+l-1,k=1|x0(t)),F(-T+l-1,k=2|x0(t)),…,F(-T+l-1,k

[0108] =K|x0(t))

[0109] Among them, F(-T+l-1,k=1|x0(t)) represents the cumulative probability of competing risks when fault type k is 1 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l-1, F(-T+l-1,k=2|x0(t)) represents the cumulative probability of competing risks when fault type k is 2 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l-1, and F(-T+l-1,k=K|x0(t)) represents the cumulative probability of competing risks when fault type k is K when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l-1.

[0110] S53. Shift the model segment of the competing risk time series chain graph degradation model backward by one time period. That is, when S=-T+1, the model segment of the competing risk time series chain graph degradation model is:

[0111]

[0112] Among them, when S=-T+1, the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is: -T+1.

[0113] S54. Based on the competition risk cumulative occurrence function, calculate the competition risk cumulative probability of each fault type when the model time slice E is -T+1, that is:

[0114] F(-T+l,k=1|x0(t)),F(-T+l,k=2|x0(t)),…,F(-T+l,k=K|x0(t))

[0115] Among them, F(-T+l,k=1|x0(t)) represents the cumulative probability of competing risks when the fault type k is 1 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l, F(-T+l,k=2|x0(t)) represents the cumulative probability of competing risks when the fault type k is 2 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l, and F(-T+l,k=K|x0(t)) represents the cumulative probability of competing risks when the fault type k is K when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is -T+l.

[0116] S55. Repeat steps S51-S54, each time shifting the model segment of the competing risk time series chain graph degradation model backward by one time period until the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is 0, and calculate the competing risk cumulative probability of each fault type when the model time slice E is 0, that is:

[0117] F(0,k=1|x0(t)),F(0,k=2|x0(t)),…,F(0,k=K|x0(t))

[0118] Among them, F(0,k=1|x0(t)) represents the cumulative probability of competing risks when the fault type k is 1 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is 0, F(0,k=2|x0(t)) represents the cumulative probability of competing risks when the fault type k is 2 when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is 0, and F(0,k=K|x0(t)) represents the cumulative probability of competing risks when the fault type k is K when the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t) is 0.

[0119] S56. Calculate the empirical distribution function of the remaining life of the heat pump system based on the cumulative probability of competing risks for each fault type under different model time slices, namely:

[0120]

[0121] in, represents the empirical distribution function of the fault type k of the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t), F(E+1,k|x0(t)) represents the cumulative probability of competing risks of the fault type k of the model time slice E+1 corresponding to the end of the time series of the monitoring variable time series data x0(t), and F(E,k|x0(t)) represents the cumulative probability of competing risks of the fault type k of the model time slice E corresponding to the end of the time series of the monitoring variable time series data x0(t).

[0122] S57. Based on the empirical distribution function of the remaining life of the heat pump system, by comparing the empirical distribution functions of different remaining lifespans of the heat pump system, find the time period and fault type corresponding to the maximum value and record it as e * and k * , then: The predicted value of the remaining life of the heat pump system is The prediction value of the corresponding fault type is

[0123] In summary, the present invention proposes a method for predicting heat pump system faults based on a time series chain graph. Based on the time series chain graph, a competing risk time series chain graph degradation model is constructed, which has flexible structural properties and can accurately characterize dynamics; at the same time, it improves the accuracy and precision of heat pump fault detection and diagnosis, and reduces the instability of prediction capabilities caused by the complexity and diversity of heat pump system fault types, the interconnection and propagation between faults, and the high correlation between faults and dynamically changing working conditions; in addition, while reducing modeling costs and improving modeling efficiency, the model can also be used to present the degradation mechanism and regular changes of the heat pump system throughout its life cycle; finally, through data-driven combined with mechanism models, it can provide a basis for the predictive maintenance function of the heat pump system of the digital intelligent energy operation and maintenance management system.

[0124] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0125] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A heat pump system fault prediction method based on a time series chain diagram, characterized in that: The following steps are involved: S1. Configure and deploy a sensor group to collect monitoring variable data of the heat pump system from normal operation to the moment of failure, record the failure type of the heat pump system, and construct a time series dataset of heat pump system degradation; Among them, the monitoring variable data includes operating status data and operating environment data; S2. Based on the heat pump system degradation time series dataset, a time series chain graph degradation model of the heat pump system from normal operation to the moment of failure is constructed; S3. Based on the time series chain graph degradation model, a hierarchical graph model of monitoring variables, remaining life, and fault types is constructed; S4. Based on the hierarchical graph model and the time series chain graph degradation model, a competing risk time series chain graph degradation model is constructed, and the competing risk cumulative occurrence function is calculated; S5. Obtain the time series data monitored during the operation of the heat pump system, and predict the remaining life and fault type of the heat pump system based on the competing risk cumulative occurrence function.

2. The heat pump system fault prediction method based on time series chain diagram according to claim 1 is characterized in that: Step S1 specifically includes: S11. Configure and deploy a first sensor group on the main equipment, components, and parts of the heat pump system, and configure and deploy a second sensor group in the operating environment of the heat pump system; The first sensor group is used to monitor operating status data composed of several monitoring variables including temperature, pressure, flow, and vibration; The second sensor group is used to monitor operating environment data composed of several monitoring variables including temperature and humidity; S12. Set the sampling frequency and number of sampling times, use the first sensor group and the second sensor group to collect monitoring variable data of the heat pump system from normal operation to the time of failure, and record the failure type of the heat pump system to construct a heat pump system degradation time series data set, that is: in, represents the heat pump system degradation time series dataset, Indicates the total number of sampling times, Indicates the number of sampling times, Indicates the The monitoring variable data of the subsamples, Indicates the The heat pump system fault type of the sub-samples, represents the sampling time point, Indicates The sampling time point The operating environment data of the subsamples, Indicates The sampling time point The running status data of the subsamples, Indicates the heat pump system fault type. Indicates the total number of heat pump system fault types, Indicates the The sampling time is The data of the first operating environment monitoring variable at the sampling time point, Indicates the The sampling time is The data of the second operating environment monitoring variable at the sampling time point, Indicates the The sampling time is Sampling time point Data of operating environment monitoring variables, Indicates the The sampling time is The data of the first operating status monitoring variable at the sampling time point, Indicates the The sampling time is The data of the second operating status monitoring variable at the sampling time point, Indicates the The sampling time is Sampling time point The data of the operating status monitoring variables.

3. The heat pump system fault prediction method based on time series chain diagram according to claim 2 is characterized in that: Step S2 specifically includes: S21. Based on the heat pump system degradation time series dataset, a probabilistic graphical model structure learning method is used to establish an undirected graphical model for each monitoring variable; Among them, the undirected graph model of each monitoring variable is composed of all variable nodes and edges connecting variable nodes, and the variable nodes are nodes of the monitoring variables; S22. Based on the undirected graph model of each monitoring variable, a time series chain graph degradation model of the heat pump system from normal operation to the moment of failure is constructed, specifically including: S221, converting the undirected edges connecting the variable nodes at different sampling time points in the undirected graph model into directed edges pointing from the earlier sampling time point to the later sampling time point; S222, converting the undirected edges connecting the variable nodes of different fault types at the same sampling time point in the undirected graph model into directed edges pointing from the operating environment data to the operating status data; S223. The undirected edges connecting the variable nodes of the same fault type at the same sampling time point in the undirected graph model are retained unchanged, and a time series chain graph degradation model of the heat pump system from normal operation to the moment of fault occurrence is obtained, namely: in, represents the time series chain graph degradation model, A set of variable nodes representing the time series chain graph degradation model, represents the set of directed and undirected edges in the time series chain graph degradation model, Indicates Monitoring variable data at the sampling time point, Indicates Operating environment data at the sampling time point, Indicates Operation status data at the sampling time point, Indicates The data of the first operating environment monitoring variable at the sampling time point, Indicates The data of the second operating environment monitoring variable at the sampling time point, Indicates Sampling time point Data of operating environment monitoring variables, Indicates The data of the first operating status monitoring variable at the sampling time point, Indicates The data of the second operating status monitoring variable at the sampling time point, Indicates Sampling time point The data of the operating status monitoring variables.

4. The heat pump system fault prediction method based on time series chain diagram according to claim 3 is characterized in that: Step S3 specifically includes: S31, based on the time series chain graph degradation model, for each time sampling point , add nodes 、 、 、 as well as ; in, 、 、 and , Indicates Parameters of the probability distribution of heat pump system failure types at sampling time points, 、 、 Respectively expressed in The scale, shape and location parameters of the probability distribution of the remaining life of the heat pump system for a given fault type at the sampling time point, 、 、 Respectively expressed in If a failure occurs in the heat pump system at the sampling time, the failure type is 、 as well as The probability of 、 、 Respectively expressed in The fault type given at the sampling time point is 、 as well as When , the scale parameter of the probability distribution of the remaining life of the heat pump system is, 、 、 Respectively expressed in The fault type given at the sampling time point is 、 as well as When , the shape parameter of the probability distribution of the remaining life of the heat pump system is, 、 、 Respectively expressed in The fault type given at the sampling time point is 、 as well as When , the location parameter of the probability distribution of the remaining life of the heat pump system; in, Indicates Remaining life of the heat pump system at the sampling time point and fault type , and the fault type is subject to the parameter 、 、 Classification distribution of ; At the same time, according to the correlation between the remaining life of the heat pump system and the fault type, at the sampling time point , if the fault type is given , then the remaining life of the heat pump system is The scale parameter is , the shape parameters are and the positional parameters are The generalized gamma distribution of ; S32, add from each variable node Point to the nodes respectively 、 、 、 ; S33. Add slave nodes 、 、 、 Point to the nodes respectively The directed edges of are used to obtain the hierarchical graph model of monitoring variables, remaining life and fault types.

5. The heat pump system fault prediction method based on time series chain diagram according to claim 4 is characterized in that: Step S4 specifically includes: S41. Based on the hierarchical graph model and the time series chain graph degradation model, a competing risk time series chain graph degradation model is constructed, namely: in, represents the competing risk time series chain graph degradation model, 、 、 、 All represent hidden variables that relate the monitored variables to the remaining life of the heat pump system under different fault types; S42. Calculate the model parameters of the competing risk time series chain graph degradation model, and obtain the joint probability distribution function of all monitoring variables of the competing risk time series chain graph degradation model in each time period; S43. Establish a multivariate random forest regression model for the latent variables of the competing risk time series chain graph degradation model and calculate the probability distribution function of the remaining life and failure type of the heat pump system; S44. Calculate the joint probability distribution function of the monitoring variables, the remaining life of the heat pump system, and the failure type based on the probability distribution function of the remaining life of the heat pump system and the failure type and the joint probability distribution function of all monitoring variables in the competing risk time series chain graph degradation model at each time period; S45. Calculate the cumulative occurrence function of competitive risk based on the joint probability distribution function of the monitored variables, the remaining life of the heat pump system, and the failure type.

6. The heat pump system fault prediction method based on time series chain diagram according to claim 5 is characterized in that: Step S42 specifically includes: S421, remove the time series chain graph degradation model All directed edges in the graph are removed, and the graph with directed edges removed is obtained. ; Among them, Figure Each maximal connected subgraph connected by undirected edges in , each chain component From the picture Variable nodes in Composition, and 、 , Represents a graph with directed edges removed The set of all chain components in; S422, if the chain components There is no parent node, that is , then the joint probability distribution function of the variable nodes contained in the chain component is: in, Represents a chain component The parent node set of represents the empty set, Represents a chain component The parameter set of the joint probability distribution function of the variable nodes included, Represents a chain component The joint probability distribution function of the included variable nodes; S423, if the chain components If there is a parent node, then for the variable node with a parent node and , then the conditional probability distribution function of the variable node with a parent node is: in, Representation diagram The parent node set of the variable node, Represents a variable node with a parent node The parameter set of the conditional probability distribution function, Represents a variable node with a parent node The conditional probability distribution function of ; S424. Using the Bayesian parameter estimation method, calculate the parameter set of the joint probability distribution function of the variable nodes included in the chain component and the parameter set of the conditional probability distribution function of the variable nodes with parent nodes, that is: in, Represents a chain component The parameter set of the joint probability distribution function of the contained variable nodes or the variable nodes with parent nodes The parameter set of the conditional probability distribution function, Represents a heat pump system degradation time series dataset Parameter set The probability distribution function of Indicates in the parameter set Generate a heat pump system degradation time series dataset under the conditions of The probability of Indicates that the parameter set is initialized based on experience The probability distribution of Indicates that the heat pump system degradation time series dataset is collected probability; S425, time series chain graph degradation model Decompose it into a subgraph consisting of each variable node and its parent node, and calculate the posterior distribution and probability distribution of the time series chain graph degradation model parameters, namely: in, Represents a variable node With variable nodes The subgraph parameters composed of the parent nodes, Represents a heat pump system degradation time series dataset Variable Node and variable nodes The parent node set The corresponding monitoring variable value, Represents a variable node With variable nodes The posterior distribution of the subgraph parameters composed of the parent nodes, Represents a variable node With variable nodes The probability distribution of the subgraph parameters composed of the parent nodes; S426, according to the conditional probability distribution function of the variable node with a parent node , the joint probability distribution function of the variable nodes contained in the chain component , calculate the time series chain graph degradation model The joint probability distribution function of all monitored variables in each time period is: in, Indicates Monitoring variable data at the sampling time point, Indicates Monitoring variable data at the sampling time point, Indicates Monitoring variable data at the sampling time point, Represents the time series chain graph degradation model All monitored variables The joint probability distribution function at each time period is, Represents a chain component The joint probability distribution function of .

7. The heat pump system fault prediction method based on time series chain diagram according to claim 6 is characterized in that: Step S43 specifically includes: S431, initialize hidden variables 、 、 、 The initial points and initial covariance matrix are used, and the multivariate random forest regression model is used for several iterations; S432, in At the iteration, according to the current parameter sample 、 、 、 With the current covariance matrix , generate candidate parameter samples 、 、 、 , and use the Metropolis-Hastings method to calculate the probability of receiving the candidate parameter sample, that is: in, 、 、 、 Respectively represent The hidden variables of the iteration, 、 、 、 Respectively represent The candidate latent variables generated by the iteration are represents the probability of receiving the candidate parameter sample, Indicates taking the minimum value, Represents a time series dataset based on the degradation of the heat pump system Inferred candidate parameter samples 、 、 、 The probability of Represents a time series dataset based on the degradation of the heat pump system Infer the current parameter sample 、 、 、 probability; S433. If the probability of receiving the candidate parameter sample is greater than the set critical value, then accept the first Generate candidate parameter samples at the iteration 、 、 、 , and update the The parameter sample of the iteration is Otherwise, keep The parameter sample of the iteration remains unchanged, which is ; in, 、 、 、 Respectively represent Hidden variables of the iteration; S434, according to the previous The parameter samples generated by the iteration are updated The covariance matrix of the iteration ; Among them, the former The parameter samples generated by the iteration are ; in, 、 、 、 Respectively represent the hidden variables of the first iteration; S435, repeat the iterative process of steps S432-S434, when the difference between the parameter samples of two adjacent iterations is less than the set minimum critical value, , calculate the latent variables 、 、 、 , we can get the probability distribution function of the remaining life of the heat pump system and the failure type, namely: in, Represents a time series dataset based on the degradation of the heat pump system The latent variable is inferred to be 、 、 、 The probability distribution of Indicates that the latent variable is 、 、 、 Generate a heat pump system degradation time series dataset under the conditions of The probability of 、 、 、 They represent the initialization of latent variables based on experience 、 、 、 The probability distribution of Indicates The fault type at the sampling time point is The conditional probability distribution of the remaining life of the heat pump system is The latent variable is 、 、 The generalized gamma distribution function of Indicates that the heat pump system fault type is The probability distribution of The latent variable is The classification distribution function of .

8. The heat pump system fault prediction method based on time series chain diagram according to claim 7 is characterized in that: The calculation formula of the joint probability distribution function of the monitoring variables, the remaining life of the heat pump system, and the fault type in step S44 is: in, Indicates monitoring variables With The remaining life of the heat pump system at the sampling time is , the fault type is The joint probability distribution function of Indicates the specific sampling time point, Indicates Remaining life of the heat pump system at the sampling time point and fault type The joint probability distribution of Indicates that given the latent variable is 、 、 、 Time Remaining life of the heat pump system at the sampling time point and fault type The conditional probability distribution of The latent variable is 、 The joint probability distribution of Indicates that when the monitoring variable group is observed Time-hidden variables 、 The conditional probability distribution of Indicates that the monitoring variable group is observed The probability distribution of Indicates that the observed monitoring variables The probability distribution of Indicates the given monitoring variables When the parent node set of monitoring variables The conditional probability distribution of .

9. The heat pump system fault prediction method based on time series chain diagram according to claim 8, characterized in that: The calculation formula of the competition risk cumulative occurrence function in step S45 is: in, , ; in, Indicates time, Indicates the time series data value monitored during the operation of the heat pump system. Indicates a set of time series data values monitored during the operation of a known heat pump system , Remaining life and fault type The cumulative probability distribution function of Represents the time series data value of the monitoring variables during the operation of the given heat pump system , the remaining life of the heat pump system is less than And the fault type is The conditional probability of Indicates that the remaining life of the heat pump system is less than , the fault type is And the time series data values of the monitoring variables during the operation of the heat pump system are observed The joint probability of Indicates the time series data values of the monitoring variables observed during the operation of the heat pump system The probability of Represents the joint probability distribution of time series data, remaining life, and fault type monitored during the operation of the heat pump system.

10. The heat pump system fault prediction method based on time series chain diagram according to claim 9, characterized in that: Step S5 specifically includes: S51. Obtaining time series data monitored during the operation of the heat pump system ,in, Indicates the The time series data of monitoring variables during the operation of the heat pump system observed at all times, Indicates the type of failure observed when the heat pump system fails; If the monitoring variable time series data The length of time is , monitoring variable time series data The actual start time of the degradation phase is ,when When , the model fragment of the competing risk time series chain graph degradation model is: Among them, when When monitoring variable time series data The model time slice corresponding to the end of the time series for: ; S52, based on the competition risk cumulative occurrence function, calculate the time slice of each fault type in the model for The cumulative probability of competing risks when , that is: in, Represents monitoring variable time series data The model time slice corresponding to the end of the time series for Time Fault Type The cumulative probability of competing risks is 1, Represents monitoring variable time series data The model time slice corresponding to the end of the time series for Time Fault Type The cumulative probability of competing risks is 2, Represents monitoring variable time series data Model time slice corresponding to the end of the time series for Time Fault Type for The cumulative probability of competing risks; S53. Shift the model segment of the competing risk time series chain graph degradation model backward by one time period, that is, when When , the model fragment of the competing risk time series chain graph degradation model is: Among them, when When monitoring variable time series data The model time slice corresponding to the end of the time series for: ; S54, based on the competition risk cumulative occurrence function, calculate the time slice of each fault type in the model for The cumulative probability of competing risks when , that is: in, Represents monitoring variable time series data The model time slice corresponding to the end of the time series for Time Fault Type The cumulative probability of competing risks is 1, Represents monitoring variable time series data The model time slice corresponding to the end of the time series for Time Fault Type The cumulative probability of competing risks is 2, Represents monitoring variable time series data The model time slice corresponding to the end of the time series for Time Fault Type for The cumulative probability of competing risks; S55, repeat steps S51-S54, each time shifting the model segment of the competing risk time series chain graph degradation model backward by one time period until the monitoring variable time series data The model time slice corresponding to the end of the time series When it is 0, calculate the time slice of each fault type in the model for The cumulative probability of competing risks when , that is: in, Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type when it is 0 The cumulative probability of competing risks is 1, Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type when it is 0 The cumulative probability of competing risks is 2, Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type when it is 0 for The cumulative probability of competing risks; S56. Calculate the empirical distribution function of the remaining life of the heat pump system based on the cumulative probability of competing risks for each fault type under different model time slices, namely: in, Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type The empirical distribution function of Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type The cumulative probability of competing risks, Represents monitoring variable time series data The model time slice corresponding to the end of the time series Fault type The cumulative probability of competing risks; S57. Based on the empirical distribution function of the remaining life of the heat pump system, by comparing the empirical distribution functions of different remaining lifespans of the heat pump system, find the time period and fault type corresponding to the maximum value and record them as and , then: The estimated remaining life of the heat pump system is , the corresponding prediction value of the fault type is .

Images