Method for predicting service life of electromechanical equipment

Abstract

The invention belongs to the technical field of electromechanical equipment management and maintenance, and particularly relates to an electromechanical equipment service life prediction method. The method mainly comprises the following steps: determining a life expectancy association attribute sequence according to the type of electromechanical equipment; determining to-be-analyzed data and initial reference data; screening samples from the samples to establish a sample database, selecting a proper data period, intercepting corresponding periodic monitoring data by adopting a sliding time window, performing similarity analysis to determine similarity, extracting effective data, calculating periodic change rate deviation, and optimizing the samples to establish an effective reference database; establishing a prediction weight and prediction based on the time sequence similarity of the sample data; the method for predicting the life of the electromechanical equipment is mainly used for providing a method which is suitable for novel equipment, can better consider the relevance between historical measurement data and the life cycle, displays the relation between the running states of the electromechanical equipment in different life stages and the final expected life, and better analyzes and predicts the expected residual life of the electromechanical equipment.

Description

Technical Field

[0001] The invention belongs to the technical field of electromechanical equipment management and maintenance, and in particular relates to a method for predicting the life of electromechanical equipment. Background Art

[0002] The life prediction of electromechanical equipment refers to predicting the remaining life of the equipment based on the various operating parameters or attributes of the current equipment and the operating data of similar equipment, so as to plan and design the replacement and update process of electromechanical equipment in advance, ensure the continuity of equipment functions, and avoid unnecessary downtime and other problems. The current life prediction of electromechanical equipment is mainly based on the current attributes or parameters of the equipment for analysis and prediction, and rarely considers the correlation between historical cycle data and equipment life. This is because the expected life of most equipment is often strongly related to the available state of a small number of core structures. These core structures often have a relatively fixed expected life and have a very clear and predictable state change process when they are damaged. Therefore, by paying attention to the current attributes or parameters and then determining the state of the core structure, the expected remaining life of the equipment can be estimated. However, with the use of flexible equipment, supplemented by increasingly mature equipment maintenance technology and capabilities, the life cycle of electromechanical equipment will be further extended in the future, and its expected life will also be further related to the specific operation process at different stages of its entire life cycle. In this case, the traditional life prediction method cannot well reflect the correlation of the corresponding life cycle, and the effectiveness of the equipment life prediction results cannot be further improved. Summary of the invention

[0003] The purpose of the present invention is to propose a method for better predicting the life of new electromechanical equipment by considering historical measurement data during the life cycle of the equipment in combination with actual needs.

[0004] To achieve the above purpose, the present invention adopts the following technical solution.

[0005] The present invention provides a method for predicting the life of electromechanical equipment, which mainly includes the following steps:

[0006] Stp1, determine the expected life associated attribute sequence S = {s1, s2...s j ....s J}; Based on the aforementioned associated attribute sequence, determine the existing monitoring data containing each attribute in the associated attribute sequence from the monitoring data of the electromechanical equipment a0 to be analyzed as the data to be analyzed; from the existing electromechanical equipment samples of the same type {a i} selects historical monitoring data covering the corresponding period or stage as initial reference data;

[0007] Stp2, select the optimal data to establish a database, select samples from existing samples to establish a sample database; select a suitable data cycle T based on the characteristics of the data collection cycle of the sample to be analyzed and the same type of equipment, T≥max(k j T j );T j Any attribute element s j The sampling period, k j To ensure that the attribute element s j The minimum period for which sample data is valid;

[0008] For sample a i Attribute sequence S = {s1, s2...s j ....s J Each attribute element s in j , using a sliding time window ΔT = T j Intercept the corresponding periodic monitoring data, and intercept the attribute element s based on the sliding time window ΔT j The monitoring data generates a periodic monitoring data curve, and performs similarity analysis with the periodic monitoring data curve of the corresponding attribute element of the sample to be analyzed to determine the similarity of each attribute element s j The time window with the largest similarity of periodic monitoring data is extracted and retained as valid data to obtain sample a. i Attribute elements of j The closest attribute value sequence X in the corresponding period ij ={x ij1 ,x ij2 ,x ij3 ...x ijn ...x ijNj}, Draw the parameter change curve within the time window ΔT and determine the sample a i Attribute elements of j The periodic rate of change w ij ;

[0009] in

[0010] Similarly, obtain the parameter periodic change rate of the sample a0 to be analyzed

[0011] Calculate historical data to calculate the sample a0 to be analyzed and each existing sample a i The cycle rate deviation where x ijn represents the sampling data of the j-th parameter of sample i in the n-th acquisition cycle; T j For attribute element s j The sampling period of

[0012] According to the principle of increasing the periodic change rate deviation from small to large, the existing sample a i Sorting and selecting several existing samples, forming a valid data set for each sample with corresponding valid data, and organizing each sample to establish a valid reference database containing all samples;

[0013] Stp3, establish prediction weights and predictions based on the time series similarity of sample data, including obtaining a sample matrix to be analyzed expressed by a multidimensional vector based on the data to be analyzed And the sample matrix to be analyzed

[0014]

[0015] Based on the previous step sample a i Attribute elements of j The periodic rate of change w ij The calculation method determines the sample period change matrix W to be analyzed j ′ and the sample matrix to be analyzed W ij ;in:

[0016]

[0017] Since the change of equipment life cycle is often accompanied by the continuous change of equipment performance indicators, the attribute data of each time node cannot well reflect the time series change trend of the corresponding attribute characteristics. According to experience, when the life stage of a certain attribute data is closer to the current life stage of the product to be analyzed, the life expectancy of the equipment under the current data is also closer to the life expectancy of the product to be analyzed; without reflecting this influence, a stage matching index ω is established to characterize the degree of matching between the life stages of the sample and the product to be analyzed, 0<ω≤1, and the stage matching index ω characterizes the degree of closeness between the life cycle node of the data and the current life cycle node of the equipment to be analyzed. The closer the stage matching index ω is, the larger the stage matching index ω is, and vice versa. The faster the equipment life cycle changes, the larger the amplitude of the stage matching index ω is, and vice versa.

[0018] Based on the similarity measurement principle, determine the periodic change matrix W of the sample to be analyzed j ′ and the reference sample matrix W ij The similarity can be expressed as For the similarity d(W J ′,W iJ ) is dimensionless, and the reference a considering the life cycle matching is obtained. i The degree of influence of the sample relative to the sample to be analyzed a0

[0019] In order to determine the influence weights of multiple reference samples on the life prediction of the sample to be analyzed, and to ensure that the weight value range is [0,1], a positive real number β is given, and the reference a is obtained. i The weight expression of the sample relative to the sample to be analyzed a0 After normalization, its weight is expressed as Where I is the total number of reference samples;

[0020] Then the predicted life of the sample to be analyzed based on the reference sample can be expressed as:

[0021]

[0022] Among them, RUL(a0|t) represents the expected remaining life of the sample a0 to be analyzed at time node t, RUL(a i |t) represents the reference sample a at time node t i remaining life.

[0023] Further improvement or specific implementation of the above-mentioned electromechanical equipment life prediction method, the data to be analyzed and the initial reference data include each associated attribute s in the associated attribute sequence j The original monitoring data shall include multiple groups of original monitoring data under different working conditions of the normal operation of the electromechanical equipment.

[0024] Further improvement or specific implementation of the above-mentioned electromechanical equipment life prediction method, in step Stp3, since each attribute element s j The total amount of attribute data is different due to different collection cycles. In the actual implementation process, the matrix parameters are optimized by zero-filling and normalization to facilitate calculation and processing.

[0025] Further improvement or specific implementation of the above-mentioned electromechanical equipment life prediction method, in the step Stp1, in order to ensure the validity of each attribute element in the associated attribute sequence S and ensure that the attribute element at the screening position can represent the expected life of the electromechanical equipment, each element in the associated attribute is determined based on the following method:

[0026] A1. First, obtain the original monitoring data and perform necessary preprocessing, including deleting and revising erroneous data and normalizing the data. The original monitoring data should include multiple sets of original monitoring data under different working conditions of the normal operation of the electromechanical equipment;

[0027] After necessary preprocessing, the original data is segmented to obtain the training data set and the verification data set;

[0028] A2. As shown in FIG. XXX, a multi-layer directed feature extraction model consisting of multiple restricted Boltzmann machine network models (RBN) is established, wherein the restricted Boltzmann machine network model includes a visible layer v and a hidden layer h, the visible layer is used as the input side to obtain sample data, and the hidden layer is used as the output end to extract the main features of the sample and output it in the form of probability distribution;

[0029] In order to optimize the data feature extraction effect, enhance the integrity of the data and avoid falling into the local optimum, the global weight change D is used j (m) and local weight changes control the weight changes of hidden layer neurons, and the algorithm is optimized by increasing the number of hidden layers and hidden layer neurons. Specifically:

[0030] The condition for increasing the number of hidden layer neurons can be expressed as:

[0031] The condition for increasing the number of hidden layers can be expressed as:

[0032] Where D j (m)=||d j (m),d j (m-1)||2;

[0033] d j (m) is the number of neurons in hidden layer h after m iterations j The weight vector of refers to the neuron h j The global weight change D for the nth training sample j (m); R is the total number of samples, and R′ is the change in global weight during the last iteration D j (m) increases the number of samples; where f(m) is the maximum global weight change D that ensures that each hidden layer neuron can correspond to a training sample after the number of hidden layer neurons increases j The threshold function of (m), f(m) can be expressed as:

[0034]

[0035] m max is the maximum number of iterations, ρ m Represents the slope of the variable threshold function curve at the corresponding position, f max is the maximum value of the threshold function, f min is the minimum value of the threshold function, e l (v r ,h l θ l ) represents the energy function, is the preset control threshold; lRBM Indicates the number of hidden layers, maxl RBM Indicates the maximum number of hidden layers;

[0036] During the training process, the hidden layer neurons h are analyzed. j If the condition of increasing the number of neurons in the hidden layer is met, a new neuron is added based on the neuron and the parameters of the two neurons are reset to zero. RBM Whether the conditions for increasing the number of hidden layers are met. If so, copy the hidden layer and keep the original parameters unchanged;

[0037] A3. Based on the above steps, a multi-layer directed factor extraction model is established and optimized, and the training data is input into the model for training. The model is optimized and trained using the validation data set to obtain the final model. The data change characteristics of the sample health status change process are extracted through the model, and the correlation between these characteristic attributes and the associated characteristics of the health status of electromechanical equipment is used to screen and determine the preferred associated attribute sequence. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic diagram of a flow chart of a method for predicting the life of electromechanical equipment in an embodiment;

[0039] Figure 2 It is a schematic diagram of a multi-layer directed feature extraction model based on a multi-RBN network. DETAILED DESCRIPTION

[0040] The electromechanical equipment life prediction method of the present application is mainly used to provide a method suitable for new equipment, which can better consider the correlation between historical measurement data and life cycle, visualize the relationship between the operating status of electromechanical equipment at different life stages and the final expected life, and better analyze and predict the expected remaining life of electromechanical equipment.

[0041] The present invention is described in detail below in conjunction with specific embodiments.

[0042] like Figure 1 As shown, the electromechanical equipment life prediction method of the present application mainly includes the following steps:

[0043] Step Stp1: Necessary data collection;

[0044] Determine the expected life span associated attribute sequence S = {s1, s2...s j ....s J}; Based on the aforementioned associated attribute sequence, determine the existing monitoring data containing each attribute in the associated attribute sequence from the monitoring data of the electromechanical equipment a0 to be analyzed as the data to be analyzed;

[0045] From the existing samples of electromechanical equipment of the same type {ai} selects historical monitoring data covering the corresponding period or stage as initial reference data;

[0046] The data to be analyzed and the initial reference data include the associated attributes s in the associated attribute sequence j Historical monitoring data;

[0047] In particular, in this embodiment, in order to ensure the validity of each attribute element in the associated attribute sequence S and ensure that the attribute element at the screening location can characterize the expected life of the electromechanical device, the present application determines each element in the associated attribute based on the following method:

[0048] A1. First, obtain the original monitoring data and perform necessary preprocessing, including deleting and revising erroneous data and normalizing the data. The original monitoring data should include multiple sets of original monitoring data under different working conditions of the normal operation of the electromechanical equipment;

[0049] After necessary preprocessing, the original data is segmented to obtain the training data set and the verification data set;

[0050] A2. Figure 2 As shown, a multi-layer directed feature extraction model consisting of multiple restricted Boltzmann machine network models (RBN) is established, wherein the restricted Boltzmann machine network model includes a visible layer v and a hidden layer h, the visible layer is used as the input side to obtain sample data, and the hidden layer is used as the output end to extract the main features of the sample and output it in the form of probability distribution;

[0051] In order to optimize the data feature extraction effect, enhance the integrity of the data and avoid falling into the local optimum, so as to retain the correlation between the attribute parameters and the equipment life in the whole life cycle of the equipment, the global weight change D j (m) and local weight changes control the weight changes of hidden layer neurons, and optimize the algorithm model by increasing the number of hidden layers and hidden layer neurons. Specifically:

[0052] The condition for increasing the number of hidden layer neurons can be expressed as:

[0053] The condition for increasing the number of hidden layers can be expressed as:

[0054] Where D j (m)=||d j (m),d j (m-1)||2;

[0055] d j (m) is the number of neurons in hidden layer h after m iterations j The weight vector of refers to the neuron h j The global weight change D for the nth training sample j (m); R is the total number of samples, and R′ is the change in global weight during the last iteration D j (m) increases the number of samples; where f(m) is the maximum global weight change D that ensures that each hidden layer neuron can correspond to a training sample after the number of hidden layer neurons increases j The threshold function of (m), f(m) can be expressed as:

[0056]

[0057] m max is the maximum number of iterations, ρ m Represents the slope of the variable threshold function curve at the corresponding position, f max is the maximum value of the threshold function, f min is the minimum value of the threshold function, e l (v r ,h l θ l ) represents the energy function, θ maxlRBM is the preset control threshold; l RBM Indicates the number of hidden layers, maxl RBM Indicates the maximum number of hidden layers;

[0058] During the training process, the hidden layer neurons h are analyzed j If the condition of increasing the number of neurons in the hidden layer is met, a new neuron is added based on the neuron and the parameters of the two neurons are reset to zero. RBM Whether the conditions for increasing the number of hidden layers are met. If so, copy the hidden layer and keep the original parameters unchanged;

[0059] A3. Based on the above steps, a multi-layer directed factor extraction model is established and optimized, and the training data is input into the model for training. The model is optimized and trained using the validation data set to obtain the final model. The data change characteristics of the sample health status change process are extracted through the model, and the correlation between these characteristic attributes and the associated features of the health status of the electromechanical equipment is screened to determine the preferred associated attribute sequence;

[0060] Based on the aforementioned associated attribute sequence, the existing monitoring data containing each attribute in the associated attribute sequence is determined from the monitoring data of the electromechanical equipment a0 to be analyzed as the data to be analyzed;

[0061] From the existing samples of electromechanical equipment of the same type {a i} selects historical monitoring data covering the corresponding period or stage as initial reference data;

[0062] The data to be analyzed and the initial reference data include the associated attributes s in the associated attribute sequence j Historical monitoring data;

[0063] Select an appropriate data period T based on the data collection period characteristics of the sample to be analyzed and the same type of equipment, T≥max(k j T j );T j Any attribute element s j The sampling period, k j To ensure that the attribute element s j The minimum period for which sample data is valid;

[0064] For sample a i Attribute sequence S = {s1, s2...s j ....s J Each attribute element s in j , using a sliding time window ΔT = T j Intercept the corresponding periodic monitoring data, and intercept the attribute element s based on the sliding time window ΔT j The monitoring data generates a periodic monitoring data curve, and performs similarity analysis with the periodic monitoring data curve of the corresponding attribute element of the sample to be analyzed to determine the similarity of each attribute element s j A time window with the greatest similarity in periodic monitoring data;

[0065] Taking into account the time scalability of attribute element data, the attribute data in different devices at a certain stage may have different duration periods. Therefore, based on the dynamic time warping (DTW) method, the data intercepted by the time window ΔT can be used to form a data curve. Whether the corresponding data have a close life cycle can be determined by whether there is a high temporal correspondence between the data curves, and then the most matching time window can be determined.

[0066] Extract the periodic monitoring data within the time window as valid data and retain it to obtain sample a i Attribute elements of j The closest attribute value sequence X in the corresponding period ij ={x ij1 ,x ij2 ,x ij3 ...x ijn ...x ijNj}, Draw the parameter change curve within the time window ΔT and determine the sample a i Attribute elements of j The periodic rate of change w ij ;

[0067] in

[0068] Similarly, obtain the parameter periodic change rate of the sample a0 to be analyzed

[0069] Calculate historical data to calculate the sample a0 to be analyzed and each existing sample a i The cycle rate deviation where x ijn represents the sampling data of the j-th parameter of sample i in the n-th acquisition cycle; T j For attribute element s j The sampling period of

[0070] According to the principle of increasing the periodic change rate deviation from small to large, the existing sample a i Sorting and selecting several existing samples, forming a valid data set for each sample with corresponding valid data, and organizing each sample to establish a valid reference database containing all samples;

[0071] Combined with the previous step, the slope of the line between two consecutive data at two adjacent time nodes can be used to determine the data change trend within the detection period. The slope matching degree of the product to be analyzed and the sample data can be used to determine the matching degree of the life characteristics of the two. Then, the consistency of the life changes represented by the data of different samples and the product to be analyzed is analyzed, and this is used as a weight to predict the expected life of the product to be analyzed through the actual life of the sample;

[0072] Specifically, based on the data to be analyzed, the sample matrix to be analyzed expressed by a multidimensional vector is obtained. And the sample matrix to be analyzed

[0073]

[0074]

[0075] Among them, since each attribute element s j The total amount of attribute data is different due to different collection cycles. In the actual implementation process, the matrix parameters can be optimized by zero-filling, normalization, etc. to facilitate calculation and processing;

[0076] Based on the previous step sample a i Attribute elements of j The periodic rate of change w ij The calculation method determines the sample period change matrix W to be analyzed j ′ and the sample matrix to be analyzed W ij ;in:

[0077]

[0078] Since the change of equipment life cycle is often accompanied by the continuous change of equipment performance indicators, the attribute data of each time node cannot well reflect the time series change trend of the corresponding attribute characteristics. According to experience, when the life stage of a certain attribute data is closer to the current life stage of the product to be analyzed, the life expectancy of the equipment under the current data is also closer to the life expectancy of the product to be analyzed; without reflecting this influence, a stage matching index ω is established to characterize the degree of matching between the life stages of the sample and the product to be analyzed, 0<ω≤1, and the stage matching index ω characterizes the degree of closeness between the life cycle node of the data and the current life cycle node of the equipment to be analyzed. The closer the stage matching index ω is, the larger the stage matching index ω is, and vice versa. The faster the equipment life cycle changes, the larger the amplitude of the stage matching index ω is, and vice versa.

[0079] Based on the similarity measurement principle, determine the periodic change matrix W of the sample to be analyzed j ′ and the reference sample matrix W ij The similarity can be expressed as For the similarity d(W J ′,W iJ ) is dimensionless, and the reference a considering the life cycle matching is obtained. i The degree of influence of the sample relative to the sample to be analyzed a0

[0080] In order to determine the influence weights of multiple reference samples on the life prediction of the sample to be analyzed, and to ensure that the weight value range is [0,1], a positive real number β is given, and the reference a is obtained. i The weight expression of the sample relative to the sample to be analyzed a0 After normalization, its weight is expressed as Where I is the total number of reference samples;

[0081] Then the predicted life of the sample to be analyzed based on the reference sample can be expressed as:

[0082]

[0083] Among them, RUL(a0|t) represents the predicted remaining life of the sample to be analyzed at time node t, RUL(a i |t) represents the remaining life of the reference sample at time node t.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for predicting the life of electromechanical equipment, characterized in that: The steps include: Stp1, determine the expected life associated attribute sequence S = {s1, s2...s j ....s J }; Based on the aforementioned associated attribute sequence, determine the existing monitoring data containing each attribute in the associated attribute sequence from the monitoring data of the electromechanical equipment a0 to be analyzed as the data to be analyzed; From the existing samples of electromechanical equipment of the same type {a i } selects historical monitoring data covering the corresponding period or stage as initial reference data; Stp2, select the optimal data to establish a database, select samples from existing samples to establish a sample database; select a suitable data cycle T based on the characteristics of the data collection cycle of the sample to be analyzed and the same type of equipment, T≥max(k j T j );T j Any attribute element s j The sampling period, k j To ensure that the attribute element s j The minimum period for which sample data is valid; For sample a i Attribute sequence S = {s1, s2...s j ....s J Each attribute element s in j , using a sliding time window ΔT = T j Intercept the corresponding periodic monitoring data, and intercept the attribute element s based on the sliding time window ΔT j The monitoring data generates a periodic monitoring data curve, and performs similarity analysis with the periodic monitoring data curve of the corresponding attribute element of the sample to be analyzed to determine the similarity of each attribute element s j The time window with the largest similarity of periodic monitoring data is extracted and retained as valid data to obtain sample a. i Attribute elements of j The closest sequence of attribute values ​​within the corresponding period Draw the parameter change curve within the time window ΔT and determine the sample a i Attribute elements of j The periodic rate of change w ij ; in Similarly, obtain the parameter periodic change rate of the sample a0 to be analyzed Calculate historical data to calculate the sample a0 to be analyzed and each existing sample a i The cycle rate deviation where x ijn represents the sampling data of the j-th parameter of sample i in the n-th acquisition cycle; T j For attribute element s j The sampling period of According to the principle of increasing the periodic change rate deviation from small to large, the existing sample a i Sorting and selecting several existing samples, forming a valid data set for each sample with corresponding valid data, and organizing each sample to establish a valid reference database containing all samples; Stp3, establish prediction weights and predictions based on the time series similarity of sample data, including obtaining a sample matrix to be analyzed expressed by a multidimensional vector based on the data to be analyzed And the sample matrix to be analyzed Based on the previous step sample a i Attribute elements of j The periodic rate of change w ij The calculation method determines the sample period change matrix W to be analyzed j ′ and the sample matrix to be analyzed W ij ,in: Since the change of equipment life cycle is often accompanied by the continuous change of equipment performance indicators, the attribute data of each time node cannot well reflect the time series change trend of the corresponding attribute characteristics. According to experience, when the life stage of a certain attribute data is closer to the current life stage of the product to be analyzed, the life expectancy of the equipment under the current data is also closer to the life expectancy of the product to be analyzed; without reflecting this influence, a stage matching index ω is established to characterize the degree of matching between the life stages of the sample and the product to be analyzed, 0<ω≤1, and the stage matching index ω characterizes the degree of closeness between the life cycle node of the data and the current life cycle node of the equipment to be analyzed. The closer the stage matching index ω is, the larger the stage matching index ω is, and vice versa. The faster the equipment life cycle changes, the larger the amplitude of the stage matching index ω is, and vice versa. Based on the similarity measurement principle, determine the periodic change matrix W of the sample to be analyzed j ′ and the reference sample matrix W ij The similarity can be expressed as For the similarity d(W′ J ,W iJ ) is dimensionless, and the reference a considering the life cycle matching is obtained. i The degree of influence of the sample relative to the sample to be analyzed a0 In order to determine the influence weights of multiple reference samples on the life prediction of the sample to be analyzed, and to ensure that the weight value range is [0,1], a positive real number β is given, and the reference a is obtained. i The weight expression of the sample relative to the sample to be analyzed a0 After normalization, its weight is expressed as Where I is the total number of reference samples; Then the predicted life of the sample to be analyzed based on the reference sample can be expressed as: Among them, RUL(a0|t) represents the expected remaining life of the sample a0 to be analyzed at time node t, RUL(a i |t) represents the reference sample a at time node t i remaining life.

2. The method for predicting the life of electromechanical equipment according to claim 1, characterized in that: The data to be analyzed and the initial reference data include the associated attributes s in the associated attribute sequence j The original monitoring data shall include multiple groups of original monitoring data under different working conditions of the normal operation of the electromechanical equipment.

3. The method for predicting the life of electromechanical equipment according to claim 1, characterized in that: In step Stp3, since each attribute element s j The total amount of attribute data is different due to different collection cycles. In the actual implementation process, the matrix parameters are optimized through zero-value filling and normalization processing to facilitate calculation and processing.

4. The method for predicting the life of electromechanical equipment according to claim 1, characterized in that: In the step Stp1, in order to ensure the validity of each attribute element in the associated attribute sequence S and to ensure that the attribute element at the screening position can characterize the expected life of the electromechanical equipment, each element in the associated attribute is determined based on the following method: A1. First, obtain the original monitoring data and perform necessary preprocessing, including deleting and revising erroneous data and normalizing the data. The original monitoring data should include multiple sets of original monitoring data under different working conditions of the normal operation of the electromechanical equipment; After necessary preprocessing, the original data is segmented to obtain the training data set and the verification data set; A2. Establish a multi-layer directed factor extraction model consisting of multiple restricted Boltzmann machine network models (RBN), wherein the restricted Boltzmann machine network model includes a visible layer v and a hidden layer h, the visible layer is used as the input side to obtain sample data, and the hidden layer is used as the output end to extract the main features of the sample and output it in the form of probability distribution; In order to optimize the data feature extraction effect, enhance the integrity of the data and avoid falling into the local optimum, the global weight change D is used j (m) and local weight changes control the weight changes of hidden layer neurons, and the algorithm is optimized by increasing the number of hidden layers and hidden layer neurons. Specifically: The condition for increasing the number of hidden layer neurons can be expressed as: The condition for increasing the number of hidden layers can be expressed as: Where D j (m)=||d j (m),d j (m-1)||2; d j (m) is the number of neurons in hidden layer h after m iterations j The weight vector of refers to the neuron h j The global weight change D for the nth training sample j (m); R is the total number of samples, and R′ is the change in global weight during the last iteration D j (m) increases the number of samples; where f(m) is the maximum global weight change D that ensures that each hidden layer neuron can correspond to a training sample after the number of hidden layer neurons increases j The threshold function of (m), f(m) can be expressed as: m max is the maximum number of iterations, ρ m Represents the slope of the variable threshold function curve at the corresponding position, f max is the maximum value of the threshold function, f min is the minimum value of the threshold function, e l (v r ,h l θ l ) represents the energy function, is the preset control threshold; l RBM Indicates the number of hidden layers, maxl RBM Indicates the maximum number of hidden layers; During the training process, the hidden layer neurons h are analyzed j If the condition of increasing the number of neurons in the hidden layer is met, a new neuron is added based on the neuron and the parameters of the two neurons are reset to zero. RBM Whether the conditions for increasing the number of hidden layers are met. If so, copy the hidden layer and keep the original parameters unchanged; A3. Based on the above steps, a multi-layer directed factor extraction model is established and optimized, and the training data is input into the model for training. The model is optimized and trained using the validation data set to obtain the final model. The data change characteristics of the sample health status change process are extracted through the model, and the correlation between these characteristic attributes and the associated characteristics of the health status of electromechanical equipment is used to screen and determine the preferred associated attribute sequence.

Images