Elevator safety risk self-adaptive analysis method

Abstract

The invention discloses an elevator safety risk self-adaptive analysis method. The method comprises the steps that a fault prediction model is obtained; obtaining an error sample set based on the sample verification set, and calculating similarity loss based on an error sample set fault prediction result and a real fault type of each sample in the error sample set; constructing a first target loss function based on cross entropy loss, constructing a second target loss function based on similarity loss, and optimizing the fault prediction model to obtain a first fault prediction optimization model and a second fault prediction optimization model; real-time operation parameters are collected, the minimum similarity between the real-time operation parameters and the error samples is confirmed, when the minimum similarity is larger than a preset similarity threshold value, the second fault prediction optimization model is adopted for prediction, otherwise, the first fault prediction optimization model is adopted for prediction, and the elevator risk level is calculated according to the prediction result. The elevator can be monitored in real time, and elevator abnormity and potential risks can be recognized in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of elevator safety detection, and in particular relates to an elevator safety risk adaptive analysis method. Background Art

[0002] With the rapid growth of the elevator market in the consumer sector, the diagnosis of elevator systems is a major challenge. Existing elevator safety maintenance is usually carried out by staff regularly inspecting or installing surveillance cameras to monitor the operating status of the elevator in real time. When an abnormal situation occurs, the maintenance personnel are immediately notified for processing. The traditional elevator diagnosis strategy focuses on fault detection or potential fault prediction.

[0003] With the emergence of artificial intelligence and big data technology, the use of sensors to collect various elevator operation data and the emergence of elevator fault diagnosis models based on deep learning models. However, due to the complexity of model input parameters and the large amount of data, how to achieve more accurate fault diagnosis and analysis is the focus of various studies. In addition, data processing, such as incorrect setting of loss functions, will also affect the training process and prediction results of the model. At the same time, the cross entropy loss commonly used in classification problems is prone to overfitting and poor calibration effects. These have posed certain threats to the safety monitoring of elevators, so the need to achieve adaptive detection of elevator operation conditions is becoming more and more urgent. Summary of the invention

[0004] In view of the above problems, the present invention proposes an elevator safety risk adaptive analysis method. To solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0005] An elevator safety risk adaptive analysis method comprises the following steps:

[0006] S1, collect the historical operating parameters of the elevator and the corresponding fault types, build a sample training set and a sample verification set, and obtain a fault prediction model based on the sample training set;

[0007] S2, inputting the sample verification set obtained in step S1 into the fault prediction model to obtain an error sample set, and calculating the similarity loss based on the fault prediction result of the error sample set and the actual fault type of each sample in the error sample set;

[0008] S3, constructing a first target loss function based on the cross entropy loss, constructing a second target loss function based on the similarity loss obtained in step S2, and optimizing the fault prediction model based on the first target loss function and the second target loss function to obtain a first fault prediction optimization model and a second fault prediction optimization model;

[0009] S4, collect real-time operating parameters, confirm the minimum similarity between the real-time operating parameters and the samples in the error sample set, and use the second fault prediction optimization model for prediction when the minimum similarity is greater than the preset similarity threshold, otherwise use the first fault prediction optimization model for prediction, and calculate the elevator risk level based on the prediction results.

[0010] The step S2 comprises the following:

[0011] S2.1, according to the prediction results of each sample in the error sample set, the samples of the same fault type are combined to form corresponding clusters, and the cluster center of each cluster is calculated according to the samples;

[0012] S2.2, using a similarity calculation method to calculate the similarity between each sample in the error sample set and the corresponding cluster center obtained in step S2.1, and selecting the similarity as the first similarity of the sample;

[0013] S2.3, confirm the cluster to which the sample belongs based on the actual fault type of each sample in the error sample set, use the similarity calculation method to calculate the similarity between the cluster center and the sample as the second similarity of the corresponding sample, and calculate the similarity loss based on the first similarity and the second similarity.

[0014] The formula corresponding to the second objective loss function is:

[0015]

[0016] Where, L CE represents the standard cross entropy loss, represents the true fault type label corresponding to the lth sample in the error sample set, y l represents the prediction result of the lth sample in the error sample set, ξ represents the polynomial coefficient, α is the hyperparameter, L represents the number of samples in the error sample set, and f(l) represents the similarity loss of the lth sample in the error sample set;

[0017] The formula corresponding to the first objective loss function is:

[0018]

[0019] In the formula, Indicates the true fault type label corresponding to the i-th sample in the sample training set, y i It represents the prediction result of the i-th sample in the sample training set, and I represents the number of samples in the sample training set.

[0020] The calculation formula of the elevator risk level is:

[0021]

[0022] In the formula, θ represents the elevator risk index factor, represents the elevator risk index of the corresponding fault type output by the fault prediction optimization model, y represents the fault probability output by the fault prediction optimization model, a, b and c are all risk factors, Represents the equipment maintenance cost rate.

[0023] The elevator risk index The calculation formula is:

[0024]

[0025] In the formula, represents the total running time of the elevator, and t represents the failure time of the corresponding fault type output by the fault prediction optimization model during the running time of the elevator.

[0026] The fault prediction model includes a normalization module, an MLP perceptron, a fully connected layer and a Softmax function layer which are connected in sequence.

[0027] In step S4, before using the second fault prediction optimization model for prediction, the error influence of each operating parameter is first calculated based on the error sample set, and the MLP perceptron updates the weight of the first hidden layer of the MLP perceptron based on the obtained error influence, and then prediction is performed based on the updated second fault prediction optimization model;

[0028] The expression of the updated weight of the first hidden layer is:

[0029] w mv =λw′ mv ;

[0030] In the formula, λ represents the adjustment coefficient, w′ mv represents the initial weight between the mth operating parameter of the input data and the vth neuron of the first hidden layer, v represents the number of neurons in the first hidden layer, and w mv Represents the updated weight between the mth operating parameter of the input data and the vth neuron of the first hidden layer.

[0031] , the calculation formula of the adjustment coefficient λ is:

[0032] Υ m =λw′ m1 +λw′ m2 +…λw′ mv ;

[0033] Where w′ m1 Represents the initial weight between the mth operating parameter of the input data and the first neuron of the first hidden layer, Υ mIt represents the error influence of the mth operating parameter on the fault prediction result.

[0034] The step of calculating the error influence of each operating parameter based on the error sample set comprises the following steps:

[0035] ①, increasing or decreasing each operating parameter in each sample of the error sample set to generate a corresponding number of first impact assessment samples and second impact assessment samples;

[0036] ②, inputting the first impact assessment sample and the second impact assessment sample obtained in step ① into the fault prediction model to obtain the corresponding fault prediction result, and calculating the impact value according to the obtained fault prediction result;

[0037] ③, calculate the error impact according to the fault prediction results;

[0038] The calculation formula of the error influence degree is:

[0039]

[0040] In the formula, Υ m represents the error influence of the mth operating parameter on the fault prediction result, γ m represents the influence value of the mth operating parameter on the fault prediction result, and M represents the number of operating parameters of each sample in the sample training set and the sample verification set;

[0041] The impact value γ m The calculation formula is:

[0042]

[0043] In the formula, It means that the second impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result, It indicates that the first impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result.

[0044] Beneficial effects of the present invention:

[0045] 1. The construction of an LSTM model based on the attention mechanism can extract more accurate features of the elevator's historical operating parameters. Different loss functions are constructed based on different samples. After analyzing the collected real-time data, different fault prediction optimization models are adapted to predict elevator faults, which improves the efficiency of fault prediction. It can monitor the elevator operation in real time, identify elevator anomalies and potential risks in a timely manner, and effectively reduce the occurrence of elevator accidents.

[0046] 2. The loss function is designed to reduce the difference between the predicted results and the actual results, so that the final model can fit the data more accurately, effectively improve the classification accuracy of the samples, reduce the misclassification problem, and improve the prediction accuracy, robustness and generalization ability of the model;

[0047] 3. Construct an elevator risk level calculation function, which can quantitatively evaluate the elevator risk based on the elevator risk index, equipment maintenance cost rate, and elevator fault type, making it easier for maintenance personnel to deal with the elevator according to the elevator risk level;

[0048] 4. Based on each error sample in the error sample set, an adaptive increase and decrease algorithm is used to generate an impact assessment sample, and the error impact of each operating parameter is calculated based on the impact assessment sample. The second fault prediction optimization model is updated based on the error impact, so that the model pays more attention to the operating parameters that affect the prediction results, which greatly improves the model prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 It is a schematic diagram of the process of the present invention.

[0051] Figure 2 Schematic diagram of the network structure of the fault prediction model.

[0052] Figure 3 It is a schematic diagram of the module structure of the present invention.

[0053] Figure 4 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] An adaptive analysis method for elevator safety risks, such as Figure 1 As shown, the following steps are included:

[0056] S1, collect the historical operating parameters of the elevator and the corresponding fault types, build a sample training set and a sample verification set, and obtain a fault prediction model based on the sample training set;

[0057] The historical operating parameters include traction motor current, brake motor current, car vibration signal and acceleration signal, and may also include inclination angle, car running speed, etc. The measurement of the above types of signals can be obtained by corresponding sensors. The traction motor current measurement is carried out by a current sensor, which is installed on the power line of the elevator system to measure the flow of current. The brake motor current is the same. The car vibration signal is measured by a single-axis acceleration sensor, which is installed on each wall of the car, the car door and the bottom of the car. The acceleration signal is measured by a three-axis acceleration sensor, which is placed in the center of the vertical elevator car floor to measure the car acceleration signal. The inclination angle can be measured by a tilt sensor, which can be installed on the side wall of the car.

[0058] The fault types include: traction system, guide system, car, electrical control system, electric traction system. Specifically, the fault types can be further subdivided according to various fault types. For example, the traction system may include brake failure, traction machine winding shell grounding short circuit, traction machine brake coil iron core suction abnormality, etc. Each fault type is encoded, and different fault types are represented by different coding numbers.

[0059] The sample training set and sample verification set are respectively train and S verify Indicates that S train ={X1,X2,…,X i ,…,X I}, X verify ={s1,x2,…,x h ,…,x H}, X i represents the i-th sample data set in the sample training set, which is obtained by combining the historical operating parameters at the corresponding time. h represents the hth sample data set in the sample validation set, I represents the number of samples in the sample training set, H represents the number of samples in the sample validation set, the label of each sample is its corresponding fault type, and K represents the number of fault types of all samples in the sample training set and the sample validation set.

[0060] A fault prediction model is obtained based on the sample training set. Specifically, the fault prediction model includes a normalization module, an MLP perceptron, a fully connected layer, and a Softmax function layer connected in sequence. The normalization module is used to normalize the historical operating parameters of each sample in the sample training set and the sample verification set. Specifically, the minimum-maximum normalization method, Z-score normalization, etc. can be used based on the data at all sampling times. This is the prior art and will not be described in detail in this embodiment. MLP is also known as a multilayer perceptron. In addition to an input layer and an output layer, there may be multiple hidden layers in the middle. It is a multi-layer neuron structure with full connections between layers. Data is received through the input layer, and the hidden layer performs feature extraction and transformation. MLP can handle nonlinear problems and can be widely used in classification tasks such as machine learning and deep learning. The model finally converts the output of the neural network into a probability distribution through a multi-classification Softmax function, ensuring the normalization of the output. In this embodiment, the input layer X∈R I×M , M represents the number of operating parameters for each sample in the sample training set and sample validation set.

[0061] S2, inputting the sample verification set obtained in step S1 into the fault prediction model to obtain an error sample set, and calculating the similarity loss based on the fault prediction result of the error sample set and the actual fault type of each sample in the error sample set;

[0062] The error sample set refers to the set of samples whose output results obtained after inputting the sample verification set into the fault prediction model for fault prediction are inconsistent with the sample labels in the sample verification set, that is, the actual fault types. error indicates that, and s error ∈S verify For example, when the label of a sample in the sample validation set is a traction system, and the prediction result obtained after inputting the sample into the fault prediction model is not a traction system, the sample is added to the error sample set, otherwise it is not added.

[0063] The similarity loss calculation based on the fault prediction result of the error sample set and the actual fault type of each sample in the error sample set comprises the following steps:

[0064] S2.1, according to the prediction results of each sample in the error sample set, the samples of the same fault type are aggregated to form corresponding clusters, and the cluster center is calculated according to the vector corresponding to each sample;

[0065] Specifically, according to the prediction results of all samples in the error sample set, the prediction result, that is, the fault type with the highest prediction probability, is regarded as the predicted fault type of each sample, and the samples with the same predicted fault type are combined to form the corresponding cluster. The cluster center is the center point of each cluster, which is obtained by calculating the mean value of the vectors corresponding to all data points in the cluster. The fault type corresponding to the cluster center of each cluster is consistent with the predicted fault type of all samples in the cluster, and this fault type is the fault type corresponding to the cluster.

[0066] S2.2, using the similarity calculation method to calculate the similarity between each sample in the error sample set and the corresponding cluster center obtained in step S2.1, and select the similarity as the first similarity of the sample. The corresponding calculation formula is:

[0067]

[0068] In the formula, Represents the error sample set s error The first similarity of the lth sample in s error,l Represents the error sample set s error The vector corresponding to the lth sample in , sim(·) represents the similarity function. The similarity calculation method can be cosine similarity, Euclidean distance, Pearson correlation coefficient, etc. Based on the vector corresponding to the sample and the vector corresponding to the cluster center, any method can be selected for calculation. C k represents the cluster center of the kth cluster, and the fault type and error sample set s of the kth cluster error The fault prediction result of the lth sample in is consistent with the predicted fault type.

[0069] S2.3, based on the real fault type of each sample in the error sample set, confirm the cluster to which the sample belongs, use the similarity calculation method to calculate the similarity between the cluster center of the cluster and the sample as the second similarity of the sample, and calculate the similarity loss based on the first similarity and the second similarity. The corresponding calculation formula is:

[0070]

[0071] Where f(l) represents the error sample set s error The similarity loss of the lth sample in C k′ Represents the cluster center of the k′th cluster, k′≠k.

[0072] Specifically, according to the real fault types of all samples in the error sample set, samples with the same real fault type are combined to form corresponding clusters. The cluster center is also the center point of each cluster, which is obtained by calculating the mean value of the vector corresponding to all data points in the cluster. The first similarity value reflects the difference between the operating parameters of each sample in the error sample set and each sample under the same predicted fault type corresponding to the fault prediction result. The smaller the first similarity value, the greater the difference between them. The second similarity value reflects the difference between the operating parameters of each sample in the error sample set and each sample under the same real fault type. The smaller the second similarity value, the greater the difference between them. The similarity loss reflects the difference in sample data between the predicted result of the fault prediction model and the real result.

[0073] S3, constructing a first target loss function based on the cross entropy loss, constructing a second target loss function based on the similarity loss obtained in step S2, and optimizing the fault prediction model based on the first target loss function and the second target loss function to obtain a first fault prediction optimization model and a second fault prediction optimization model;

[0074] The formula corresponding to the second objective loss function is:

[0075]

[0076] Where, L CE represents the standard cross entropy loss, represents the true fault type label corresponding to the lth sample in the error sample set, y l represents the prediction result of the lth sample in the error sample set, ξ represents the polynomial coefficient, α is the hyperparameter, and L represents the number of samples in the error sample set. The loss function is determined by cross entropy loss, sample classification similarity loss, and mean square error loss, and the loss function minimization is used as the optimization goal to train the fault prediction model.

[0077] Cross entropy loss aims to minimize the difference between the predicted results and the true results, so that the model can fit the training data more accurately. Sample classification similarity loss aims to improve the classification tightness and accuracy of samples and reduce misclassification caused by category boundary problems. Mean square error loss evaluates the prediction accuracy of the model by calculating the average of the square of the difference between the predicted value and the true value, improving the robustness and generalization ability of the model.

[0078] The formula corresponding to the first objective loss function is:

[0079]

[0080] In the formula, Indicates the true fault type label corresponding to the i-th sample in the sample training set, yi Representation sample prediction results.

[0081] In the backward propagation process of training, the total loss is first calculated by formula (3) or formula (4), and then the gradient of the loss function is updated by calculating the network parameters through the back propagation algorithm to minimize the loss. When the loss function converges, the fault prediction model training is completed and the fault prediction optimization model is obtained.

[0082] S4, collecting real-time operating parameters, confirming the minimum similarity between the real-time operating parameters and the samples in the error sample set, and using the second fault prediction optimization model for prediction when the minimum similarity is greater than a preset similarity threshold, otherwise using the first fault prediction optimization model for prediction, and calculating the elevator risk level according to the prediction results;

[0083] The calculation formula corresponding to the elevator risk level is:

[0084]

[0085] In the formula, θ represents the elevator risk index factor, represents the elevator risk index of the corresponding fault type output by the fault prediction optimization model, y represents the fault probability output by the fault prediction optimization model, a, b and c are all risk factors, Represents the equipment maintenance cost rate.

[0086] The elevator risk index The calculation formula is:

[0087]

[0088] In the formula, represents the total running time of the elevator, and t represents the failure time of the corresponding fault type output by the fault prediction optimization model during the running time of the elevator.

[0089] The equipment maintenance cost rate It is obtained by dividing the equipment maintenance cost of the corresponding fault type by the original value of the equipment, that is, the purchase cost.

[0090] When confirming the minimum similarity between the real-time operating parameters and the samples in the error sample set, in order to simplify the calculation method, the K-means clustering method can also be used to cluster the samples in the error sample set. Specifically, first initialize Random Selection Data points are used as the initial cluster centers, and the distance between each sample and each cluster center is calculated. According to the calculated distance results, it is assigned to the nearest cluster center; according to the assignment results, the center point of each cluster is recalculated, which generally refers to the mean of all data points in the cluster. Until the cluster center no longer changes. This is the prior art, and the present application embodiment will not be elaborated in detail. After the clustering is completed, the similarity between the real-time operating parameters and each cluster center can be directly calculated.

[0091] Specifically, an over-threshold alarm reminder can be issued according to the preset elevator risk threshold, and the predicted probability of the fault type output by the fault prediction model can be displayed separately on the elevator monitoring terminal, so that the monitoring personnel can conduct targeted detection before the fault occurs and determine the cause of the fault as soon as possible after the fault occurs.

[0092] Preferably, in order to further improve the prediction results of the second fault prediction optimization model, before using the second fault prediction optimization model for prediction, the error influence of each operating parameter can be first calculated based on the error sample set, and the MLP perceptron updates the weight of the first hidden layer based on the obtained error influence, and then the prediction is made based on the updated second fault prediction optimization model. Since each neuron in the input layer of the MLP is fully connected to each neuron in the adjacent hidden layer, the weight of the first hidden layer receiving the input data is updated based on the error influence to replace the initial weight of the model. The corresponding calculation formula is:

[0093] Υ m =λw′ m1 +λw′ m2 +…λw′ mv ; (7)

[0094] In the formula, λ represents the adjustment coefficient, w′ m1 Represents the initial weight between the mth running parameter of the input data and the first neuron of the first hidden layer, w′ mv represents the initial weight between the mth operating parameter of the input data and the vth neuron of the first hidden layer, and v represents the number of neurons in the first hidden layer. According to formula (7), λ can be calculated, and then the updated weight of the vth neuron of the first hidden layer can be calculated. mv The corresponding expression is:

[0095] w mv =λw′ mv (8)

[0096] After inputting the data set consisting of real-time operation parameters, the expression of the output of the vth neuron in the first hidden layer is:

[0097]

[0098] Where b v represents the bias of the vth neuron, X′ mrepresents the mth operating parameter in the real-time operating parameter dataset, and f(·) represents the activation function.

[0099] The step of calculating the error influence of each operating parameter based on the error sample set comprises the following steps:

[0100] ①, increasing or decreasing each operating parameter in each sample of the error sample set to generate a corresponding number of first impact assessment samples and second impact assessment samples;

[0101] The expression of the first impact assessment sample set is:

[0102]

[0103] In the formula, represents the first impact assessment sample corresponding to the nth sample in the error sample set, which is obtained by multiplying the mth operating parameter of the nth sample in the error sample set by 1-θ. Specifically, at least the mth operating parameter of at least one sampling moment in the sample is multiplied by 1-θ, where θ represents the percentage change, and x n,m Represents the mth operating parameter in the nth sample in the error sample set.

[0104] The expression of the second impact assessment sample set is:

[0105]

[0106] In the formula, The second impact assessment sample set corresponding to the nth sample in the error sample set is represented by multiplying the mth operating parameter of the nth sample in the error sample set by 1+θ. Specifically, at least the mth operating parameter of at least one sampling moment in the sample is multiplied by 1+θ.

[0107] ②, inputting the first impact assessment sample and the second impact assessment sample obtained in step ① into the fault prediction model to obtain the corresponding fault prediction result, and calculating the impact value according to the obtained fault prediction result;

[0108] The impact value γ m The calculation formula is:

[0109]

[0110] In the formula, It means that the second impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result, Indicates that the first impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result, γ m Represents the impact value of the mth operating parameter on the fault prediction result.

[0111] ③, calculate the error impact according to the fault prediction results;

[0112] The calculation formula of the error influence degree is:

[0113]

[0114] In the formula, Υ m It represents the error influence of the m-th operating parameter on the fault prediction result. It reflects the error influence of the fault prediction model on the fault prediction result when the m-th operating parameter changes. The larger the value, the greater the influence of the operating parameter change on the fault prediction result. Conversely, the smaller the value.

[0115] The present application also provides an elevator safety risk adaptive analysis device, such as Figure 3 As shown, including:

[0116] Data acquisition module: used to collect the historical operating parameters of the elevator and the corresponding real fault types, and build sample training sets and sample verification sets;

[0117] Model building module: used to obtain fault prediction model based on sample training set;

[0118] Model optimization module: used to obtain an error sample set according to the sample verification set obtained by the data acquisition module and the fault prediction model constructed by the model construction module, calculate the similarity loss based on the fault prediction result of the error sample set and the actual fault type of each sample in the error sample set, construct a first target loss function based on the cross entropy loss, construct a second target loss function based on the similarity loss, and optimize the fault prediction model based on the first target loss function and the second target loss function to obtain a first fault prediction optimization model and a second fault prediction optimization model;

[0119] Risk calculation module: used to collect real-time operating parameters, confirm the minimum similarity between the real-time operating parameters and each sample in the error sample set, and use the second fault prediction optimization model for prediction when it is greater than the preset similarity threshold; otherwise, use the first fault prediction optimization model for prediction, and calculate the elevator risk level based on the prediction results.

[0120] The optimization method of this embodiment may also include a parameter impact optimization module: used to calculate the error impact of each operating parameter on the model prediction result based on the error sample set, and update the second fault prediction optimization model of the model optimization module based on the error impact.

[0121] The present application also provides an electronic device, such as Figure 4As shown, the device may include: at least one processor and a memory, the memory being used to store a program. Specifically, the program may include a program code, and the program code includes computer operating instructions. The memory may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. The processor is used to execute the computer execution instructions stored in the memory to implement the method described in the embodiment of the above-mentioned elevator safety risk adaptive analysis method. Among them, the processor may be a central processing unit (Central Processing Unit, referred to as CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, referred to as ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0122] An embodiment of the present invention further provides a computer-readable storage medium, which may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media that can store computer programs. Specifically, a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, it is used to implement the steps of any one of the elevator safety risk adaptive analysis methods in the above embodiments. The specific implementation method and technical effect are similar and will not be repeated here.

[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An elevator safety risk adaptive analysis method, characterized in that: The steps include: S1, collect the historical operating parameters of the elevator and the corresponding fault types, build a sample training set and a sample verification set, and obtain a fault prediction model based on the sample training set; S2, inputting the sample verification set obtained in step S1 into the fault prediction model to obtain an error sample set, and calculating the similarity loss based on the fault prediction result of the error sample set and the actual fault type of each sample in the error sample set; S3, constructing a first target loss function based on the cross entropy loss, constructing a second target loss function based on the similarity loss obtained in step S2, and optimizing the fault prediction model based on the first target loss function and the second target loss function to obtain a first fault prediction optimization model and a second fault prediction optimization model; S4, collect real-time operating parameters, confirm the minimum similarity between the real-time operating parameters and the samples in the error sample set, and use the second fault prediction optimization model for prediction when the minimum similarity is greater than the preset similarity threshold, otherwise use the first fault prediction optimization model for prediction, and calculate the elevator risk level based on the prediction results.

2. The elevator safety risk adaptive analysis method according to claim 1 is characterized in that: The step S2 comprises the following: S2.1, according to the prediction results of each sample in the error sample set, the samples of the same fault type are combined to form corresponding clusters, and the cluster center of each cluster is calculated according to the samples; S2.2, using a similarity calculation method to calculate the similarity between each sample in the error sample set and the corresponding cluster center obtained in step S2.1, and selecting the similarity as the first similarity of the sample; S2.3, confirm the cluster to which the sample belongs based on the actual fault type of each sample in the error sample set, use the similarity calculation method to calculate the similarity between the cluster center and the sample as the second similarity of the corresponding sample, and calculate the similarity loss based on the first similarity and the second similarity.

3. The elevator safety risk adaptive analysis method according to claim 1 is characterized in that: The formula corresponding to the second objective loss function is: Where, L CE represents the standard cross entropy loss, represents the true fault type corresponding to the lth sample in the error sample set, y l represents the prediction result of the lth sample in the error sample set, ξ represents the polynomial coefficient, α is the hyperparameter, L represents the number of samples in the error sample set, and f(l) represents the similarity loss of the lth sample in the error sample set; The formula corresponding to the first objective loss function is: In the formula, Indicates the true fault type label corresponding to the i-th sample in the sample training set, y i It represents the prediction result of the i-th sample in the sample training set, and I represents the number of samples in the sample training set.

4. The elevator safety risk adaptive analysis method according to claim 1, characterized in that: The calculation formula of the elevator risk level is: In the formula, θ represents the elevator risk index factor, represents the elevator risk index of the corresponding fault type output by the fault prediction optimization model, y represents the fault probability output by the fault prediction optimization model, a, b and c are all risk factors, Represents the equipment maintenance cost rate.

5. The elevator safety risk adaptive analysis method according to claim 1, characterized in that: The elevator risk index The calculation formula is: In the formula, represents the total running time of the elevator, and t represents the failure time of the corresponding fault type output by the fault prediction optimization model during the running time of the elevator.

6. The elevator safety risk adaptive analysis method according to claim 1, characterized in that: The fault prediction model includes a normalization module, an MLP perceptron, a fully connected layer and a Softmax function layer which are connected in sequence.

7. The elevator safety risk adaptive analysis method according to claim 6 is characterized in that: In step S4, before using the second fault prediction optimization model for prediction, the error influence of each operating parameter is first calculated based on the error sample set, and the MLP perceptron updates the weight of the first hidden layer of the MLP perceptron based on the obtained error influence, and then prediction is performed based on the updated second fault prediction optimization model; The expression of the updated weight of the first hidden layer is: In mv =λw′ mv ; In the formula, λ represents the adjustment coefficient, w ′ mv represents the initial weight between the mth operating parameter of the input data and the vth neuron of the first hidden layer, v represents the number of neurons in the first hidden layer, and w mv Represents the updated weight between the mth operating parameter of the input data and the vth neuron of the first hidden layer.

8. The elevator safety risk adaptive analysis method according to claim 7, characterized in that: The calculation formula of the adjustment coefficient λ is: Υ m =λw′ m1 +λw′ m2 +…λw′ mv ; In the formula, w ′ m1 Represents the initial weight between the mth operating parameter of the input data and the first neuron of the first hidden layer, Υ m It represents the error influence of the mth operating parameter on the fault prediction result.

9. The elevator safety risk adaptive analysis method according to claim 7, characterized in that: The step of calculating the error influence of each operating parameter based on the error sample set comprises the following steps: ①, increasing or decreasing each operating parameter in each sample of the error sample set to generate a corresponding number of first impact assessment samples and second impact assessment samples; ②, inputting the first impact assessment sample and the second impact assessment sample obtained in step ① into the fault prediction model to obtain the corresponding fault prediction result, and calculating the impact value according to the obtained fault prediction result; ③, calculate the error impact according to the fault prediction results; The calculation formula of the error influence degree is: In the formula, Υ m represents the error influence of the mth operating parameter on the fault prediction result, γ m represents the influence value of the mth operating parameter on the fault prediction result, and M represents the number of operating parameters of each sample in the sample training set and the sample verification set; The impact value γ m The calculation formula is: In the formula, It means that the second impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result, It indicates that the first impact assessment sample corresponding to the mth operating parameter is input into the fault prediction model to obtain the fault prediction result.

Images