CN117933499A - Invasion risk prediction method, device and storage medium for high-speed railway catenary - Google Patents
Invasion risk prediction method, device and storage medium for high-speed railway catenary Download PDFInfo
- Publication number
- CN117933499A CN117933499A CN202410332226.5A CN202410332226A CN117933499A CN 117933499 A CN117933499 A CN 117933499A CN 202410332226 A CN202410332226 A CN 202410332226A CN 117933499 A CN117933499 A CN 117933499A
- Authority
- CN
- China
- Prior art keywords
- sample data
- network
- weight
- neural networks
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 230000009545 invasion Effects 0.000 title claims description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 147
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 48
- 239000013598 vector Substances 0.000 claims abstract description 32
- 230000000946 synaptic effect Effects 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000007667 floating Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000007613 environmental effect Effects 0.000 description 5
- 238000012423 maintenance Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 208000034699 Vitreous floaters Diseases 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000003442 weekly effect Effects 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- XSQUKJJJFZCRTK-UHFFFAOYSA-N Urea Chemical compound NC(N)=O XSQUKJJJFZCRTK-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000004202 carbamide Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010891 electric arc Methods 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The disclosure relates to an intrusion risk prediction method, device and storage medium for a high-speed railway catenary, wherein the method comprises the following steps: constructing T initial neural networks, and training the T initial neural networks to obtain T trained target neural networks; respectively inputting a preset sample data set into each target neural network, and obtaining an output matrix according to the output vectors of the T target neural networks; calculating a pseudo-inverse matrix of the output matrix, carrying out regularization treatment on the pseudo-inverse matrix, and determining a network weight matrix corresponding to the T target neural networks according to the regularization treatment result; determining the network weight of each target neural network according to the weight vector in the network weight matrix; and constructing an integrated network according to the network weight and the T target neural networks so as to predict the risk of the high-speed railway overhead line system according to the integrated network. Therefore, risk prediction of the contact network is realized, and accuracy and robustness of risk prediction of the high-speed railway contact network are improved.
Description
Technical Field
The disclosure relates to the technical field of traffic safety, in particular to an intrusion risk prediction method, an intrusion risk prediction device and a storage medium for a high-speed railway overhead line system.
Background
At present, an overhead contact line without a backup device on a high-speed railway is used as a key carrier for traction power transmission and mainly comprises a contact suspension, a supporting device, a positioning device, a support and a foundation. The safety performance of the catenary is of great concern due to various operational risks, such as natural aging, human factors, external weather conditions, and the like. According to the statistical analysis of the historical failure records, the contact net is more easily affected by external weather conditions, especially wind, because the contact net is completely exposed outdoors. Wind has become one of the important factors causing the failure of the overhead contact line, and the overhead contact line spans a wide range due to the fact that the wind changes along with the climate period and various terrains, so that the overhead contact line can be interfered by different wind speeds and wind directions. More specifically, the kite, the plastic bag, the urea ribbon, the cloth strip, the reflective film and other floaters are easy to hang on the contact net under the action of wind, and are collectively called wind-induced floaters to invade. The contact net failure caused by the invasion of the floats is mainly divided into the following failure scenes: 1) A trip accident. The wind-induced drift intrusion causes spatial electric field distortion, reducing the insulation margin of the contact net wires. When severe, the air gap is directly short-circuited, causing the wire to ground, the wire to wire, the wire to post discharge, and thus causing a trip event. 2) Secondary flashover accident. In general, because the contact net distance is far, the invasion of the floater caused by wind can not be cleared in time, and the overvoltage generated by reclosing can cause a secondary flashover accident in an air gap, which can cause the burning out or even fusing of a lead. 3) The pantograph collides and even the catenary collapses. When a high-speed train passes through a floating object invasion zone, floating objects wound around the pantograph are more likely to cause the pantograph to collide and arc discharge, even cause disconnection accidents of the contact net, and the consequences are disastrous, so that huge economic loss and social negative effects are often caused. These accidents can also have serious adverse effects on the safety and elasticity of the contact network. If the risk level of the wind-caused floating object can be effectively predicted, the operation and maintenance management department of the overhead contact line can make a corresponding operation and maintenance plan, such as personnel arrangement, inspection strategies and the like, so as to reduce the operation risk of the overhead contact line caused by the floating object. Therefore, in order to make a predictive operation and maintenance decision, and further improve the safety and the elastic performance of the overhead contact system, it is necessary to predict the invasion risk of the wind-caused floating object.
In the related art, the method is mainly focused on an intrusion monitoring method, and the existing foreign matter intrusion monitoring method comprises a video identification method. The video recognition method is to analyze pictures shot by a video recorder in real time, judge whether foreign matters exist or not by utilizing a video recognition technology, intuitively feel actual conditions on site, realize monitoring of the whole road section by installing the video recognition method at intervals, easily suffer from external interference such as weather and the like, and have low accuracy.
Disclosure of Invention
In order to solve the technical problems or at least partially solve the technical problems, the disclosure provides an intrusion risk prediction method, an intrusion risk prediction device and a storage medium for a high-speed railway catenary, so as to solve the technical problem of lower intrusion risk detection accuracy for the high-speed railway catenary in the prior art.
The present disclosure provides an intrusion risk prediction method for a high-speed railway catenary, comprising constructing T initial neural networks, and training the T initial neural networks according to a preset sample data set to obtain trained T target neural networks, wherein T is a natural number greater than 1; respectively inputting the preset sample data set into each target neural network, and obtaining an output matrix according to the output vectors of the T target neural networks; calculating a pseudo-inverse matrix of the output matrix, regularizing the pseudo-inverse matrix, and determining a network weight matrix corresponding to the T target neural networks according to regularized results; determining the network weight of each target neural network according to the weight vector in the network weight matrix; and constructing an integrated network according to the network weight and the T target neural networks, so as to predict the risk of the high-speed railway catenary according to the integrated network.
The present disclosure provides an intrusion risk prediction apparatus for a high-speed railway catenary, comprising: the first training module is used for constructing T initial neural networks and training the T initial neural networks according to a preset sample data set to obtain T trained target neural networks, wherein T is a natural number greater than 1; the second training module is used for respectively inputting the preset sample data sets into each target neural network and obtaining an output matrix according to the output vectors of the T target neural networks; the weight matrix determining module is used for calculating a pseudo-inverse matrix of the output matrix, regularizing the pseudo-inverse matrix and determining a network weight matrix corresponding to the T target neural networks according to regularized results; the weight determining module is used for determining the network weight of each target neural network according to the weight vector in the network weight matrix; and the integrated network construction module is used for constructing an integrated network according to the network weight and the T target neural networks so as to predict the risk of the high-speed railway catenary according to the integrated network.
The present disclosure provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the above method.
Compared with the prior art, the technical scheme provided by the embodiment of the disclosure has the following advantages:
Constructing T initial neural networks, training the T initial neural networks according to a preset sample data set to obtain T trained target neural networks, wherein T is a natural number larger than 1, further, respectively inputting the preset sample data set to each target neural network, obtaining an output matrix according to output vectors of the T target neural networks, calculating a pseudo-inverse matrix of the output matrix, carrying out regularization treatment on the pseudo-inverse matrix, determining a network weight matrix corresponding to the T target neural networks according to regularization treatment results, determining network weight of each target neural network according to weight vectors in the network weight matrix, and constructing an integrated network according to the network weight and the T target neural networks so as to facilitate risk prediction of a high-speed railway catenary according to the integrated network. According to the technical scheme, risk prediction of the contact network is realized, and the capability of predicting uncertainty and stronger robustness are realized under the conditions of unbalance and a small data set, so that the accuracy and the robustness of risk prediction of the contact network of the high-speed railway are improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure.
In order to more clearly illustrate the embodiments of the present disclosure or the solutions in the prior art, the drawings that are required for the description of the embodiments or the prior art will be briefly described below, and it will be obvious to those skilled in the art that other drawings can be obtained from these drawings without inventive effort.
Fig. 1 is a schematic flow chart of an intrusion risk prediction method for a high-speed railway catenary according to an embodiment of the present disclosure;
fig. 2 is a schematic diagram of a training scenario of an initial neural network according to an embodiment of the present disclosure;
Fig. 3 is a schematic diagram of a training process of an initial neural network according to an embodiment of the disclosure;
fig. 4 is a schematic structural diagram of an integrated network according to an embodiment of the disclosure;
Fig. 5 is a schematic structural diagram of an intrusion risk prediction device for a high-speed railway catenary according to an embodiment of the present disclosure.
Detailed Description
In order that the above objects, features and advantages of the present disclosure may be more clearly understood, a further description of aspects of the present disclosure will be provided below. It should be noted that, without conflict, the embodiments of the present disclosure and features in the embodiments may be combined with each other.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure, but the present disclosure may be practiced otherwise than as described herein; it will be apparent that the embodiments in the specification are only some, but not all, embodiments of the disclosure.
The intrusion risk prediction method of the high-speed railway catenary according to the embodiment of the disclosure is described below with reference to the accompanying drawings, and the intrusion risk prediction method of the high-speed railway catenary can be applied to an intrusion risk prediction device of the high-speed railway catenary. As shown in fig. 1, the intrusion risk prediction method for the high-speed railway catenary includes:
step 101, constructing T initial neural networks, and training the T initial neural networks according to a preset sample data set to obtain T trained target neural networks, wherein T is a natural number greater than 1.
In order to ensure accuracy of subsequent risk prediction, the preset sample data set in the embodiment of the disclosure includes multiple types of sample data, where the multiple types of sample data include at least one of average maximum wind speed, maximum wind speed, sum of included angles between the line and wind direction, sum of environmental sensitive parameters, and corresponding invasion times of wind, so that a correlation model of subsequent training can be ensured to fully utilize factors such as space-time correlation, uncertainty of the maximum wind speed and wind direction, environmental sensitive parameters, and the like, and the method has capability of predicting uncertainty and stronger robustness under the conditions of unbalance and small data set.
In some possible embodiments, the collection of the sample data set is performed based on wind-induced float intrusion records including fault location (strut number and fault kilometer sign), date and time of float intrusion, description of float intrusion, catenary assembly, etc., wherein three key issues should be considered in order to study the complex relationship between the number of float intrusions and the numerical data of the wind. The first is to determine the weather station from which the wind data is obtained. Considering that the overhead line system line spans a wide area, a plurality of weather stations are arranged in different cities. Thus, a weather station located near the catenary is selected as the wind data source. The second is to determine the prediction period. And setting a prediction period to be one week according to investigation on the invasion times of the floats and the maintenance intervals of the overhead contact system. The third is to classify the risk level of wind-induced drift intrusion. The weekly wind-induced drift intrusion frequency of the catenary is distributed between 0 and 11.
It is apparent that most weekly drift intrusion counts are medium and low. Thus, the counts associated with wind-induced drift intrusion fall into three categories. These three categories account for 90.70%, 6.98% and 2.33%, respectively. The risk levels of 0 and 1 account for more than 97%. Because of the maximum proportion, the number of times of invasion of the floating objects caused by wind is zero is classified as one type. A medium risk level indicates that the wind induced drift intrusion is one to two times. In addition, because input data of the catenary wind-induced drift invasion risk prediction has universality and space-time correlation, a candidate sample data set for model training needs to be selected by performing space-time similarity test.
Wherein, spatial similarity test: the aim is to evaluate the mutual spatial similarity existing in the wind data of different catenary galleries. The purpose of the spatial similarity check is to delete contact net galleries from the candidate list that are not related to other galleries. The time correlation test is to select a high rank wind data set of the catenary corridor and remove redundancy. And finally obtaining a candidate sample data set of the catenary through spatial similarity and time correlation detection.
In some possible embodiments, four key features of the wind variables, namely, the average maximum wind speed around, the maximum wind speed around, the circumferential sum of angles to the wind direction, and the circumferential sum of environmental sensitive parameters (the sum of angles to the wind direction around, and the sum of environmental sensitive parameters around), may be selected, and these four wind features are closely related to the weekly float invasion risk level of each catenary corridor, and based on these four wind features, the corresponding input and the corresponding wind-induced float invasion risk level, the four wind features corresponding to the candidate sample data set and the number of times of invasion of the corresponding wind may be used as the preset sample data set for model training in this embodiment.
In one embodiment of the present disclosure, T initial neural networks are constructed, where T is a natural number greater than 1, each initial neural network being a multi-layer neural network having a single hidden layer.
In this embodiment, T initial neural networks are trained according to a preset sample data set to obtain T target neural networks after training.
The training manner of the T target neural networks may be any adaptive enhancement training method for training the neural networks, which is exemplified as follows:
In some possible embodiments, a first synaptic strength of an output layer and a single hidden layer of each initial neural network and a second synaptic strength of the single hidden layer and the corresponding output layer are obtained, a first bias corresponding to the first synaptic strength and a second bias corresponding to the second synaptic strength are obtained, and an output function of each initial neural network is constructed according to the first synaptic strength, the second synaptic strength, the first bias and the second bias, wherein the output function may be represented as the following formula (1), in this embodiment, when the input is The predicted output is:
Formula (1)
Wherein O is an output layer, j is a j-th hidden neuron, H is a hidden layer, M is the number of hidden neurons,To input the number of X, the ith component of X is/>Wherein the component comprises, for example, one of average maximum wind speed, maximum wind speed, line angle to wind direction, environmental sensitivity parameter, etc./>, within a weekFor the second synaptic strength,/>For the first synaptic strength,/>Is a first bias,/>Is a second bias, function/>As a common S-shaped function, as/>。
And acquiring a target sample data weight of each preset sample data corresponding to each initial neural network, and iteratively updating the first synaptic strength, the second synaptic strength, the first bias and the second bias according to the output function, the preset sample data set and the target sample data weight to acquire updated first synaptic strength, updated second synaptic strength, updated first bias and updated second bias.
Wherein the relevant synaptic strength and bias may be updated in a random gradient descent iterative update, in some possible embodiments, the update rules may be summarized as the following equation (2), wherein,Output result for prediction of relevant network output,/>For the true output result of the corresponding sample data:
Formula (2)
Wherein,For any of the updated first synaptic strength, the updated second synaptic strength, the updated first bias and the updated second bias,/>For presetting the learning rate, the learning rate is an important parameter in the neural network training, and determines the size of each step of adjustment when the weight of the model is updated in the optimization process. The correct choice of learning rate is critical to the convergence speed and stability of the model. A smaller learning rate may result in the model taking longer to find the optimal solution, while a larger learning rate may result in the model skipping the optimal point or sinking into a local minimum. The choice of learning rate generally requires balancing two aspects: one aspect is to avoid overfitting: too little learning rate can cause the model to slowly change during the initial period of training, which can lead to the model having difficulty capturing global features of the data, thereby affecting generalization ability. Another aspect is to reduce the computational cost: excessive learning rates can increase computational overhead because the model needs to perform more weight updates in each iteration.
In practice, it is common practice to set the learning rate to a small value, such as 0.01 or 0.001, to ensure that the model is trained effectively with limited computing resources. n is the nth actual training sample data set,Sample dataset weights for the nth actual training sample dataset,/>To correspond to the actual output of the neural network,Is the predicted output.
In this embodiment, the target neural network corresponding to each initial neural network is determined according to the updated first synaptic strength, the updated second synaptic strength, the updated first bias and the updated second bias.
In this embodiment, modeling of a single neuron, D is the number of inputs x; the synaptic strength is used to control the strength of influence of one neuron on another, as well as the direction of influence; the S-type function is an activation function and must be nonlinear, which is to introduce nonlinearity into the neural network and increase the computational complexity of the neural network. The training of the neural network is to update the corresponding synaptic strength and bias, and pi is the synaptic strength or the corresponding bias obtained by each training.
In this example, when the initial neural network currently trained is the initial neural network trained first, the total number of samples of the preset sample data set is determined, and the inverse of the total number of samples is determined as the target sample data weight of each preset sample data. That is, as shown in fig. 2, the T neural networks are trained one by one in sequence, and when the first initial neural network is trained, the target sample data weight of each preset sample data corresponding to the first initial neural network is shown in the following formula (3), where in the formula (3), N is the total number of samples of the preset sample data set, N is the nth preset sample data,Target sample data weights for the nth preset sample data of the nth neural network:
formula (3)
I.e. in this embodiment the initial neural network for the first training receives an equal amount of training for each preset sample data.
With continued reference to fig. 2, in the case of not the initial neural network for the first training, the number of error samples of the target neural network for which the last training was completed is determined, and the network error rate is determined according to the number of error samples and the total number of samples.
And determining the absolute value of the difference value between the predicted output result and the actual output result of each preset sample data corresponding to the target neural network after the previous training, calculating the ratio of the absolute value of the difference value to the corresponding actual output result, obtaining the historical target sample data weight of the target neural network after the previous training corresponding to each preset sample data, comparing the historical target sample data with the first preset sample data with the comparison value larger than the preset error threshold value, determining that the historical target sample data weight of the first preset sample data is the candidate sample data weight of the first preset sample data, comparing the second preset sample data with the comparison value not larger than the preset error threshold value, determining that the product value of the historical target sample data weight of the second preset sample data and the network error rate is the candidate sample data weight of the second preset sample data, and carrying out normalization processing on the candidate sample data weights of the first preset sample data and the second preset sample data to obtain the target sample data weights of the first preset sample data and the second preset sample data.
In the present embodiment, for each subsequent index, it isEach sample weight/>Is based on the previous sample as/>(1 /)Fractional error generated by the individual neural networks, use/>Representation of(I.e./>)) />, In a single input output training sampleAnd each. To do this, the algorithm maintains a threshold/>. When the absolute relative error is at/>The initial neural network output of the preset sample data is considered error-free, and the error calculation is referred to as the following equation (4):
Formula (4)
I.e. in the present embodiment, new weightsFrom the previous/>According to the following formula (5), wherein in the formula (5), the preset sample data which is not predicted to be correct is given a larger target sample data weight, that is, a larger probability is taken by the following initial neural network to train,/>For the corresponding initial neural networkNetwork error rates generated at the end of training, e.g. determining a sample set/>, of training errors for each initial neural networkWherein/>The determination of (a) is referred to the following equation (6)/(b)For example, the following formula (7) is obtained, for example, 10 preset sample data, the prediction is performed on 6 (the first preset sample data), the error rate is 40%, the original weight of the historical target sample data is 0.1, the weight of the historical target sample data is reduced if the updated 4 prediction errors are the weight of 0.1,6 prediction errors (the second preset sample data), the weight of the target sample data is 0.1×0.4=0.04, and the like, so that the purpose is to make the prediction error have more damaged sample data, to train the following initial neural network and ensure the reliability of the following initial neural network:
Formula (5)
Formula (6)
Formula (7)
As mentioned in the above example, after the preset sample data set includes 10 preset Yang Gen data, if the target sample weight of the second preset sample data is updated to 0.04, it is obvious that the sum of the target sample weights corresponding to the 10 preset sample data is not 1, that is, the sum of 4 0.1 and 6 0.04 is not 1, so that the candidate sample data weights of the first preset sample data and the second preset sample data are normalized, and the sum of the target sample data weights of the first preset sample data and the second preset sample data after the normalization is 1.
In different application scenarios, the normalization processing manners of the candidate sample data weights of the first preset sample data and the second preset sample data are different, and in some possible embodiments, the normalization processing may be performed by adopting the following formula (8), where the sum of the target sample data weights of the first preset sample data and the second preset sample data after the normalization processing is 1 as shown in the formula (9).
Formula (8)
Formula (9)
Summarizing, in one embodiment of the present disclosure, referring to fig. 3, when training t=1 to T for T initial neural networks, the network is first trained using equations (1) and (2)Calculating a network error rate using equation (6) and equation (7)Calculating a target sample data weight/>, using equation (5), for each preset sample data for each initial declaration networkPair distribution/>, using equation (8)And (5) carrying out normalization processing.
Step 102, respectively inputting a preset sample data set to each target neural network, and obtaining an output matrix according to the output vectors of the T target neural networks.
In one embodiment of the present disclosure, after obtaining T target neural networks, the T target neural networks are integrated and trained, that is, a preset sample data set is input to each target neural network, and an output matrix is obtained according to output vectors of the T target neural networks.
Referring to fig. 4, the predicted output result of the integrated networkIs all/>The weighted sum of the individual target neural networks, in this embodiment, the output of each network may be organized as one/>Vector/>Output matrix/>。
And 103, calculating a pseudo-inverse matrix of the output matrix, regularizing the pseudo-inverse matrix, and determining a network weight matrix corresponding to the T target neural networks according to regularized results.
Step 104, determining the network weight of each target neural network according to the weight vector in the network weight matrix.
In the present embodiment, considering that in the conventional adaptive enhancement algorithm, in the output of the integrated network, the weights received by the target neural networks are considered to be proportional to the logarithm of their inverse error rates, the network weights set for the t target neural networks are the following formula (10):
formula (10)
Thus, the output of the integrated network available based on the architecture of the integrated network in fig. 4 is shown in the following formula (11):
formula (11)
In this embodiment, in order to minimize the sum of square differences between all preset sample data, so as to reduce the sensitivity of the target neural network to the samples, increase the robustness, and improve the determination manner of the network weights. Are respectively provided withColumn vector/>And/>Wherein/>Vector representation of the actual output of the integrated network,/>For vector representation of the predicted output of the integrated network, then the sum of squares of the errors between the actual output and the preset output can be represented as the following equation (12), where,/>, as described aboveRepresenting the true output result of the sample,/>The calculation result obtained by the algorithm of the disclosure is a vector representation of a result of prediction output of an integrated network, and the square sum of errors is a variance represented by the following formula:
Formula (12)
The output of each target neural network may be organized as oneVector/>In a similar manner, the network weights may be arranged as one/>Vector/>. Definition/>Output matrix/>Thus, in a modified algorithm, the output vector can be expressed as formula (13):
Formula (13)
Wherein, equation (13) can be regarded as the vector identification of equation (11), and in this embodiment, if E is minimized, the corresponding network weight matrix can be obtained by the following equation (14):
Formula (14)
Equation (14) isPseudo-inverse rule of matrix/>Is/>Is a pseudo-inverse of (a).
In this embodiment, regularization is performed on the pseudo-inverse matrix, network weight matrices corresponding to the T target neural networks are determined according to the regularization result, and network weights of each target neural network are determined according to weight vectors in the network weight matrices.
In one embodiment of the present disclosure, regularization may be added for numerical stability of the matrix inversion, at which timeCan be obtained,/>Is a preset small constant. That is, a network weight matrix is obtained by the formula (15), where the network weight matrix is a matrix vector of t×1, and each matrix vector represents a network weight of a corresponding target neural network, so that the network weight of each target neural network can be determined according to the weight vector in the network weight matrix:
Formula (15)
And 105, constructing an integrated network according to the network weight and the T target neural networks so as to predict the risk of the high-speed railway catenary according to the integrated network.
In this embodiment, the foregoing fig. 4 is continuously adopted, and an integrated network may be constructed according to the network weights and T target neural networks, so as to perform risk prediction of the high-speed railway catenary according to the integrated network.
In an embodiment of the present disclosure, a preset sample data set includes multiple types of intrusion parameter sample data of a high-speed railway catenary, for example, including the above-mentioned cycle average maximum wind speed, cycle maximum wind speed, cycle sum of angles between the line and the wind direction, cycle sum of environment sensitive parameters, and the like, when performing risk prediction of the high-speed railway catenary according to an integrated network, multiple types of intrusion parameter actual values of the high-speed railway catenary are obtained, the multiple types of intrusion parameter actual values are respectively input into each target neural network of the integrated network to obtain T initial predicted values, where the initial predicted values in the embodiment may be the number of times of occurrence of intrusion risk, etc., product values of the initial predicted values of each target neural network and corresponding network weights are calculated to obtain T product values, and the T product values are summed and calculated to obtain the risk predicted value of the high-speed railway catenary. Here, the risk prediction value may be the number of intrusion risks or the like.
Therefore, in the invasion risk prediction method of the high-speed railway catenary, various invasion parameters are comprehensively considered, for example, invasion risk prediction is carried out by combining the circumferential sum of the circumferential average maximum wind speed, the circumferential maximum wind speed and the included angle between the line and the wind direction and the circumferential sum of environment sensitive parameters, the four parameters are closely related to the invasion risk level of each floating object per each catenary corridor, the prediction accuracy and the robustness are improved, and the integrated network can fully utilize factors such as space-time correlation, the uncertainty of the maximum wind speed and the wind direction, the environment sensitive parameters and the like, and has the capability of predicting the uncertainty and stronger robustness under the conditions of unbalance and a small data set. The integrated network in the present disclosure has smaller errors, smaller uncertainty ratio, and larger prediction uncertainty coverage. The method can obtain a robust and reliable prediction result, has higher confidence, is convenient for a decision maker to trust the result provided by the prediction model based on the improved self-adaptive learning algorithm, and provides a feasible thought for guiding the operation and maintenance plan by using the estimated prediction uncertainty. Meanwhile, the provided prediction model has lower operation cost, and can effectively reduce the fault time of the overhead line system caused by the floats.
In summary, according to the intrusion risk prediction method of the high-speed railway catenary of the embodiment of the disclosure, T initial neural networks are constructed, the T initial neural networks are trained according to preset sample data sets to obtain trained T target neural networks, and then the preset sample data sets are respectively input to each target neural network, output matrixes are obtained according to output vectors of the T target neural networks, pseudo-inverse matrixes of the output matrixes are calculated, regularization processing is performed on the pseudo-inverse matrixes, network weight matrixes corresponding to the T target neural networks are determined according to regularization processing results, network weights of each target neural network are determined according to weight vectors in the network weight matrixes, and an integrated network is constructed according to the network weights and the T target neural networks, so that risk prediction of the high-speed railway catenary is performed according to the integrated network. According to the technical scheme, risk prediction of the contact network is realized, and the capability of predicting uncertainty and stronger robustness are realized under the conditions of unbalance and a small data set, so that the accuracy and the robustness of risk prediction of the contact network of the high-speed railway are improved.
In order to achieve the above embodiment, the present disclosure further provides an intrusion risk prediction apparatus for a high-speed railway catenary.
Fig. 5 is a schematic structural diagram of an intrusion risk prediction device for a high-speed railway catenary according to an embodiment of the present disclosure, where the intrusion risk prediction device may be implemented by software and/or hardware, and may be generally integrated in an electronic device. As shown in fig. 5, the apparatus includes: a first training module 510, a second training module 520, a weight matrix determination module 530, a weight determination module 540, an integrated network construction module 550, wherein,
The first training module 510 is configured to construct T initial neural networks, and train the T initial neural networks according to a preset sample data set to obtain trained T target neural networks, where T is a natural number greater than 1;
the second training module 520 is configured to input a preset sample data set to each target neural network, and obtain an output matrix according to output vectors of the T target neural networks;
The weight matrix determining module 530 is configured to calculate a pseudo-inverse matrix of the output matrix, perform regularization processing on the pseudo-inverse matrix, and determine a network weight matrix corresponding to the T target neural networks according to a regularization result;
A weight determining module 540, configured to determine a network weight of each target neural network according to the weight vector in the network weight matrix;
and the integrated network construction module 550 is used for constructing an integrated network according to the network weights and the T target neural networks so as to predict the risk of the high-speed railway catenary according to the integrated network.
The intrusion risk prediction device for the high-speed railway overhead line system provided by the embodiment of the disclosure can execute the intrusion risk prediction method for the high-speed railway overhead line system provided by any embodiment of the disclosure, has corresponding functional modules and beneficial effects of the execution method, and is similar in implementation principle and method and not described in detail herein.
To achieve the above embodiments, the present disclosure also proposes a computer program product comprising a computer program/instruction which, when executed by a processor, implements the intrusion risk prediction method of the high-speed railway catenary in the above embodiments.
In order to achieve the above embodiment, the present disclosure further provides an electronic device, where the electronic device includes the intrusion risk prediction device of the high-speed railway catenary.
In order to implement the above embodiment, the present disclosure further proposes a computer-readable storage medium having stored thereon a computer program that, when executed by a processor, implements the intrusion risk prediction method of the high-speed railway catenary described above.
It should be noted that in this document, relational terms such as "first" and "second" and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing is merely a specific embodiment of the disclosure to enable one skilled in the art to understand or practice the disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the disclosure. Thus, the present disclosure is not intended to be limited to the embodiments shown and described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. The intrusion risk prediction method for the high-speed railway overhead line system is characterized by comprising the following steps of:
constructing T initial neural networks, and training the T initial neural networks according to a preset sample data set to obtain T trained target neural networks, wherein T is a natural number greater than 1;
Respectively inputting the preset sample data set into each target neural network, and obtaining an output matrix according to the output vectors of the T target neural networks;
Calculating a pseudo-inverse matrix of the output matrix, regularizing the pseudo-inverse matrix, and determining a network weight matrix corresponding to the T target neural networks according to regularized results;
Determining the network weight of each target neural network according to the weight vector in the network weight matrix;
And constructing an integrated network according to the network weight and the T target neural networks, so as to predict the risk of the high-speed railway catenary according to the integrated network.
2. The method of claim 1, wherein each of the initial neural networks is a multi-layer neural network having a single hidden layer, and the training process of each initial neural network comprises:
acquiring first synaptic strength of an output layer and a single hidden layer of each initial neural network and second synaptic strength of the single hidden layer and a corresponding output layer;
Acquiring a first bias corresponding to the first synaptic strength and a second bias corresponding to the second synaptic strength;
Constructing an output function of each of the initial neural networks based on the first synaptic strength, the second synaptic strength, the first bias and the second bias;
acquiring a target sample data weight of each preset sample data corresponding to each initial neural network;
iteratively updating the first synaptic strength, the second synaptic strength, the first bias and the second bias according to the output function, the preset sample data set and the target sample data weight to obtain updated first synaptic strength, updated second synaptic strength, updated first bias and updated second bias;
And determining a target neural network corresponding to each initial neural network according to the updated first synaptic strength, the updated second synaptic strength, the updated first bias and the updated second bias.
3. The method of claim 2, wherein the output function of each of the initial neural networks comprises:
wherein y is an output function, M is the number of neurons contained in the single hidden layer, j is the corresponding hidden layer neurons, O is an output layer, As a squeezing function,/>For/>，/>For the first synaptic strength,/>Is the second synaptic strength, where D is the number of inputs x, the ith component of input x is/>,/>For the first bias,/>Is a second bias.
4. The method of claim 3, wherein the iterative result of each iterative update of the first synaptic strength, the second synaptic strength, the first bias and the second bias based on the output function and the actual training sample data set is:
wherein, For any of the updated first synaptic strength, the updated second synaptic strength, the updated first bias and the updated second bias,/>For the preset learning rate, n is the nth actual training sample data set,/>Sample dataset weights for the nth actual training sample dataset.
5. The method of any one of claims 2-4, wherein the obtaining the target sample data weight of each preset sample data corresponding to each initial neural network includes:
Determining whether the currently trained initial neural network is a first trained initial neural network;
and under the condition of the initial neural network trained for the first time, determining the total sample number of the preset sample data set, and determining the reciprocal of the total sample number as the target sample data weight of each preset sample data.
6. The method as recited in claim 5, further comprising:
under the condition that the target neural network is not the initial neural network for the first training, determining the number of error samples of the target neural network for which the last training is completed;
Determining a network error rate according to the number of error samples and the total number of samples, and determining an absolute value of a difference value between a predicted output result and an actual output result of each preset sample data corresponding to the target neural network after the last training is completed;
Calculating the ratio of the absolute value of the difference value to the corresponding actual output result, and acquiring the historical target sample data weight of the target neural network which is completed by the previous training and corresponding to each preset sample data;
For first preset sample data with the ratio larger than a preset error threshold value, determining that the historical target sample data weight of the first preset sample data is the candidate sample data weight of the first preset sample data;
For second preset sample data with the ratio not larger than a preset error threshold, determining that the product value of the historical target sample data weight of the second preset sample data and the network error rate is the candidate sample data weight of the second preset sample data;
And carrying out normalization processing on the candidate sample data weights of the first preset sample data and the second preset sample data to obtain target sample data weights of the first preset sample data and the second preset sample data.
7. The method of claim 1, wherein the network weight matrix corresponding to the T target neural networks determined according to the regularization result comprises:
wherein, Is a preset constant,/>Is a unitary matrix,/>For the pseudo-inverse matrix,/>Is the output matrix.
8. The method of claim 1, wherein the pre-set sample data set comprises a plurality of types of intrusion parameter sample data for the high-speed rail catenary; the risk prediction of the high-speed railway catenary according to the integrated network comprises the following steps:
acquiring actual values of multiple types of invasion parameters of the high-speed railway catenary;
Inputting the actual values of the multiple intrusion parameters into each target neural network of the integrated network respectively to obtain T initial predicted values;
calculating the product value of the initial predicted value of each target neural network and the corresponding network weight to obtain T product values;
And summing and calculating the T product values to obtain a risk prediction value of the high-speed railway catenary.
9. An intrusion risk prediction device for a high-speed railway catenary, comprising:
the first training module is used for constructing T initial neural networks and training the T initial neural networks according to a preset sample data set to obtain T trained target neural networks, wherein T is a natural number greater than 1;
the second training module is used for respectively inputting the preset sample data sets into each target neural network and obtaining an output matrix according to the output vectors of the T target neural networks;
the weight matrix determining module is used for calculating a pseudo-inverse matrix of the output matrix, regularizing the pseudo-inverse matrix and determining a network weight matrix corresponding to the T target neural networks according to regularized results;
The weight determining module is used for determining the network weight of each target neural network according to the weight vector in the network weight matrix;
and the integrated network construction module is used for constructing an integrated network according to the network weight and the T target neural networks so as to predict the risk of the high-speed railway catenary according to the integrated network.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410332226.5A CN117933499B (en) | 2024-03-22 | 2024-03-22 | Invasion risk prediction method, device and storage medium for high-speed railway catenary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410332226.5A CN117933499B (en) | 2024-03-22 | 2024-03-22 | Invasion risk prediction method, device and storage medium for high-speed railway catenary |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117933499A true CN117933499A (en) | 2024-04-26 |
| CN117933499B CN117933499B (en) | 2024-09-06 |
Family
ID=90751140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410332226.5A Active CN117933499B (en) | 2024-03-22 | 2024-03-22 | Invasion risk prediction method, device and storage medium for high-speed railway catenary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117933499B (en) |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105989408A (en) * | 2015-03-18 | 2016-10-05 | 国际商业机器公司 | A system and a method for mapping a neural network onto a neurosynaptic substrate |
| CN108449295A (en) * | 2018-02-05 | 2018-08-24 | 西安电子科技大学昆山创新研究院 | Combined modulation recognition methods based on RBM networks and BP neural network |
| CN109034387A (en) * | 2018-06-29 | 2018-12-18 | 北京师范大学 | A kind of approximation method for quickly training self-encoding encoder based on pseudo- reversal learning |
| CN112529166A (en) * | 2020-12-25 | 2021-03-19 | 中国科学院西安光学精密机械研究所 | Fusion neuron model, neural network structure, training and reasoning method, storage medium and device |
| CN113158582A (en) * | 2021-05-24 | 2021-07-23 | 苏州大学 | Wind speed prediction method based on complex value forward neural network |
| CN114429246A (en) * | 2022-01-21 | 2022-05-03 | 广东电网有限责任公司 | Power grid potential safety determination method and device based on heterogeneous neural network |
| CN114462597A (en) * | 2022-02-13 | 2022-05-10 | 北京师范大学 | Method, device, equipment and storage medium for quickly training deep neural network |
| CN114662657A (en) * | 2022-03-03 | 2022-06-24 | 中国科学院重庆绿色智能技术研究院 | Polynomial tree process neural network and prediction method thereof |
| CN115081592A (en) * | 2022-06-13 | 2022-09-20 | 华设设计集团股份有限公司 | Road low visibility estimation method based on genetic algorithm and feedforward neural network |
| CN115311355A (en) * | 2022-09-20 | 2022-11-08 | 中国铁建电气化局集团有限公司 | Contact network foreign matter risk early warning method, device, equipment and storage medium |
| CN115809700A (en) * | 2022-06-09 | 2023-03-17 | 电子科技大学 | Spiking neural network learning method based on synapse-threshold synergy |
| CN116758767A (en) * | 2023-08-21 | 2023-09-15 | 中南大学 | Traffic signal lamp control method based on multi-strategy reinforcement learning |
| CN116757465A (en) * | 2023-04-21 | 2023-09-15 | 国网浙江省电力有限公司金华供电公司 | Line risk assessment method and device based on double training weight distribution model |
| US20240046089A1 (en) * | 2022-08-08 | 2024-02-08 | National Chengchi University | Adaptive learning algorithm |
-
2024
- 2024-03-22 CN CN202410332226.5A patent/CN117933499B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105989408A (en) * | 2015-03-18 | 2016-10-05 | 国际商业机器公司 | A system and a method for mapping a neural network onto a neurosynaptic substrate |
| CN108449295A (en) * | 2018-02-05 | 2018-08-24 | 西安电子科技大学昆山创新研究院 | Combined modulation recognition methods based on RBM networks and BP neural network |
| CN109034387A (en) * | 2018-06-29 | 2018-12-18 | 北京师范大学 | A kind of approximation method for quickly training self-encoding encoder based on pseudo- reversal learning |
| WO2022134391A1 (en) * | 2020-12-25 | 2022-06-30 | 中国科学院西安光学精密机械研究所 | Fusion neuron model, neural network structure and training and inference methods therefor, storage medium, and device |
| CN112529166A (en) * | 2020-12-25 | 2021-03-19 | 中国科学院西安光学精密机械研究所 | Fusion neuron model, neural network structure, training and reasoning method, storage medium and device |
| CN113158582A (en) * | 2021-05-24 | 2021-07-23 | 苏州大学 | Wind speed prediction method based on complex value forward neural network |
| CN114429246A (en) * | 2022-01-21 | 2022-05-03 | 广东电网有限责任公司 | Power grid potential safety determination method and device based on heterogeneous neural network |
| CN114462597A (en) * | 2022-02-13 | 2022-05-10 | 北京师范大学 | Method, device, equipment and storage medium for quickly training deep neural network |
| CN114662657A (en) * | 2022-03-03 | 2022-06-24 | 中国科学院重庆绿色智能技术研究院 | Polynomial tree process neural network and prediction method thereof |
| CN115809700A (en) * | 2022-06-09 | 2023-03-17 | 电子科技大学 | Spiking neural network learning method based on synapse-threshold synergy |
| CN115081592A (en) * | 2022-06-13 | 2022-09-20 | 华设设计集团股份有限公司 | Road low visibility estimation method based on genetic algorithm and feedforward neural network |
| US20240046089A1 (en) * | 2022-08-08 | 2024-02-08 | National Chengchi University | Adaptive learning algorithm |
| CN115311355A (en) * | 2022-09-20 | 2022-11-08 | 中国铁建电气化局集团有限公司 | Contact network foreign matter risk early warning method, device, equipment and storage medium |
| CN116757465A (en) * | 2023-04-21 | 2023-09-15 | 国网浙江省电力有限公司金华供电公司 | Line risk assessment method and device based on double training weight distribution model |
| CN116758767A (en) * | 2023-08-21 | 2023-09-15 | 中南大学 | Traffic signal lamp control method based on multi-strategy reinforcement learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117933499B (en) | 2024-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492830B (en) | Mobile pollution source emission concentration prediction method based on time-space deep learning | |
| Chen et al. | Learning deep representation from big and heterogeneous data for traffic accident inference | |
| CN112561191B (en) | Prediction model training method, prediction device, prediction apparatus, prediction program, and program | |
| Basu et al. | Understanding household vehicle ownership in Singapore through a comparison of econometric and machine learning models | |
| CN113469457B (en) | Power transmission line fault probability prediction method integrating attention mechanism | |
| US20210142161A1 (en) | Systems and methods for model-based time series analysis | |
| CN109214604A (en) | The haze calamity risk forecast method and system of power transmission and transforming equipment based on Adaboost | |
| CN114594532B (en) | Cold weather prediction method and device, electronic equipment and computer readable medium | |
| CN115019209A (en) | Method and system for detecting state of electric power tower based on deep learning | |
| CN113516304A (en) | Space-time joint prediction method and device for regional pollutants based on space-time graph network | |
| CN116451862A (en) | Model training method and device and traffic event occurrence probability evaluation method and device | |
| CN116884192A (en) | Power production operation risk early warning method, system and equipment | |
| US20220101273A1 (en) | Methods and systems for forecasting estimated time of restoration during service interruption | |
| CN117933499B (en) | Invasion risk prediction method, device and storage medium for high-speed railway catenary | |
| Shen et al. | Risk response for critical infrastructures with multiple interdependent risks: A scenario-based extended CBR approach | |
| Zhang et al. | Off-deployment traffic estimation—a traffic generative adversarial networks approach | |
| Huo et al. | Traffic anomaly detection method based on improved GRU and EFMS-Kmeans clustering | |
| CN117726162A (en) | Community risk level assessment method and system based on multi-mode data fusion | |
| CN117408598A (en) | Expressway large-piece transportation path rapid decision method based on dynamic road network | |
| Xue et al. | A data aggregation-based spatiotemporal model for rail transit risk path forecasting | |
| CN112529315B (en) | Landslide prediction method, landslide prediction device, landslide prediction equipment and storage medium | |
| CN117010892A (en) | Payment risk detection method, device, electronic equipment and readable medium | |
| KR20230123574A (en) | Transfer learning driven sequential forecasting and ventilation control of PM2.5 associated health risk levels in underground public facilities | |
| CN108985526B (en) | Transportation capacity prediction method and device, computer readable storage medium and terminal | |
| Zhu et al. | Demand estimation and spatio-temporal clustering for urban road networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |