Road material equipment residual life prediction method and system based on neural network

Abstract

The invention provides a road material equipment residual life prediction method and system based on a neural network, and relates to the technical field of equipment residual life prediction, and the method comprises the steps: calibrating each piece of road material equipment as a node on a two-dimensional plane, distributing a two-dimensional coordinate identifier, and collecting the operation parameters and environmental parameters of each piece of road material equipment; representing state information of each node by using a multi-dimensional feature vector, classifying related nodes into the same equipment group through comprehensive association degree, and marking unassociated nodes as independent abnormal equipment; acquiring historical operation parameters, environmental parameters and residual life, constructing a time sequence sample set, and training a neural network model based on the sample set to predict the residual life of the equipment; inputting the parameters of each group and the independent equipment into a model, outputting a residual life prediction value, and identifying the predicted residual life of the equipment groups and the independent equipment by using different colors on a construction site map; according to the method, prediction precision is improved through space-time correlation analysis, and intelligent operation and maintenance management of equipment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment remaining life prediction, and in particular to a method and system for predicting the remaining life of highway material equipment based on a neural network. Background Art

[0002] In the process of modern highway construction and maintenance, the effective management of highway materials and equipment is particularly important. With the expansion of engineering project scale and the extension of construction period, the frequency of equipment use continues to increase, and the problem of equipment failure and damage is becoming increasingly prominent. Traditional equipment management methods usually rely on regular maintenance and manual inspections. This method is not only inefficient but also lacks real-time and accuracy. The failure to detect potential equipment failures in a timely manner leads to unplanned equipment downtime during use, which in turn affects the construction progress and overall project quality. In addition, current equipment management is mostly based on experience and simple statistical analysis, which cannot fully consider the changes in the status of equipment under different operating environments, resulting in a significant reduction in the accuracy of the prediction model. This makes construction units face high maintenance costs and time losses, and there is an urgent need for a more scientific and effective method for predicting the remaining life of equipment.

[0003] Existing equipment remaining life prediction technologies are often limited to linear models or simple regression analysis based on historical data, failing to fully utilize multidimensional feature information. Many methods focus only on a single operating parameter of the equipment, ignoring the impact of environmental factors such as temperature, humidity, and rainfall on equipment performance. In addition, the lack of dynamic real-time monitoring of equipment status makes it impossible to adjust maintenance strategies in a timely manner during equipment operation. This static analysis method not only reduces prediction accuracy, but also prevents the equipment from achieving optimal utilization efficiency. In addition, traditional methods often use isolated analysis when dealing with the relationship between equipment, failing to consider the mutual influence between equipment groups, resulting in insufficient overall equipment management. Therefore, there is an urgent need for a new technical solution that can comprehensively consider the multidimensional status information and environmental factors of the equipment, while realizing dynamic monitoring and analysis of the equipment, so as to improve the accuracy of equipment remaining life prediction and the management level.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting the remaining life of road material equipment based on a neural network to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The remaining life prediction method of highway material equipment based on neural network includes the following specific steps:

[0008] Step 1: Each piece of highway material equipment is calibrated as a node on a two-dimensional plane and assigned a unique two-dimensional coordinate identifier. The operating parameters and environmental parameters of each node are collected. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall.

[0009] Step 2: Based on the operating parameters and environmental parameters of each node, a multidimensional feature vector is used to represent the device status information of each node. The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified as the same device group, and the rest are marked as independent abnormal devices.

[0010] Step 3: Obtain multiple sets of historical operating parameters, environmental parameters, and remaining life of highway materials equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model;

[0011] Step 4: Input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value;

[0012] Step 5: Based on the two-dimensional coordinate identification of the equipment, the remaining life prediction value of each equipment group and independent abnormal equipment is marked with different colors on the construction site map. When the remaining life prediction value is lower than the preset life threshold, an early warning is triggered.

[0013] Furthermore, based on GPS positioning, a unique two-dimensional coordinate identifier is assigned to each piece of road material equipment. The specific logic is as follows:

[0014] A plane rectangular coordinate system is established for the construction site, with the southwest corner of the site as the coordinate origin, the east-west direction as the X-axis, and the north-south direction as the Y-axis. An RTK-GPS receiver is used to measure the actual position of each device and record its plane coordinates, which serve as the device's two-dimensional coordinate identifier. As the device moves, the coordinate data is updated in real time using the Beidou positioning terminal installed on the device.

[0015] Each piece of highway material equipment is calibrated in the two-dimensional plane coordinate system of the construction site, and a unique two-dimensional coordinate identifier is assigned to each device. The operating parameters and environmental parameters of each node are collected in real time. The operating parameters include the service life of the equipment, the cumulative working hours and the number of maintenance times. The environmental parameters include temperature, humidity and rainfall. The service life of the equipment refers to the length of time the equipment has been in use. The temperature, humidity and rainfall are daily average data.

[0016] Both operating parameters and environmental parameters are obtained through on-site surveys and sensor monitoring as device attributes of each node.

[0017] Furthermore, based on the operating parameters and environmental parameters of each node, a multidimensional feature vector is used to represent the device status information of each node. The expression of the multidimensional feature vector is as follows:

[0018]

[0019] Where, F i is the multidimensional feature vector of node i, i is the index of the node, f use,i and f env,i Represent the operating parameters and environmental parameters of node i, X i 、Y i and Z i are the service life, cumulative working hours and maintenance times of the equipment at node i, T i 、S i and R i They are the temperature, humidity and rainfall of node i respectively;

[0020] The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. The formula is as follows:

[0021]

[0022] Where, EIE ij Represents the comprehensive correlation between node i and node j, i and j are the indexes of the nodes, and i≠j; Duse(f use,i ,f use,j ) and Denv(f env,i ,f env,j ) represent the difference in operation characteristics and environment characteristics between node i and node j respectively; f env,i and f env,j They represent the operating parameters and environmental parameters of node j, e is a natural constant, X j 、Y j and Z j are the service life, cumulative working hours and maintenance times of the equipment at node j, T j 、S j and R j are the temperature, humidity and rainfall of node j respectively, ω1 and ω2 are preset weights, ω1>ω2>0, and ω1+ω2=1.

[0023] Furthermore, two nodes with a comprehensive correlation greater than the dynamic correlation critical value are grouped into the same equipment group, and the rest are marked as independent abnormal equipment. The specific logic is as follows: the comprehensive correlation between all node pairs is calculated, and the dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations; an undirected graph containing all nodes is constructed, and node pairs with a comprehensive correlation greater than the dynamic correlation critical value are connected by edges. A depth-first search algorithm is used to identify connected subgraphs in the graph to form equipment groups. At the same time, isolated nodes without any edge connections are marked as independent abnormal equipment. Finally, a division result containing various equipment groups and independent abnormal equipment is output, thereby dividing the target highway material equipment into multiple equipment groups and independent abnormal equipment.

[0024] The dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations, and the formula is as follows:

[0025] θ=μ+k*σ

[0026] Where k is the preset adjustment coefficient, and the value range of k is [0.2, 0.8].

[0027] Furthermore, the method for generating the time series sample set is: mapping each set of operating parameters, environmental parameters and the remaining life of the equipment one by one to form a data chain, and recording the formed data chain as the time series sample set;

[0028] Based on the data in the time series sample set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model.

[0029] The Tanh function expression is as follows:

[0030]

[0031] Wherein, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's input, that is, the result of the weighted summation of the input received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of training samples per batch is set to 256, and the number of hidden layer neurons is 32;

[0032] The time series sample set is randomly divided into a training set and a test set, and a device remaining life prediction model is constructed based on a deep learning algorithm. The operating parameters and environmental parameters in the training set are used as input, and the remaining life of the equipment is used as a label. The model is trained to obtain a trained device remaining life prediction model. The operating parameters and environmental parameters in the test set are substituted into the trained model to obtain the corresponding prediction result; the error between the prediction result and the actual value in the test set is calculated; and it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the device remaining life prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set;

[0033] The process of randomly dividing the data into training and test sets is as follows: the time series sample set is randomly sorted, 80% of the sorted time series sample set is used as the training set, and the remaining 20% ​​is used as the test set;

[0034] The trained equipment remaining life prediction model takes operating parameters and environmental parameters as input and outputs the equipment remaining life;

[0035] Among them, historical operating parameters and environmental parameters are aligned at a daily granularity, and the remaining life of the equipment is marked as the actual number of days from the current day to scrapping.

[0036] Furthermore, the mean operating parameters and environmental parameters of all nodes in each device group are calculated as the feature input of the group. At the same time, the original operating parameters and environmental parameters of independent abnormal devices are directly used as input. The processed feature data are input into the trained equipment remaining life prediction model. The model outputs the remaining life prediction value of each device group and independent abnormal device through forward propagation calculation. The prediction results are stored in a structured data format and bound to the device's unique identifier.

[0037] Furthermore, based on the two-dimensional coordinate identification of the equipment, the remaining life prediction values ​​of each equipment group and independent abnormal equipment are marked with different colors on the construction site map. The specific logic is as follows: a heat map is drawn on the digital construction site map based on the two-dimensional coordinate identification of the equipment, and the remaining life prediction values ​​are represented by a color gradient, where red indicates a remaining life prediction value between [0, 30] days, yellow indicates a remaining life prediction value between (30, 90] days, and green indicates a remaining life prediction value between (90, +∞] days;

[0038] When the remaining life prediction value is lower than the preset life threshold, an early warning is automatically triggered, and the early warning information containing the equipment location, predicted remaining life and recommended maintenance measures is pushed to the responsible personnel's mobile terminal through the Internet of Things platform.

[0039] The present invention also provides another neural network-based remaining life prediction system for highway materials and equipment. The neural network-based remaining life prediction system for highway materials and equipment is used to implement the above-mentioned neural network-based remaining life prediction method for highway materials and equipment, including:

[0040] The equipment node calibration module is used to calibrate each piece of highway material equipment as a node on a two-dimensional plane and assign it a unique two-dimensional coordinate identifier. The module also collects the operating parameters and environmental parameters of each node. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall.

[0041] The device grouping module is used to characterize the device status information of each node using a multidimensional feature vector based on the operating parameters and environmental parameters of each node. The module also defines the comprehensive correlation between any two nodes based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified into the same device group, and the rest are marked as independent abnormal devices.

[0042] The model building and training module is used to obtain the historical operating parameters, environmental parameters, and remaining life of multiple sets of highway material equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model;

[0043] The life prediction module is used to input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value;

[0044] The visual early warning module is used to identify the remaining life prediction value of each equipment group and independent abnormal equipment with different colors on the construction site map based on the two-dimensional coordinate identification of the equipment. When the remaining life prediction value falls below the preset life threshold, an early warning is triggered.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] Through multi-dimensional feature modeling and dynamic correlation analysis, the present invention achieves intelligent grouping of highway material equipment and precise identification of abnormal equipment, significantly improving the accuracy and adaptability of remaining life prediction. Compared with traditional methods, this solution can more comprehensively capture the correlation between equipment operating status and environmental factors, effectively reducing prediction errors, while enhancing the model's generalization capabilities in complex construction scenarios. In addition, the present invention's visual early warning mechanism provides intuitive and real-time decision support for equipment management, can identify potential failure risks in advance and provide graded early warnings, thereby optimizing equipment maintenance strategies. This solution not only improves the operation and maintenance efficiency of highway construction equipment, but also provides reliable technical support for intelligent construction management, and has significant engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0048] Figure 2 This is a schematic diagram of the overall system module of the present invention;

[0049] Figure 3-4 They are the fitting curves of operating characteristic differences and comprehensive correlation, and the fitting curves of environmental characteristic differences and comprehensive correlation, respectively. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0051] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0052] Example:

[0053] See also Figure 1 , the present invention provides a technical solution:

[0054] The remaining life prediction method of highway material equipment based on neural network includes the following specific steps:

[0055] Step 1: Each piece of highway material equipment is calibrated as a node on a two-dimensional plane and assigned a unique two-dimensional coordinate identifier. The operating parameters and environmental parameters of each node are collected. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall.

[0056] In this embodiment, a unique two-dimensional coordinate identifier is assigned to each road material equipment based on GPS positioning. The specific logic is as follows:

[0057] A plane rectangular coordinate system is established for the construction site, with the southwest corner of the site as the coordinate origin, the east-west direction as the X-axis, and the north-south direction as the Y-axis. An RTK-GPS receiver is used to measure the actual position of each device and record its plane coordinates, which serve as the device's two-dimensional coordinate identifier. As the device moves, the coordinate data is updated in real time using the Beidou positioning terminal installed on the device.

[0058] Each piece of highway material equipment is calibrated in the two-dimensional plane coordinate system of the construction site, and a unique two-dimensional coordinate identifier is assigned to each device. The operating parameters and environmental parameters of each node are collected in real time. The operating parameters include the service life of the equipment, the cumulative working hours and the number of maintenance times. The environmental parameters include temperature, humidity and rainfall. The service life of the equipment refers to the length of time the equipment has been in use. The temperature, humidity and rainfall are daily average data.

[0059] Both operating parameters and environmental parameters are obtained through on-site surveys and sensor monitoring as device attributes of each node.

[0060] The advantage of step 1 is that by calibrating each piece of road material equipment as a node on a two-dimensional plane and assigning it a unique two-dimensional coordinate identifier, it enables precise positioning and real-time monitoring of the equipment. Compared to traditional equipment management methods, this approach more effectively integrates equipment operating parameters with environmental parameters, forming a data foundation that provides high-quality input for subsequent analysis and decision-making. This real-time monitoring capability enables construction units to promptly understand equipment status, reduce equipment failures, and thus improve management efficiency.

[0061] In this solution, implementing Step 1 lays the data foundation for the overall solution, enabling more accurate equipment status analysis and remaining life prediction in subsequent steps. By acquiring and integrating specific operating and environmental information for each device, the subsequent multidimensional feature vector construction and comprehensive correlation calculation can be more targeted and accurate, thereby improving the effectiveness of device grouping and the reliability of the prediction model. The accuracy of this foundational data directly impacts the performance of the entire prediction system and promotes the implementation of intelligent equipment management.

[0062] Step 2: Based on the operating parameters and environmental parameters of each node, a multidimensional feature vector is used to represent the device status information of each node. The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified as the same device group, and the rest are marked as independent abnormal devices.

[0063] In this embodiment, a multidimensional feature vector is used to represent the device status information of each node based on the operating parameters and environmental parameters of each node. The expression of the multidimensional feature vector is as follows:

[0064]

[0065] Where, F i is the multidimensional feature vector of node i, i is the index of the node, f use,i and f env,i Represent the operating parameters and environmental parameters of node i, X i 、Y i and Z i are the service life, cumulative working hours and maintenance times of the equipment at node i, T i 、S i and R i They are the temperature, humidity and rainfall of node i respectively;

[0066] The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. The formula is as follows:

[0067]

[0068] Where, EIE ij Represents the comprehensive correlation between node i and node j, i and j are the indexes of the nodes, and i≠j; Duse(f use,i ,f use,j ) and Denv(f env,i ,f env,j ) represent the difference in operation characteristics and environment characteristics between node i and node j respectively; f env,i and f env,j They represent the operating parameters and environmental parameters of node j, e is a natural constant, X j 、Y j and Z j are the service life, cumulative working hours and maintenance times of the equipment at node j, T j 、S j and R jThese are the temperature, humidity, and rainfall at node j, respectively. ω1 and ω2 are preset weights: ω1 = 0.6 and ω2 = 0.4. These weights are set because operational characteristics generally have a more significant impact on device status than environmental characteristics. Operational parameters such as equipment age, accumulated operating hours, and number of repairs directly reflect the equipment's usage and health, making them crucial for predicting its remaining life. Environmental parameters such as temperature, humidity, and rainfall, while also having some influence, are relatively indirect. Therefore, increasing the weight of the operational characteristic, ω1, can more effectively capture the key factors influencing device performance, thereby improving the accuracy of the comprehensive correlation and the performance of the prediction model.

[0069] In the first formula given, the dependent variable EIE ij It represents the comprehensive correlation between node i and node j, reflecting the similarity between the two nodes in terms of operating characteristics and environmental characteristics. Specifically, the closer its value is to 1, the higher the similarity between the nodes and the stronger the correlation, while the closer its value is to 0, the lower the similarity. Its technical effect is to quantify the relationship between nodes, facilitate the classification and group management of equipment, thereby providing data support for fault prediction and maintenance decision-making, and improving the intelligent level of equipment management. use,i ,f use,j ) and Denv(f env,i ,f env,j ) represent the operational characteristic difference and environmental characteristic difference between node i and node j, respectively. The two are calculated by the service life, cumulative working hours, maintenance times, temperature, humidity and rainfall of the equipment. The correlation between these independent variables and the dependent variable is reflected in that when the difference between node i and node j is small, its comprehensive correlation EIE ij A higher value indicates that the nodes have a stronger similarity; on the contrary, when the difference increases, the comprehensive correlation decreases, reflecting that the similarity is weakened. use,i ,f use,j )、Denv(f env,i ,f env,j ) and EIE ij This formula effectively quantifies the comprehensive correlation EIE between node i and node j by comprehensively considering the difference in operating characteristics and environmental characteristics and introducing preset weights. ij Formally, an exponential function is used to smooth the effect of differences on correlation, making the comprehensive correlation more sensitive when differences are small, while effectively suppressing the decline in correlation when differences increase. Furthermore, the comprehensive correlation value is normalized to the range of (0, 1) by taking the reciprocal, ensuring the interpretability and comparability of the results, thereby ensuring the rationality and practicality of the formula.

[0070] Table 1: Comprehensive correlation statistics

[0071] Operational characteristic difference Environmental characteristic difference Comprehensive correlation 1.73 2.89 0.462 2.24 3.16 0.399 2.83 5.39 0.368 3.1 6.08 0.238 3.16 6.18 0.23 3.61 6.32 0.215 4 7 0.193 4.47 7.07 0.185 4.5 7.81 0.17 5.1 8.06 0.162 5.74 8.37 0.151 5.83 9 0.143 6.45 10 0.133 7.5 11.18 0.118 7.87 12.25 0.107

[0072] See also Figure 3-Figure 4 In this data analysis, the dynamic response characteristics of the multidimensional feature correlation model can be clearly observed through analysis of measured data on the operational characteristic differences, environmental characteristic differences, and comprehensive correlation between device nodes. As the operational characteristic differences increase from 1.73 to 7.87 and the environmental characteristic differences increase from 2.89 to 12.25, the comprehensive correlation shows a regular decreasing trend, from 0.462 to 0.107. This verifies the rationality of the weight coefficient setting ω1>ω2, that is, the operational characteristics have a greater impact on device correlation than the environmental characteristics.

[0073] In the low-difference range (i.e., when the operating characteristic difference is <3 and the environmental characteristic difference is <6), the comprehensive correlation remains above 0.238, indicating that there is significant functional similarity between the devices, which meets the conditions for dividing the device groups in step 2. When the difference enters the medium range (operating characteristic difference is 3-5, and environmental characteristic difference is 6-8), the correlation drops to 0.17-0.215, at which point there is only partial similarity in the operating conditions between the devices. In the high-difference range (operating characteristic difference >5, environmental characteristic difference >8), the comprehensive correlation falls below 0.15. Such devices should be marked as independent abnormal devices, which is consistent with the abnormality handling mechanism in step 4.

[0074] The mean of the comprehensive correlation of all data is 0.215, with a standard deviation of 0.108. When the adjustment coefficient k = 0.5, the dynamic correlation critical value is 0.269. Approximately 30% of the node pairs in the measured data have correlations exceeding this threshold, which is consistent with the typical distribution ratio of equipment groups in actual engineering. This proves that the calculation formula θ = μ + kσ used in step 4 is applicable to engineering projects.

[0075] Two nodes with a comprehensive correlation greater than the dynamic correlation critical value are grouped together, and the rest are marked as independent abnormal devices. The specific logic is as follows: the comprehensive correlation between all node pairs is calculated, and the dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations; an undirected graph containing all nodes is constructed, and node pairs with a comprehensive correlation greater than the dynamic correlation critical value are connected by edges. A depth-first search algorithm is used to identify connected subgraphs in the graph to form equipment groups. Isolated nodes without any edge connections are marked as independent abnormal devices. The final output includes the division results of various equipment groups and independent abnormal devices, thereby dividing the target highway material equipment into multiple equipment groups and independent abnormal devices.

[0076] The dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations, and the formula is as follows:

[0077] θ=μ+k*σ

[0078] Wherein, k is the preset adjustment coefficient, k=0.5.

[0079] This formula is used to determine the dynamic correlation threshold, which is calculated based on the mean μ and standard deviation σ of all the combined correlations. Specifically, the mean μ represents the central tendency of the combined correlations between all nodes, reflecting the overall level of similarity between nodes, while the standard deviation σ measures the dispersion of these correlation values, that is, the variability of similarity between nodes.

[0080] By introducing the adjustment factor k, the formula effectively adjusts the mean upward by a value proportional to the standard deviation. This design helps to statistically form a dynamic threshold θ, which not only reflects the overall similarity between nodes but also appropriately accounts for the impact of fluctuations in similarity between different nodes. Once θ is calculated, it is used for subsequent node classification, helping to identify nodes whose comprehensive correlation exceeds the normal range, allowing for targeted analysis and treatment. The advantage of this approach is that it can dynamically adapt to changes in the relationships between nodes, improving the flexibility and accuracy of the system.

[0081] Step 2 achieves refined characterization of device status and intelligent grouping by constructing a multidimensional feature vector and calculating dynamic correlation. Compared to traditional clustering methods that simply rely on distance or a single metric, this step comprehensively considers the interactive effects of device operating parameters and environmental factors. By weightedly integrating Euclidean distance and feature similarity, the grouping is more consistent with engineering practice. Its innovation lies in the adaptive calculation mechanism of the dynamic correlation threshold, which can automatically adjust the classification criteria based on the distribution of device group characteristics, avoiding the subjectivity of manually setting fixed thresholds.

[0082] Existing technologies often use static clustering algorithms or simple threshold methods to group equipment, which are difficult to adapt to the complex and changing environments of construction sites. This step, by introducing a comprehensive correlation model, effectively addresses the problems of traditional methods' lack of sensitivity to abnormal equipment and rigid grouping. In particular, by differentially weighting operational and environmental characteristics, it more accurately reflects the causal relationship between equipment degradation and environmental impacts, improving the ability to identify patterns of coordinated equipment degradation compared to traditional methods.

[0083] This step provides a scientifically sound input data organization structure for subsequent neural network predictions. By grouping devices with similar degradation patterns, the LSTM model's efficiency in extracting common features is significantly enhanced. Simultaneously, the identification of individual abnormal devices ensures prediction reliability under unique operating conditions. This dual-track mechanism of "group commonality learning + individual specificity processing" gives the entire prediction system both generalizability and flexibility, laying the data foundation for high-precision predictions in step 4 and ultimately achieving full optimization from data to decision-making.

[0084] Step 3: Obtain multiple sets of historical operating parameters, environmental parameters, and remaining life of highway materials equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model;

[0085] In this embodiment, the method for generating the time series sample set is: mapping each set of operating parameters, environmental parameters and the remaining life of the equipment one by one to form a data chain, and recording the formed data chain as a time series sample set;

[0086] Based on the data in the time series sample set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model.

[0087] The Tanh function expression is as follows:

[0088]

[0089] Wherein, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's input, that is, the result of the weighted summation of the input received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of training samples per batch is set to 256, and the number of hidden layer neurons is 32;

[0090] The time series sample set is randomly divided into a training set and a test set, and a device remaining life prediction model is constructed based on a deep learning algorithm. The operating parameters and environmental parameters in the training set are used as input, and the remaining life of the equipment is used as a label. The model is trained to obtain a trained device remaining life prediction model. The operating parameters and environmental parameters in the test set are substituted into the trained model to obtain the corresponding prediction result; the error between the prediction result and the actual value in the test set is calculated; and it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the device remaining life prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set;

[0091] The process of randomly dividing the data into training and test sets is as follows: the time series sample set is randomly sorted, 80% of the sorted time series sample set is used as the training set, and the remaining 20% ​​is used as the test set;

[0092] The trained equipment remaining life prediction model takes operating parameters and environmental parameters as input and outputs the equipment remaining life;

[0093] Among them, historical operating parameters and environmental parameters are aligned at a daily granularity, and the remaining life of the equipment is marked as the actual number of days from the current day to scrapping.

[0094] Step 3 extracts time series features of the device degradation process by constructing a time series sample set and training an LSTM neural network model. This method innovatively aligns device operating parameters with environmental parameters in time series, capturing the long-term dependencies of device performance degradation through the LSTM model's unique gating mechanism. In particular, the combination of the Tanh activation function and the Adam optimizer effectively addresses the vanishing gradient problem and accelerates model convergence, enabling the prediction model to more accurately reflect the nonlinear changes in device performance over time.

[0095] Compared to traditional static prediction models, the LSTM time series modeling approach employed in this step overcomes the limitations of existing technologies for extracting time-dimensional features. Existing methods often simply stack or statistically analyze time series data, resulting in the loss of critical temporal dynamic information. However, this solution, through a three-layer LSTM network structure and 200 training iterations, accurately models the degradation trajectory of the device throughout its lifecycle. In particular, the sensitivity for detecting sudden performance degradation is significantly improved compared to traditional methods. Furthermore, an intelligent 80%-20% data partitioning strategy ensures the model's combined learning and generalization capabilities.

[0096] The time series prediction model constructed in this step is the technical core of the entire solution, providing a reliable computing engine for the real-time predictions in Step 4. By combining device group characteristics with the time series processing capabilities of the LSTM, it not only preserves the common characteristics of the devices classified in Step 2, but also captures the time series evolution patterns of individual devices. This dual modeling mechanism of "spatial correlation + temporal evolution" enables prediction results to reflect the overall trends of the device group while adapting to the specific operating conditions of individual devices. Ultimately, it achieves a precise mapping from historical data to future predictions, providing a scientific basis for early warning decisions in Step 5.

[0097] Step 4: Input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value;

[0098] In this embodiment, the operating parameters and environmental parameters of all nodes in each device group are calculated as the mean of the group's feature input. At the same time, the original operating parameters and environmental parameters of the independent abnormal device are directly used as input. The processed feature data are input into the trained equipment remaining life prediction model. The model outputs the remaining life prediction value of each device group and independent abnormal device through forward propagation calculation. The prediction results are stored in a structured data format and bound to the device's unique identifier.

[0099] Step 4 accurately predicts the remaining life of the equipment by inputting both the characteristics of the equipment group and the data of the individual abnormal equipment into the trained prediction model. This step innovatively employs a dual-track input strategy of "group mean + individual raw values": Averages of operating and environmental parameters are calculated for the equipment group, preserving the group's common characteristics; while the raw data of the individual abnormal equipment is directly used to ensure accurate reflection of specific operating conditions. This differentiated data processing approach improves prediction efficiency for conventional equipment while ensuring prediction accuracy for abnormal cases.

[0100] Compared to the "one-size-fits-all" data processing model of traditional prediction methods, this step addresses the existing technology's inadequate handling of device heterogeneity by dynamically distinguishing between group devices and outliers. Existing technologies typically apply a unified prediction process to all devices, resulting in significant prediction bias for outliers. However, this solution, based on intelligent grouping in step 2, optimizes the allocation of prediction resources: It leverages the "wisdom of the crowd" to enhance prediction stability for group devices, while independently analyzing outliers to avoid "group bias." Specifically, when using forward propagation, the model automatically adjusts the attention weights for different input types, ensuring that predictions are both group-consistent and individual-specific.

[0101] This step is the critical transition point for the entire method from theory to application. On the one hand, it inherits the powerful time series feature extraction capabilities of the LSTM model trained in Step 3; on the other hand, it provides targeted predictions based on the intelligent grouping results of Step 2, providing high-quality data input for the visual warnings in Step 5. This "group prediction + anomaly identification" mechanism enables the system to grasp the overall trend of equipment degradation while promptly detecting sudden signs of failure, ultimately achieving a balance between prediction accuracy and practicality. Furthermore, structured data storage ensures traceability of prediction results, providing a data foundation for subsequent model iteration and optimization.

[0102] Step 5: Based on the two-dimensional coordinates of the equipment, the remaining life prediction values ​​of each equipment group and individual abnormal equipment are marked with different colors on the construction site map. When the remaining life prediction value falls below the preset life threshold, an early warning is triggered;

[0103] In this embodiment, the remaining life prediction values ​​of each equipment group and independent abnormal equipment are marked with different colors on the construction site map according to the two-dimensional coordinate identification of the equipment. The specific logic is as follows: a heat map is drawn on the digital construction site map according to the two-dimensional coordinate identification of the equipment, and the remaining life prediction values ​​are represented by a color gradient, where red indicates a remaining life prediction value between [0, 30] days, yellow indicates a remaining life prediction value between (30, 90] days, and green indicates a remaining life prediction value between (90, +∞] days;

[0104] When the remaining life prediction value is lower than the preset life threshold, an early warning is automatically triggered, and the early warning information containing the equipment location, predicted remaining life and recommended maintenance measures is pushed to the responsible personnel's mobile terminal through the Internet of Things platform.

[0105] The advantage of step 5 is that by visualizing the equipment's remaining life predictions and color-coding them on the construction site map, managers are significantly more intuitive in their understanding and monitoring of equipment status. Compared to traditional text reports or data tables, this approach provides a more intuitive display of equipment health and warning information, allowing managers to quickly identify equipment requiring special attention and optimize decision-making.

[0106] In this solution, step 5 provides effective visualization support for the overall solution, making equipment management more intelligent and efficient. Presenting the predicted remaining life via a heat map not only improves information dissemination but also triggers timely warnings, ensuring that managers can quickly implement appropriate maintenance measures. This combination of visualization and early warning mechanisms significantly improves project safety and equipment efficiency, reduces the risks and economic losses associated with potential failures, and thus enhances overall construction management.

[0107] See also Figure 2, a highway material equipment remaining life prediction system based on neural network, comprising:

[0108] The equipment node calibration module is used to calibrate each piece of highway material equipment as a node on a two-dimensional plane and assign it a unique two-dimensional coordinate identifier. The module also collects the operating parameters and environmental parameters of each node. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall.

[0109] The device grouping module is used to characterize the device status information of each node using a multidimensional feature vector based on the operating parameters and environmental parameters of each node. The module also defines the comprehensive correlation between any two nodes based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified into the same device group, and the rest are marked as independent abnormal devices.

[0110] The model building and training module is used to obtain the historical operating parameters, environmental parameters, and remaining life of multiple sets of highway material equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model;

[0111] The life prediction module is used to input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value;

[0112] The visual early warning module is used to identify the remaining life prediction value of each equipment group and independent abnormal equipment with different colors on the construction site map based on the two-dimensional coordinate identification of the equipment. When the remaining life prediction value falls below the preset life threshold, an early warning is triggered.

[0113] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0114] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0115] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0116] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for predicting the remaining life of highway materials and equipment based on a neural network, characterized in that: The specific steps include: Step 1: Each piece of highway material equipment is calibrated as a node on a two-dimensional plane and assigned a unique two-dimensional coordinate identifier. The operating parameters and environmental parameters of each node are collected. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall. Step 2: Based on the operating parameters and environmental parameters of each node, a multidimensional feature vector is used to represent the device status information of each node. The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified as the same device group, and the rest are marked as independent abnormal devices. Step 3: Obtain multiple sets of historical operating parameters, environmental parameters, and remaining life of highway materials equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model; Step 4: Input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value; Step 5: Based on the two-dimensional coordinate identification of the equipment, the remaining life prediction value of each equipment group and independent abnormal equipment is marked with different colors on the construction site map. When the remaining life prediction value is lower than the preset life threshold, an early warning is triggered.

2. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 1, characterized in that: Based on GPS positioning, each piece of road material equipment is assigned a unique two-dimensional coordinate identifier. The specific logic is as follows: A plane rectangular coordinate system is established for the construction site, with the southwest corner of the site as the coordinate origin, the east-west direction as the X-axis, and the north-south direction as the Y-axis. An RTK-GPS receiver is used to measure the actual position of each device and record its plane coordinates, which serve as the device's two-dimensional coordinate identifier. As the device moves, the coordinate data is updated in real time using the Beidou positioning terminal installed on the device. Each piece of highway material equipment is calibrated in the two-dimensional plane coordinate system of the construction site, and a unique two-dimensional coordinate identifier is assigned to each device. The operating parameters and environmental parameters of each node are collected in real time. The operating parameters include the service life of the equipment, the cumulative working hours and the number of maintenance times. The environmental parameters include temperature, humidity and rainfall. The service life of the equipment refers to the length of time the equipment has been in use. The temperature, humidity and rainfall are daily average data. Both operating parameters and environmental parameters are obtained through on-site surveys and sensor monitoring as device attributes of each node.

3. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 2, characterized in that: According to the operating parameters and environmental parameters of each node, a multidimensional feature vector is used to represent the device status information of each node. The expression of the multidimensional feature vector is as follows: Where, F i is the multidimensional feature vector of node i, i is the index of the node, f use,i and f env,i Represent the operating parameters and environmental parameters of node i, X i 、Y i and Z i are the service life, cumulative working hours and maintenance times of the equipment at node i, T i 、S i and R i They are the temperature, humidity and rainfall of node i respectively; The comprehensive correlation between any two nodes is defined based on the multidimensional feature vector. The formula is as follows: Where, EIE ij Represents the comprehensive correlation between node i and node j, where i and j are the indexes of the nodes and i≠j; Duse(f use,i ,f use,j ) and Denv(f env,i ,f env,j ) represent the operational characteristic difference and environmental characteristic difference between node i and node j respectively; f env,i and f env,j They represent the operating parameters and environmental parameters of node j, e is a natural constant, X j 、Y j and Z j are the service life, cumulative working hours and maintenance times of the equipment at node j, T j 、S j and R j are the temperature, humidity and rainfall of node j respectively, ω1 and ω2 are preset weights, ω1>ω2>0, and ω1+ω2=1.

4. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 1, characterized in that: Two nodes with a comprehensive correlation greater than the dynamic correlation critical value are grouped together, and the rest are marked as independent abnormal devices. The specific logic is as follows: the comprehensive correlation between all node pairs is calculated, and the dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations; an undirected graph containing all nodes is constructed, and node pairs with a comprehensive correlation greater than the dynamic correlation critical value are connected by edges. A depth-first search algorithm is used to identify connected subgraphs in the graph to form equipment groups. Isolated nodes without any edge connections are marked as independent abnormal devices. The final output includes the division results of various equipment groups and independent abnormal devices, thereby dividing the target highway material equipment into multiple equipment groups and independent abnormal devices. The dynamic correlation critical value θ is determined based on the mean μ and standard deviation σ of all comprehensive correlations, and the formula is as follows: θ=μ+k*σ Where k is the preset adjustment coefficient, and the value range of k is [0.2, 0.8].

5. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 4, characterized in that: The method for generating the time series sample set is as follows: mapping each set of operating parameters, environmental parameters and the remaining life of the equipment one by one to form a data chain, and recording the formed data chain as the time series sample set; Based on the data in the time series sample set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model. The Tanh function expression is as follows: Wherein, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's input, that is, the result of the weighted summation of the input received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of training samples per batch is set to 256, and the number of hidden layer neurons is 32; The time series sample set is randomly divided into a training set and a test set, and a device remaining life prediction model is constructed based on a deep learning algorithm. The operating parameters and environmental parameters in the training set are used as input, and the remaining life of the equipment is used as a label. The model is trained to obtain a trained device remaining life prediction model. The operating parameters and environmental parameters in the test set are substituted into the trained model to obtain the corresponding prediction result; the error between the prediction result and the actual value in the test set is calculated; and it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the device remaining life prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set; The process of randomly dividing the data into training and test sets is as follows: the time series sample set is randomly sorted, 80% of the sorted time series sample set is used as the training set, and the remaining 20% ​​is used as the test set; The trained equipment remaining life prediction model takes operating parameters and environmental parameters as input and outputs the equipment remaining life; Among them, historical operating parameters and environmental parameters are aligned at a daily granularity, and the remaining life of the equipment is marked as the actual number of days from the current day to scrapping.

6. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 1, characterized in that: The operating parameters and environmental parameters of all nodes in each device group are calculated as the mean of the group's feature input. At the same time, the original operating parameters and environmental parameters of independent abnormal devices are directly used as input. The processed feature data are input into the trained equipment remaining life prediction model. The model outputs the remaining life prediction value of each device group and independent abnormal device through forward propagation calculation. The prediction results are stored in a structured data format and bound to the device's unique identifier.

7. The method for predicting the remaining life of road materials and equipment based on a neural network according to claim 1, characterized in that: According to the two-dimensional coordinate identification of the equipment, the remaining life prediction value of each equipment group and independent abnormal equipment is marked with different colors on the construction site map. The specific logic is as follows: based on the two-dimensional coordinate identification of the equipment, a heat map is drawn on the digital construction site map, and the remaining life prediction value is represented by a color gradient, where red indicates a remaining life prediction value between [0,30] days, yellow indicates a remaining life prediction value between (30,90] days, and green indicates a remaining life prediction value between (90,+∞] days; When the remaining life prediction value is lower than the preset life threshold, an early warning is automatically triggered, and the early warning information containing the equipment location, predicted remaining life and recommended maintenance measures is pushed to the responsible personnel's mobile terminal through the Internet of Things platform.

8. A neural network-based system for predicting the remaining life of highway materials and equipment, characterized by: The neural network-based remaining life prediction system for highway materials and equipment is used to implement the neural network-based remaining life prediction method for highway materials and equipment according to any one of claims 1 to 7, comprising: The equipment node calibration module is used to calibrate each piece of highway material equipment as a node on a two-dimensional plane and assign it a unique two-dimensional coordinate identifier. The module also collects the operating parameters and environmental parameters of each node. The operating parameters include the equipment's service life, cumulative working hours, and number of maintenance times. The environmental parameters include temperature, humidity, and rainfall. The device grouping module is used to characterize the device status information of each node using a multidimensional feature vector based on the operating parameters and environmental parameters of each node. The module also defines the comprehensive correlation between any two nodes based on the multidimensional feature vector. Two nodes with a comprehensive correlation greater than the dynamic correlation threshold are classified into the same device group, and the rest are marked as independent abnormal devices. The model building and training module is used to obtain the historical operating parameters, environmental parameters, and remaining life of multiple sets of highway material equipment, construct a time series sample set, map each set of operating parameters and environmental parameters to the remaining life of the equipment, establish a neural network model, and train the neural network model based on the time series sample set to obtain an equipment remaining life prediction model; The life prediction module is used to input the operating parameters and environmental parameters of each device group and independent abnormal device into the trained device remaining life prediction model, and the model outputs the remaining life prediction value; The visual early warning module is used to identify the remaining life prediction value of each equipment group and independent abnormal equipment with different colors on the construction site map based on the two-dimensional coordinate identification of the equipment. When the remaining life prediction value falls below the preset life threshold, an early warning is triggered.

Images