Distributed Internet of Things operation and maintenance terminal fault intelligent prediction method and system

Abstract

The invention relates to the technical field of Internet of Things and artificial intelligence, in particular to a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method and system, and the method comprises the steps: collecting historical fault data and real-time operation data of distributed Internet of Things terminal equipment, and processing the data to form a standardized data set; constructing a fault mode recognition model according to the convolutional neural network, training the standardized data set, and calculating to obtain a pooling layer output feature map and full connection layer output; clustering the fault data by adopting a K-means clustering algorithm, and identifying a high-frequency fault area and a high-frequency fault type; based on the identified high-frequency fault area and the high-frequency fault type, calculating a hidden state by adopting a long-short-term memory network in a time sequence prediction algorithm; and predicting the fault according to the calculation result of the hidden state of the LSTM model, generating operation and maintenance decision suggestions including fault early warning, an equipment maintenance plan and a resource allocation scheme, and pushing the prediction result and the decision suggestions to an operation and maintenance management terminal.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things and artificial intelligence technology, and specifically to a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method and system. Background Art

[0002] With the rapid development of IoT technology, IoT devices are increasingly used in the fields of industry, agriculture, medical treatment, transportation, etc. With the help of sensors, communication modules and data processing units, these devices have achieved real-time monitoring and intelligent control of the physical world, greatly promoting the intelligentization process of various industries. However, with the surge in the number of IoT devices and the increase in system complexity, its operation and maintenance management faces many challenges:

[0003] ‌Equipment failures are frequent and difficult to predict‌. Due to the complexity of the equipment operating environment and the limitations of the equipment itself, failures occur frequently. Traditional threshold or rule-based fault detection methods are difficult to accurately predict failures in complex environments, making fault prediction difficult. In addition, manual inspections and post-maintenance methods are not only inefficient, but also costly to operate and maintain.

[0004] ‌Existing technologies have limitations‌. Currently, the fault prediction and operation and maintenance management of IoT devices mainly rely on rule-based fault detection, statistical analysis, and machine learning. However, these methods all have certain shortcomings. Rule-based fault detection methods lack flexibility and accuracy; statistical analysis methods rely on the quality and quantity of data and have difficulty capturing nonlinear relationships; and although machine learning methods have achieved certain results, they still face problems such as data imbalance, difficulty in feature selection, and lack of real-time performance‌. Summary of the invention

[0005] (1) Technical issues to be resolved

[0006] The purpose of the present invention is to provide a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method and system to solve the problem of operation and maintenance terminal fault prediction.

[0007] (2) Technical solution

[0008] To achieve the above objectives, on the one hand, the present invention provides a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method, the method comprising:

[0009] Collect historical fault data and real-time operation data of distributed IoT terminal devices, the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; pre-process the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set;

[0010] A fault pattern recognition model is constructed based on a convolutional neural network, and the standardized data set is trained to calculate the output feature map of the pooling layer and the output of the fully connected layer. The K-means clustering algorithm is used to cluster the fault data to identify high-frequency fault areas and high-incidence fault types.

[0011] Combined with real-time operation data, based on the identified high-frequency fault areas and high-incidence fault types, the long short-term memory network in the time series prediction algorithm is used to calculate the hidden state; according to the calculation results of the hidden state of the long short-term memory network model, the fault is predicted, and operation and maintenance decision suggestions are generated, including fault warning, equipment maintenance plan, resource allocation plan, and the prediction results and decision suggestions are pushed to the operation and maintenance management terminal.

[0012] Furthermore, the method of preprocessing the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set includes:

[0013] S21, data cleaning: remove duplicate data and process outliers, wherein the outliers include data points that are beyond a reasonable range and mark the outliers as missing values;

[0014] S22, missing value filling: fill missing values ​​using mean or interpolation;

[0015] S23, data normalization: scale the data to the range of [0,1], the normalization formula is as follows:

[0016] ;

[0017] in, represents the normalized data value, Represents the original data, and Respectively represent the minimum and maximum values ​​of the original data;

[0018] S24, forming a standardized data set, integrating the cleaned, filled and normalized data into a unified standardized data set .

[0019] Furthermore, the method of constructing a fault mode recognition model according to a convolutional neural network, training a standardized data set, and calculating a pooling layer output feature map and a fully connected layer output includes:

[0020] Input layer: receives standardized data set As input;

[0021] Convolution layer: extract local features through convolution kernels. The calculation formula is as follows:

[0022] ;

[0023] in, represents the output feature map of the convolutional layer, represents the convolution kernel weight, Represents input data, represents the bias term;

[0024] Activation layer: using Activation function, the formula is as follows:

[0025] ;

[0026] Pooling layer output : Use the maximum pooling operation, the formula is as follows:

[0027] ;

[0028] Among them, s represents the pooling window size;

[0029] Fully connected layer output : Flatten the output of the pooling layer and connect it to the fully connected layer. The calculation formula is as follows:

[0030] ;

[0031] in, represents the weight matrix, represents the bias vector, Represents the activation function.

[0032] Furthermore, the method of clustering fault data using the K-means clustering algorithm to identify high-frequency fault areas and high-incidence fault types includes:

[0033] S41, select data and randomly initialize k cluster centers;

[0034] S42, assign each data point to the nearest cluster center, and the cluster assignment calculation formula is as follows:

[0035] ;

[0036] in, represents a data point, Indicates Cluster centers;

[0037] S43, update the cluster center to the mean of the current cluster, and the calculation formula is as follows:

[0038] ;

[0039] in, Indicates A set of data points in a cluster;

[0040] S44, repeat steps S42 and S43 until the cluster center no longer changes or the maximum number of iterations is reached.

[0041] Furthermore, the method of calculating the hidden state by using the long short-term memory network in the time series prediction algorithm based on the identified high-frequency fault areas and high-incidence fault types in combination with the real-time operation data includes:

[0042] S51, calculate the forget gate , the formula is:

[0043] ;

[0044] in, is the activation function, is the forget gate weight matrix, is the forget gate bias term, is the hidden state, For input data;

[0045] S52, calculation input gate , the formula is:

[0046] ;

[0047] in, is the input gate weight matrix, is the input gate bias term;

[0048] S53, calculate candidate memory units , the formula is:

[0049] ;

[0050] in, is the candidate memory unit weight matrix, is the candidate memory unit bias item;

[0051] S54, computational memory unit , the formula is:

[0052] ;

[0053] S55, calculate output gate , the formula is:

[0054] ;

[0055] in, is the output gate weight matrix, is the output gate bias term;

[0056] S56, calculate hidden state , the formula is:

[0057] .

[0058] Furthermore, the method of predicting faults according to the hidden state calculation results of the long short-term memory network model, generating operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and pushing the prediction results and decision suggestions to the operation and maintenance management terminal includes:

[0059] In the LSTM network model, the hidden state Contains the historical information and current status of time series data, which is used to predict the probability and trend of future failures;

[0060] Calculate the fully connected layer , the formula is:

[0061] ;

[0062] in, is the weight matrix of the fully connected layer, is the bias vector of the fully connected layer;

[0063] use The function will Convert to a probability distribution:

[0064] ;

[0065] Based on the probability distribution results, the trend of future failures is predicted and analyzed, and the following operation and maintenance decision suggestions are generated, including fault warning, equipment maintenance plan, and resource allocation plan. The prediction results and decision suggestions are pushed to the operation and maintenance management terminal.

[0066] Based on the same inventive concept, on the other hand, the present invention also provides a distributed Internet of Things operation and maintenance terminal fault intelligent prediction system, the system comprising:

[0067] The data collection and preprocessing module is used to collect historical fault data and real-time operation data of distributed IoT terminal devices, wherein the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; the collected historical fault data and real-time operation data are preprocessed, including data cleaning, missing value filling, and data normalization, to form a standardized data set;

[0068] The fault pattern recognition and clustering analysis module is used to build a fault pattern recognition model based on the convolutional neural network, train the standardized data set, and calculate the output feature map of the pooling layer and the output of the fully connected layer; the K-means clustering algorithm is used to cluster the fault data to identify high-frequency fault areas and high-incidence fault types;

[0069] The fault prediction and decision support module is used to combine real-time operation data, based on the identified high-frequency fault areas and high-incidence fault types, and use the long short-term memory network in the time series prediction algorithm to calculate the hidden state; predict faults based on the hidden state calculation results of the long short-term memory network model, generate operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and push the prediction results and decision suggestions to the operation and maintenance management terminal.

[0070] (3) Beneficial effects

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] 1. Use convolutional neural networks for fault pattern recognition, use K-means clustering algorithm for cluster analysis, identify high-frequency fault areas and high-incidence fault types, and use long short-term memory networks for time series prediction to provide real-time fault warnings.

[0073] 2. Generate operation and maintenance decision suggestions based on the prediction results, including fault warning, equipment maintenance plan and resource allocation plan, so as to effectively reduce operation and maintenance costs and improve equipment reliability BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a flowchart of a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method according to Embodiment 1 of the present invention;

[0075] Figure 2 This is a module block diagram of a distributed Internet of Things operation and maintenance terminal fault intelligent prediction system according to Example 2 of the present invention. DETAILED DESCRIPTION

[0076] 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.

[0077] Before giving examples, it is necessary to explain the application scenarios conceived by the present invention. The present invention is applied to intelligent prediction of faults in IoT operation and maintenance terminals.

[0078] Example 1: Figure 1 As shown, this embodiment provides a distributed Internet of Things operation and maintenance terminal fault intelligent prediction method, the method comprising:

[0079] S1, collecting historical fault data and real-time operation data of distributed IoT terminal devices, wherein the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; preprocessing the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set;

[0080] S2, builds a fault pattern recognition model based on the convolutional neural network, trains the standardized data set, and calculates the output feature map of the pooling layer and the output of the fully connected layer; uses the K-means clustering algorithm to cluster the fault data and identify high-frequency fault areas and high-incidence fault types;

[0081] S3, combined with real-time operation data, uses the long short-term memory network in the time series prediction algorithm to calculate the hidden state based on the identified high-frequency fault areas and high-incidence fault types; predicts faults based on the calculation results of the hidden state of the long short-term memory network model, generates operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and pushes the prediction results and decision suggestions to the operation and maintenance management terminal.

[0082] For example, suppose there is a distributed IoT system with 100 terminal devices, each of which collects operating data (such as temperature, humidity, voltage, etc.) every 5 minutes. Historical fault data includes device type, fault type, fault occurrence time, fault duration, environmental parameters, etc. Real-time operating data includes device operating status, sensor data, network communication status, etc. Our goal is to identify high-frequency fault areas and predict future fault trends through the fault prediction method of the present invention.

[0083] Set parameters: Number of devices: 100, data collection frequency: 5 minutes / time, historical fault data: 1000 records, real-time operation data: each device generates one piece of data every 5 minutes;

[0084] Convolutional neural network (CNN) parameters: convolution kernel size: 3x3; number of convolution layers: 2 layers; pooling layer: max pooling, window size 2x2; fully connected layer: 128 neurons; output layer: 5 neurons (corresponding to 5 fault types);

[0085] K-means clustering algorithm parameters: number of clusters k=2; maximum number of iterations: 100;

[0086] LSTM network parameters: input dimension: 2 (fully connected layer output and pooling layer output), number of hidden layer neurons: 2, time step: 2, output dimension: 1 (failure probability);

[0087] Collect 1,000 pieces of historical fault data and real-time operation data from 100 devices;

[0088] Example of historical fault data:

[0089] Device ID: 001, Device type: Temperature sensor, Fault type: A (overtemperature), Fault occurrence time: 2023-10-01 12:00, Fault duration: 30 minutes, Temperature: 45°C, Humidity: 60%, Voltage: 220V, Network communication status: Good;

[0090] Device ID: 002, Device type: Humidity sensor, Fault type: B (abnormal humidity), Fault occurrence time: 2023-10-01 12:05, Fault duration: 20 minutes, Temperature: 50°C, Humidity: 55%, Voltage: 210V, Network communication status: Delayed;

[0091] Device ID: 003, Device type: Voltage sensor, Fault type: C (voltage fluctuation), Fault occurrence time: 2023-10-01 12:10, Fault duration: 15 minutes, Temperature: 48°C, Humidity: 58%, Voltage: 215V, Network communication status: Good;

[0092] Device ID: 004, Device type: Temperature sensor, Fault type: A (overtemperature), Fault occurrence time: 2023-10-01 12:15, Fault duration: 25 minutes, Temperature: 47°C, Humidity: 57%, Voltage: 218V, Network communication status: Good;

[0093] Device ID: 005, Device type: Humidity sensor, Fault type: B (abnormal humidity), Fault occurrence time: 2023-10-01 12:20, Fault duration: 10 minutes, Temperature: 49°C, Humidity: 59%, Voltage: 212V, Network communication status: Good;

[0094] Examples of real-time operation data:

[0095] Device ID: 001, device type: temperature sensor, temperature: 46°C, humidity: 61%, voltage: 219V, operating status: normal, network communication status: good, collection time: 2023-10-01 12:25;

[0096] Device ID: 002, device type: humidity sensor, temperature: 51°C, humidity: 56%, voltage: 209V, operation status: abnormal, network communication status: delayed, collection time: 2023-10-01 12:30;

[0097] Device ID: 003, device type: voltage sensor, temperature: 49°C, humidity: 57%, voltage: 214V, operating status: normal, network communication status: good, collection time: 2023-10-01 12:35;

[0098] Device ID: 004, device type: temperature sensor, temperature: 48°C, humidity: 58%, voltage: 217V, operation status: abnormal, network communication status: delayed, collection time: 2023-10-01 12:40;

[0099] Device ID: 005, device type: humidity sensor, temperature: 51°C, humidity: 56%, voltage: 209V, operating status: abnormal, network communication status: delayed, collection time: 2023-10-01 12:30.

[0100] Furthermore, the method of preprocessing the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set includes:

[0101] S21, data cleaning: remove duplicate data and process outliers, wherein the outliers include data points that are beyond a reasonable range and mark the outliers as missing values;

[0102] S22, missing value filling: fill missing values ​​using mean or interpolation;

[0103] S23, data normalization: scale the data to the range of [0,1], the normalization formula is as follows:

[0104] ;

[0105] in, represents the normalized data value, Represents the original data, and Respectively represent the minimum and maximum values ​​of the original data;

[0106] S24, forming a standardized data set, integrating the cleaned, filled and normalized data into a unified standardized data set .

[0107] Exemplarily, the collected historical fault data and real-time operation data are preprocessed:

[0108] Classification data such as device ID, device type, and fault type usually do not need to be cleaned, but they need to be checked for consistency and completeness, and the check results are normal; Classification data such as device ID, device type, and fault type usually do not need to be filled, but they need to be checked for completeness, and the check results are normal; Classification data such as device ID, device type, and fault type usually do not need to be normalized;

[0109] Clean the numerical data (such as temperature, humidity, and voltage) in the collected data, fill in missing values, and check that the collected data is good and does not need to be processed;

[0110] Data normalization processing, according to the normalization formula, the historical fault data and real-time operation data are normalized and the results are as follows:

[0111] temperature: =45℃, =51℃; Humidity: =55%, =61%; Voltage: =209V, =220V;

[0112] Normalization results of historical fault data: Device ID: 001, temperature 0.0, humidity: 1.0, voltage: 1.0; Device ID: 002, temperature 1.0, humidity: 0.0, voltage: 0.0; Device ID: 003, temperature 0.6, humidity: 0.6, voltage: 0.5; Device ID: 004, temperature 0.4, humidity: 0.4, voltage: 0.75; Device ID: 005, temperature 0.8, humidity: 0.8, voltage: 0.25;

[0113] Normalization results of real-time operation data: Device ID: 001, temperature 0.1667, humidity: 1.0, voltage: 0.9091; Device ID: 002, temperature 1.0, humidity: 0.1667, voltage: 0.0; Device ID: 003, temperature 0.8, humidity: 0.4, voltage: 0.4545; Device ID: 004, temperature 0.6, humidity: 0.6, voltage: 0.7273; Device ID: 005, temperature 1.0, humidity: 0.8333, voltage: 0.1818;

[0114] Constructing a unified standardized dataset matrix : Select devices 001, 002, 003 in the historical fault data and devices 001, 002 in the real-time operation data, and construct a 4×4 matrix (the last column is the time step normalization result):

[0115] .

[0116] Furthermore, the method of constructing a fault mode recognition model according to a convolutional neural network, training a standardized data set, and calculating a pooling layer output feature map and a fully connected layer output includes:

[0117] Input layer: receives standardized data set As input;

[0118] Convolution layer: extract local features through convolution kernels. The calculation formula is as follows:

[0119] ;

[0120] in, represents the output feature map of the convolutional layer, represents the convolution kernel weight, Represents input data, represents the bias term;

[0121] Activation layer: using Activation function, the formula is as follows:

[0122] ;

[0123] Pooling layer output : Use the maximum pooling operation, the formula is as follows:

[0124] ;

[0125] Among them, s represents the pooling window size;

[0126] Fully connected layer output : Flatten the output of the pooling layer and connect it to the fully connected layer. The calculation formula is as follows:

[0127] ;

[0128] in, represents the weight matrix, represents the bias vector, Represents the activation function.

[0129] For example, a fault pattern recognition model is constructed based on a convolutional neural network, and a standardized data set is trained to identify different fault patterns and their characteristics; based on the trained fault pattern recognition model, fault pattern matching is performed on real-time operation data to identify potential fault types:

[0130] Convolution layer calculation: The input data is a 4×4 matrix X:

[0131] ;

[0132] The convolution kernel W is a 3x3 matrix:

[0133] ;

[0134] Bias b = 0.1;

[0135] Convolution output:

[0136] Location Calculation:

[0137] Input area:

[0138] ;

[0139] Calculation formula:

[0140] ;

[0141] Calculation process:

[0142] Location Calculation:

[0143] Input area:

[0144] ;

[0145] Calculation formula:

[0146] ;

[0147] Calculation process:

[0148] Location Calculation:

[0149] Input area:

[0150] ;

[0151] Calculation formula:

[0152] ;

[0153] Calculation process:

[0154] Location Calculation:

[0155] Input area:

[0156] ;

[0157] Calculation formula:

[0158] ;

[0159] Calculation process:

[0160] Output feature map:

[0161] Through the above calculation, the output feature map C of the convolutional layer is:

[0162] ;

[0163] Activation layer calculation: Apply to each element of the convolutional layer output feature map C Activation function:

[0164] ; ; ; ;

[0165] The activation layer output feature map A is:

[0166] ;

[0167] Pooling layer calculation: Assume the pooling window size is 2 2, the step size is 2, according to the maximum pooling operation formula ; The calculated output feature map P of the pooling layer is:

[0168] ;

[0169] Fully connected layer calculation: The fully connected layer flattens the output of the pooling layer and connects it to the fully connected layer. Assume that the weight matrix of the fully connected layer is and the bias vector as follows:

[0170] ;

[0171] ;

[0172] The input of the fully connected layer is the flattened vector of the output feature map P of the pooling layer: ;

[0173] The calculation formula of the fully connected layer is:

[0174] ;

[0175] in, Represents the activation function ;

[0176] Fully connected layer output for:

[0177] ;

[0178] Calculate the output feature maps of other pooling layers and the output of the fully connected layer in turn.

[0179] Furthermore, the method of clustering fault data using the K-means clustering algorithm to identify high-frequency fault areas and high-incidence fault types includes:

[0180] S41, select data and randomly initialize k cluster centers;

[0181] S42, assign each data point to the nearest cluster center, and the cluster assignment calculation formula is as follows:

[0182] ;

[0183] in, represents a data point, Indicates Cluster centers;

[0184] S43, update the cluster center to the mean of the current cluster, and the calculation formula is as follows:

[0185] ;

[0186] in, Indicates A set of data points in a cluster;

[0187] S44, repeat steps S42 and S43 until the cluster center no longer changes or the maximum number of iterations is reached.

[0188] Exemplarily, data preparation: 4 data points are set, each data point has 2 features, which are the fully connected layer output and the pooling layer output feature map;

[0189] Data Points = ;

[0190] Randomly initialize k cluster centers: Assume k=2 and randomly select 2 cluster centers;

[0191] ; ;

[0192] Assign each data point to the nearest cluster center: Calculate the distance from each data point to the cluster center and assign it to the nearest cluster center;

[0193] Data Point 1: ;

[0194] arrive Distance: ;

[0195] arrive Distance: ;

[0196] Assign to ;

[0197] Data Point 2: ;

[0198] arrive Distance: ;

[0199] arrive Distance: ;

[0200] Assign to ;

[0201] Data Point 3: ;

[0202] arrive Distance: ;

[0203] arrive Distance: ;

[0204] Assign to ;

[0205] Data Point 4: ;

[0206] arrive Distance: ;

[0207] arrive Distance: ;

[0208] Assign to ;

[0209] Update the cluster center to the mean of the current cluster;

[0210] Cluster 1: contains data points 1 and 3;

[0211] ;

[0212] Cluster 2: contains data points 2 and 4;

[0213] ;

[0214] Repeat the above steps until the cluster center no longer changes or the maximum number of iterations is reached.

[0215] Second Iteration

[0216] Data Point 1: ;

[0217] arrive Distance: ;

[0218] arrive Distance: ;

[0219] Assign to ;

[0220] Data Point 2: ;

[0221] arrive Distance: ;

[0222] arrive Distance: ;

[0223] Assign to ;

[0224] Data Point 3: ;

[0225] arrive Distance: ;

[0226] arrive Distance: ;

[0227] Assign to ;

[0228] Data Point 4: ;

[0229] arrive Distance: ;

[0230] arrive Distance: ;

[0231] Assign to ;

[0232] After two iterations, the cluster center no longer changes, the algorithm terminates, and the final clustering result is (updated cluster center):

[0233] Cluster 1: contains data points 1 and 3;

[0234] ;

[0235] The fully connected layer output of data point 1 is 1.2, the pooling layer output is 2.2, and the fault type is A; the fully connected layer output of data point 3 is 1.5, the pooling layer output is 2.5, and the fault type is B;

[0236] Cluster 2: contains data points 2 and 4;

[0237] ;

[0238] The fully connected layer output of data point 2 is 0.8, the pooling layer output is 1.8, and the fault type is C; the fully connected layer output of data point 4 is 0.9, the pooling layer output is 1.9, and the fault type is D.

[0239] Furthermore, the method of calculating the hidden state by using the long short-term memory network in the time series prediction algorithm based on the identified high-frequency fault areas and high-incidence fault types in combination with the real-time operation data includes:

[0240] S51, calculate the forget gate , the formula is:

[0241] ;

[0242] in, is the activation function, is the forget gate weight matrix, is the forget gate bias term, is the hidden state, For input data;

[0243] S52, calculation input gate , the formula is:

[0244] ;

[0245] in, is the input gate weight matrix, is the input gate bias term;

[0246] S53, calculate candidate memory units , the formula is:

[0247] ;

[0248] in, is the candidate memory unit weight matrix, is the candidate memory unit bias item;

[0249] S54, computational memory unit , the formula is:

[0250] ;

[0251] S55, calculate output gate , the formula is:

[0252] ;

[0253] in, is the output gate weight matrix, is the output gate bias term;

[0254] S56, calculate hidden state , the formula is:

[0255] .

[0256] For example, the initial parameters of the long short-term memory network model are: input dimension: 2 (full connection layer output and pooling layer output), number of hidden layer neurons: 2, time step: 2; initialized according to the above parameters: memory unit , hidden state hidden state ;

[0257] Time step 1, input data , calculated as follows:

[0258] S51, calculate the forget gate :

[0259] set up , ;

[0260] ;

[0261] S52, calculation input gate :

[0262] set up , ;

[0263] ;

[0264] S53, calculate candidate memory units :

[0265] set up , ;

[0266] ;

[0267] S54, computational memory unit :

[0268] ;

[0269] S55, calculate output gate :

[0270] set up , ;

[0271] ;

[0272] S56, calculate hidden state :

[0273] ;

[0274] Time step 2, input data , calculated as follows:

[0275] S51, calculate the forget gate :

[0276] ;

[0277] S52, calculation input gate :

[0278] ;

[0279] S53, calculate candidate memory units :

[0280] ;

[0281] S54, computational memory unit :

[0282] ;

[0283] S55, calculate output gate :

[0284] ;

[0285] S56, calculate hidden state :

[0286] .

[0287] Furthermore, the method of predicting faults according to the hidden state calculation results of the long short-term memory network model, generating operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and pushing the prediction results and decision suggestions to the operation and maintenance management terminal includes:

[0288] In the LSTM network model, the hidden state Contains the historical information and current status of time series data, which is used to predict the probability and trend of future failures;

[0289] Calculate the fully connected layer , the formula is:

[0290] ;

[0291] in, is the weight matrix of the fully connected layer, is the bias vector of the fully connected layer;

[0292] use The function will Convert to a probability distribution:

[0293] ;

[0294] Based on the probability distribution results, the trend of future failures is predicted and analyzed, and the following operation and maintenance decision suggestions are generated, including fault warning, equipment maintenance plan, and resource allocation plan. The prediction results and decision suggestions are pushed to the operation and maintenance management terminal.

[0295] For example, the calculation results , contains the historical information and current status of time series data, which is used to predict the probability and trend of future faults; the output layer is set to have 3 neurons, corresponding to 3 types of faults;

[0296] Calculate the fully connected layer :

[0297] Set the weight matrix of the fully connected layer , the bias vector of the fully connected layer ;

[0298] ;

[0299] use The function will Convert to a probability distribution:

[0300] ;

[0301] ;

[0302] ;

[0303] According to the probability distribution of failure , prediction results: probability of fault type A: 30.37%, probability of fault type B: 32.90%, probability of fault type C: 36.73%;

[0304] Predict the trend of future faults: Fault type A has a low probability of occurrence, but its changing trend needs to be paid attention to; Fault type B has a medium probability of occurrence and needs to be checked regularly; Fault type C has a high probability of occurrence and needs special attention;

[0305] Generate operation and maintenance decision suggestions, and push the prediction results and operation and maintenance decision suggestions to the operation and maintenance management terminal for reference and execution by operation and maintenance personnel: Fault type C has the highest probability of occurrence (36.73%), and fault warning information is generated to remind operation and maintenance personnel to focus on it and allocate more resources for preventive maintenance and real-time monitoring; Fault type B has a medium probability of occurrence (32.90%), and a regular inspection plan is formulated to ensure the normal operation of the equipment and allocate appropriate resources for regular inspection and maintenance; Fault type A has a low probability of occurrence (30.37%), and regular maintenance is arranged to prevent failures.

[0306] Embodiment 2: Based on the same inventive concept, Figure 2 As shown, this embodiment also provides a distributed IoT operation and maintenance terminal fault intelligent prediction system, the system comprising:

[0307] The data collection and preprocessing module is used to collect historical fault data and real-time operation data of distributed IoT terminal devices, wherein the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; the collected historical fault data and real-time operation data are preprocessed, including data cleaning, missing value filling, and data normalization, to form a standardized data set;

[0308] The fault pattern recognition and clustering analysis module is used to build a fault pattern recognition model based on the convolutional neural network, train the standardized data set, and calculate the output feature map of the pooling layer and the output of the fully connected layer; the K-means clustering algorithm is used to cluster the fault data to identify high-frequency fault areas and high-incidence fault types;

[0309] The fault prediction and decision support module is used to combine real-time operation data, based on the identified high-frequency fault areas and high-incidence fault types, and use the long short-term memory network in the time series prediction algorithm to calculate the hidden state; predict faults based on the hidden state calculation results of the long short-term memory network model, generate operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and push the prediction results and decision suggestions to the operation and maintenance management terminal.

[0310] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0311] Finally, it should be noted that: Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method, characterized in that: The method comprises: S1, collecting historical fault data and real-time operation data of distributed IoT terminal devices, wherein the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; preprocessing the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set; S2, builds a fault pattern recognition model based on the convolutional neural network, trains the standardized data set, and calculates the output feature map of the pooling layer and the output of the fully connected layer; uses the K-means clustering algorithm to cluster the fault data and identify high-frequency fault areas and high-incidence fault types; S3, combined with real-time operation data, uses the long short-term memory network in the time series prediction algorithm to calculate the hidden state based on the identified high-frequency fault areas and high-incidence fault types; predicts faults based on the calculation results of the hidden state of the long short-term memory network model, generates operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and pushes the prediction results and decision suggestions to the operation and maintenance management terminal.

2. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method as claimed in claim 1, characterized in that: The method of preprocessing the collected historical fault data and real-time operation data, including data cleaning, missing value filling, and data normalization, to form a standardized data set includes: S21, data cleaning: remove duplicate data and process outliers, wherein the outliers include data points that are beyond a reasonable range and mark the outliers as missing values; S22, missing value filling: fill missing values ​​using mean or interpolation; S23, data normalization: scale the data to the range of [0,1], the normalization formula is as follows: ； in, represents the normalized data value, Represents the original data, and Respectively represent the minimum and maximum values ​​of the original data; S24, forming a standardized data set, integrating the cleaned, filled and normalized data into a unified standardized data set .

3. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method as claimed in claim 2, characterized in that: The method of constructing a fault mode recognition model according to a convolutional neural network, training a standardized data set, and calculating a pooling layer output feature map and a fully connected layer output comprises: Input layer: receives standardized data set As input; Convolution layer: extract local features through convolution kernels. The calculation formula is as follows: ； in, represents the output feature map of the convolutional layer, represents the convolution kernel weight, Represents input data, represents the bias term; Activation layer: using Activation function, the formula is as follows: ； Pooling layer output : Use the maximum pooling operation, the formula is as follows: ； Among them, s represents the pooling window size; Fully connected layer output : Flatten the output of the pooling layer and connect it to the fully connected layer. The calculation formula is as follows: ； in, represents the weight matrix, represents the bias vector, Represents the activation function.

4. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method as claimed in claim 3, characterized in that: The method of clustering fault data using the K-means clustering algorithm to identify high-frequency fault areas and high-incidence fault types includes: S41, select data and randomly initialize k cluster centers; S42, assign each data point to the nearest cluster center, and the cluster assignment calculation formula is as follows: ； in, represents a data point, Indicates Cluster centers; S43, update the cluster center to the mean of the current cluster, and the calculation formula is as follows: ； in, Indicates A set of data points in a cluster; S44, repeating steps S42 and S43 until the cluster center no longer changes or the maximum number of iterations is reached, and identifying high-frequency fault areas and high-incidence fault types based on the cluster center.

5. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method as claimed in claim 4, characterized in that: The method of calculating the hidden state by using the long short-term memory network in the time series prediction algorithm based on the identified high-frequency fault areas and high-incidence fault types in combination with the real-time operation data includes: S51, calculate the forget gate , the formula is: ； in, is the activation function, is the forget gate weight matrix, is the forget gate bias term, is the hidden state, For input data; S52, calculation input gate , the formula is: ； in, is the input gate weight matrix, is the input gate bias term; S53, calculate candidate memory units , the formula is: ； in, is the candidate memory unit weight matrix, is the candidate memory unit bias item; S54, computational memory unit , the formula is: ； S55, calculate output gate , the formula is: ； in, is the output gate weight matrix, is the output gate bias term; S56, calculate hidden state , the formula is: 。 6. A distributed Internet of Things operation and maintenance terminal fault intelligent prediction method as claimed in claim 5, characterized in that: The method of predicting faults according to the hidden state calculation results of the long short-term memory network model, generating operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and pushing the prediction results and decision suggestions to the operation and maintenance management terminal includes: In the LSTM network model, the hidden state Contains the historical information and current status of time series data, which is used to predict the probability and trend of future failures; Calculate the fully connected layer , the formula is: ； in, is the weight matrix of the fully connected layer, is the bias vector of the fully connected layer; use The function will Convert to a probability distribution: ； Based on the probability distribution results, the trend of future failures is predicted and analyzed, and the following operation and maintenance decision suggestions are generated, including fault warning, equipment maintenance plan, and resource allocation plan. The prediction results and decision suggestions are pushed to the operation and maintenance management terminal.

7. A distributed IoT operation and maintenance terminal fault intelligent prediction system, characterized in that: The system comprises: The data collection and preprocessing module is used to collect historical fault data and real-time operation data of distributed IoT terminal devices, wherein the historical fault data includes device type, fault type, fault occurrence time, fault duration, and environmental parameters, and the real-time operation data includes device operation status, sensor data, and network communication status; the collected historical fault data and real-time operation data are preprocessed, including data cleaning, missing value filling, and data normalization, to form a standardized data set; The fault pattern recognition and clustering analysis module is used to build a fault pattern recognition model based on the convolutional neural network, train the standardized data set, and calculate the output feature map of the pooling layer and the output of the fully connected layer; the K-means clustering algorithm is used to cluster the fault data to identify high-frequency fault areas and high-incidence fault types; The fault prediction and decision support module is used to combine real-time operation data, based on the identified high-frequency fault areas and high-incidence fault types, and use the long short-term memory network in the time series prediction algorithm to calculate the hidden state; predict faults based on the hidden state calculation results of the long short-term memory network model, generate operation and maintenance decision suggestions, including fault warnings, equipment maintenance plans, and resource allocation plans, and push the prediction results and decision suggestions to the operation and maintenance management terminal.

Images