Overturning early warning method and system for overhead working truck
The overturn warning model was established through the support vector machine algorithm, and the overturn risk was judged using the pitch angle, slewing angle and telescopic arm length data of the aerial vehicle, which solved the problem of the lack of early warning system for the aerial vehicle, and achieved high accuracy and efficient overturn warning. , improves the safety of operators.
Patent Information
- Application Number
- CN202410805396.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-21
- Publication Date
- 2024-11-08
AI Technical Summary
The lack of an effective overturn warning system during the work of the high-altitude work vehicle has caused the operator to have limited reaction time after discovering the outrigger off the ground, which poses a major safety hazard.
The overturn warning model is established by using the support vector machine (SVM) algorithm. By collecting data on the pitch angle, slewing angle and telescopic arm length of the aerial vehicle, the optimal hyperplane is constructed and the predicted overturn value is output, combining the probability output value and the preset threshold value. Determine whether the overturn has occurred and issue an alarm.
It improves the accuracy and efficiency of overturned warning of high-altitude work vehicles, ensures that operators adjust or escape in a timely manner, and significantly improves the safety of life and property.
Smart Images

Figure CN118916767A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of aerial work vehicles, and in particular to a method and system for early warning of overturning of an aerial work vehicle. Background Art
[0002] With the advancement of science and technology, aerial construction operations are becoming more and more common. The traditional rope-carrying operation method is not only inefficient, but also highly dangerous during operation. Especially with the increasing height of urban buildings, aerial work vehicles with higher operating heights are the future development direction of aerial work equipment. However, in recent years, aerial work vehicles have frequently overturned, causing significant losses to people's lives and property. At present, the design level of aerial work vehicles cannot guarantee the safety of aerial work vehicles during work, and some overturning incidents often occur; since most work vehicles do not have an early warning system, when the operator finds that the outriggers are off the ground, they have limited time to deal with the work vehicle and escape from the work vehicle, which poses a great safety hazard. In order to solve the above problems, in the prior art, an aerial work vehicle anti-overturning early warning system is proposed, which judges the overturning of the aerial work vehicle by processing the data detected by the sensor through a single-chip microcomputer. However, the prior art does not disclose a specific judgment method, nor does it disclose the specific algorithm steps of the single-chip microcomputer that can judge the overturning of the aerial work vehicle.
[0003] Therefore, there is an urgent need for an aerial work vehicle overturning warning method and system to solve the above problems. Summary of the invention
[0004] Based on the above, the purpose of the present invention is to provide a method and system for early warning of overturning of an aerial work vehicle, which solves the problem of lack of early warning system for aerial work vehicles in the prior art, and has high early warning accuracy.
[0005] To achieve the above object, the present invention adopts the following technical solutions:
[0006] A method for early warning of overturning of an aerial work vehicle, comprising:
[0007] Establish a capsizing warning model, which outputs a predicted capsizing value f according to the input observation data * ;
[0008] Collect the observation data, which includes the pitch angle θ, the rotation angle α, and the telescopic arm length l of the aerial work vehicle, and bring the observation data into the overturning warning model, which outputs the predicted overturning value f * ;
[0009] The predicted overturning value f *It is converted into a probability output value P, and the probability output value P is compared with a preset threshold value R. If P≥R, it is considered that the aerial work vehicle will overturn under the observed data, and the aerial work vehicle will issue an alarm. If P<R, it is considered that the aerial work vehicle will not overturn under the observed data.
[0010] As a preferred solution for an aerial work vehicle overturning warning method, establishing a overturning warning model includes:
[0011] Construct SVM algorithm model;
[0012] The SVM algorithm model is trained by a number of training data, wherein the training data includes input data and a capsizing label y of the input data, the input data includes a pitch angle θ, a rotation angle α, and a telescopic arm length l of the aerial work vehicle, a capsizing label y of 1 indicates that the input data is a capsizing sample, and a capsizing label y of -1 indicates that the input data is a non-capsulating sample. The SVM algorithm model obtains the capsizing warning model through training learning, and the capsizing warning model distinguishes whether the input data is a capsizing sample or a non-capsulating sample according to the input data, and can output a predicted capsizing value f * .
[0013] As a preferred solution for the overturning warning method of aerial work vehicles, the SVM algorithm model is:
[0014] f=ω T x i -b (1)
[0015] Among them, x i is the three-dimensional feature vector of the input data, x i =[θ i ,α i ,l i ], ω is the normal vector of the hyperplane in the SVM algorithm model, and b is the bias;
[0016] The SVM algorithm model is trained by the plurality of training data to obtain the optimal hyperplane of the SVM algorithm model and the ω of the optimal hyperplane. * and b * , and obtain the capsizing warning model:
[0017] f * =ω *T x i -b * (2).
[0018] As a preferred solution for the overturning warning method of aerial work vehicles, the SVM algorithm model adopts a soft-boundary hyperplane selection method and uses two parallel hyperplanes to separate the two categories of data:
[0019] w T xb=1 (3)
[0020] Any sample on or above this boundary is of one class, with label 1, i.e., a positive hyperplane;
[0021] w T xb=-1 (4)
[0022] Any sample on or above this boundary is of one class and has label -1, i.e., a negative hyperplane;
[0023] The distance between a positive hyperplane and a negative hyperplane is:
[0024]
[0025] Introducing the hinge loss function:
[0026] ε i =max(0,1-y i (w T x i -b)) (6)
[0027] where y i is the i-th target category, w T x i -b is the i-th output value, and the optimization goal is:
[0028]
[0029] Among them, C is the penalty factor, which controls the degree of penalty for misclassification, C ≥ 0;
[0030] The constraints are:
[0031] y i (w T x i -b)≥1-ε i (8)
[0032] Construct the Lagrangian function:
[0033]
[0034] In order to save computing resources and obtain higher solution efficiency, it is transformed into a dual problem:
[0035]
[0036] Where α is the Lagrange multiplier vector;
[0037] The constraints that need to be satisfied by the dual problem are:
[0038]
[0039] The optimal Lagrange multiplier α* is calculated and obtained from equations (10) and (11):
[0040]
[0041] The optimal hyperplane ω is calculated by equations (12) and (13): * and b * .
[0042] As a preferred solution of an aerial work vehicle overturning warning method, training the SVM algorithm model through a number of training data also includes:
[0043] Perform Gaussian kernel function transformation on the input data:
[0044]
[0045] Converting the three-dimensional feature vector of the input data into a feature vector higher than three dimensions;
[0046] The best combination of C and γ is selected by searching through a grid or Bayesian optimization algorithm, and each combination of C and γ is tested by cross-validation, and the combination of C and γ with the best cross-validation accuracy is selected.
[0047] As a preferred method for overturning warning of aerial work vehicles, the predicted overturning value f * The conversion into probability output value P includes:
[0048] The predicted overturning value f is converted into * Converted into the probability output value P:
[0049] P(y=1|x)=h(f(x)) (15)
[0050] h is a monotonically increasing function obtained by fitting when the SVM algorithm model is trained using the plurality of training data.
[0051] As a preferred solution of an aerial work vehicle overturning warning method, the preset threshold R is determined based on a cross-validation method, including:
[0052] Dividing the plurality of input data into n subsets;
[0053] Each time, np subsets are used as training subsets and the remaining p subsets are used as validation subsets to train the SVM model, and the true positive rate and false positive rate corresponding to each training are calculated. The training is repeated n times to obtain the ROC curve, and the probability output value P that achieves the best balance between the true positive rate and the false positive rate is selected according to the ROC curve as the preset threshold R.
[0054] As a preferred solution for an aerial work vehicle rollover warning method, the sum of the true positive rate and the false positive rate is 1. When the best balance is achieved between the true positive rate and the false positive rate, the true positive rate ranges from 0.95 to 0.99.
[0055] A tipping system for an aerial work vehicle, comprising:
[0056] The signal acquisition module is used to collect the pitch angle θ, rotation angle α and telescopic arm length l of the aerial work vehicle;
[0057] A processing module is used to receive the pitch angle θ, the rotation angle α and the telescopic arm length l data collected by the signal collection module, and calculate the probability output value P through a preset algorithm to determine whether the aerial work vehicle will overturn;
[0058] An alarm module is connected to the processing module, and is used to issue an alarm according to the prediction result of whether capsizing occurs.
[0059] As a preferred solution for the overturning warning system of an aerial work vehicle, it also includes:
[0060] The storage module is used to store the data received and calculated by the processing module.
[0061] The beneficial effects of the present invention are:
[0062] The present invention provides a method for overturning warning of aerial work vehicles. By establishing a overturning warning model, the pitch angle θ, the rotation angle α, and the telescopic arm length l of the aerial work vehicle collected are brought into the overturning warning model, and the overturning warning model can output a predicted overturning value f * , and then the predicted capsizing value f * The probability output value P is converted into a probability output value P. By comparing the probability output value P with the preset threshold value, it can be judged whether the aerial work vehicle will overturn, and the judgment accuracy is relatively high. The alarm issued by the aerial work vehicle reminds the operator, so that the operator can adjust the aerial work vehicle or escape from the aerial work vehicle in time, improving the life safety of the operator and the property safety of the aerial work vehicle. The overturning warning method of the aerial work vehicle is highly intelligent and automated, with good warning accuracy, which can greatly improve the efficiency and reliability of overturning response.
[0063] The present invention also provides an aerial work vehicle overturning warning system, which includes a signal acquisition module, a processing module and an alarm module. The processing module calculates a probability output value P according to the received pitch angle θ, rotation angle α and telescopic arm length l of the aerial work vehicle, and then determines whether the aerial work vehicle will overturn. The efficiency and reliability of overturning response can be greatly improved, and the warning accuracy is better, and the intelligence and automation are higher. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.
[0065] Figure 1 It is a principle diagram of an aerial work vehicle overturning warning method provided by an embodiment of the present invention;
[0066] Figure 2 Schematic diagram of an aerial work vehicle rollover warning system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0067] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.
[0068] like Figure 1 As shown, this embodiment provides a method for early warning of overturning of an aerial work vehicle, and the method for early warning of overturning of an aerial work vehicle comprises the following steps:
[0069] S100: Establishing a capsizing warning model, which outputs a predicted capsizing value f according to the input observation data * ;
[0070] S200: Collect observation data, including the pitch angle θ, the rotation angle α, and the telescopic arm length l of the aerial work vehicle, and bring the observation data into the overturning warning model, which outputs the predicted overturning value f * ;
[0071] Among them, the pitch angle sensor, rotation angle sensor, and telescopic arm length sensor installed on the aerial work vehicle respectively detect and collect the pitch angle θ, rotation angle α, and telescopic arm length l of the aerial work vehicle. The predicted overturning value f output by the overturning warning model *, the overturning state of the aerial work vehicle can be preliminarily determined, f * If it is greater than 0, it means that the input observation data is an overturned sample, f * If it is less than 0, it means that the input observation data is a non-overturned sample.
[0072] S300: The predicted capsizing value f * It is converted into a probability output value P, and the probability output value P is compared with the preset threshold value R. If P≥R, it is considered that the aerial work vehicle will overturn under the observed data, and the aerial work vehicle will issue an alarm. If P<R, it is considered that the aerial work vehicle will not overturn under the observed data.
[0073] By setting the preset threshold, the accuracy of the final judgment result meets the expected accuracy. That is, by setting the preset threshold, the tolerance for misjudgment can be effectively controlled, and the accuracy of the aerial work vehicle overturning warning can be improved. In order to facilitate comparison with the preset threshold, the predicted overturning value f * It is converted into a probability output value P. By comparing the probability output value P with the preset threshold R, it can be accurately determined whether the aerial work vehicle will overturn.
[0074] If the aerial work vehicle is predicted to overturn, the aerial work vehicle will send out an alarm, and the alarm sent by the aerial work vehicle will remind the operator, so that the operator can adjust the aerial work vehicle or escape from the aerial work vehicle in time, thereby improving the life safety of the operator and the property safety of the aerial work vehicle. The overturning warning method of the aerial work vehicle is highly intelligent and automated, with good warning accuracy, which can greatly improve the efficiency and reliability of overturning response.
[0075] Specifically, in step S100, establishing the overturning warning model specifically includes the following steps:
[0076] S101: Construct an SVM algorithm model; the SVM (Support Vector Machine) algorithm is applicable to binary classification problems. In this embodiment, the judgment problem of early warning of overturning of aerial work vehicles is transformed into a binary classification problem. The overturning early warning model based on the SVM algorithm model aims to find the optimal hyperplane for dividing the state feature space of the aerial work vehicle.
[0077] S102: The SVM algorithm model is trained by using a number of training data, wherein the training data includes input data and a capsizing label y of the input data, the input data includes the pitch angle θ, the rotation angle α, and the telescopic arm length l of the aerial work vehicle, the capsizing label y is 1, indicating that the input data is a capsizing sample, and the capsizing label y is -1, indicating that the input data is a non-capsulating sample. The SVM algorithm model obtains a capsizing warning model through training and learning, and the capsizing warning model distinguishes whether the input data is a capsizing sample or a non-capsulating sample according to the input data, and can output a predicted capsizing value f* .
[0078] Among them, the SVM algorithm model is:
[0079] f=ω T x i -b (1)
[0080] ω is the normal vector of the hyperplane in the SVM algorithm model, and b is the bias;
[0081] When training the SVM algorithm model, given the training data (x1, y1),,,,, (x i ,y i ),,,,,,(x m ,y m ), x i is the three-dimensional feature vector of the input data, x i =[θ i ,α i ,l i ];y i The label representing the i-th specimen belongs to the set {-1, 1}. The SVM algorithm model is trained through a number of training data to obtain the optimal hyperplane of the SVM algorithm model and the ω of the optimal hyperplane. * and b * , and obtain the capsizing warning model:
[0082] f * =ω *T x i -b * (2)
[0083] The optimal hyperplane of the SVM algorithm model can be obtained, which can be used to distinguish the observed data. If the output f * If it is greater than 0, it means that the input observation data is an overturned sample, f * If it is less than 0, it means that the input observation data is a non-overturned sample.
[0084] In this embodiment, when solving the optimal hyperplane ω * and b * The specific steps include:
[0085] The SVM algorithm model uses a soft-margin hyperplane selection method and uses two parallel hyperplanes to separate the two categories of data:
[0086] w T xb=1 (3)
[0087] Any sample on or above this boundary is of one class, with label 1, i.e., a positive hyperplane;
[0088] wT xb=-1 (4)
[0089] Any sample on or above this boundary is of one class and has label -1, i.e., a negative hyperplane;
[0090] The distance between a positive hyperplane and a negative hyperplane is:
[0091]
[0092] Introducing the hinge loss function:
[0093] ε i =max(0,1-y i (w T x i -b)) (6)
[0094] where y i is the i-th target category, w T x i -b is the i-th output value, and the optimization goal is:
[0095]
[0096] Among them, C is the penalty factor, which controls the degree of penalty for misclassification, C ≥ 0;
[0097] The constraints are:
[0098] y i (w T x i -b)≥1-ε i (8)
[0099] Construct the Lagrangian function:
[0100]
[0101] In order to save computing resources and obtain higher solution efficiency, it is transformed into a dual problem:
[0102]
[0103] Where α is the Lagrange multiplier vector;
[0104] The constraints that need to be satisfied by the dual problem are:
[0105]
[0106] The optimal Lagrange multiplier α* is calculated, and:
[0107]
[0108] The optimal hyperplane ω is calculated by equations (12) and (13): * and b * .
[0109] In order to improve the accuracy of the capsizing warning model, the input data is transformed into a Gaussian kernel function:
[0110]
[0111] Here, "γ" represents the scale parameter, also known as bandwidth. It determines the "shape" and "width" of the Gaussian kernel function.
[0112] The three-dimensional feature vector of the input data is converted into a feature vector higher than three dimensions; the Gaussian kernel function maps the original input data to a higher-dimensional feature space by calculating the similarity between the input data. In the new feature space, the originally linearly inseparable input data becomes linearly separable, which is beneficial to improving the calculation accuracy of the overturning warning model.
[0113] The best combination of penalty factors C and γ is selected by searching through a grid or Bayesian optimization algorithm, and each combination of penalty factors C and γ is detected through cross-validation, and the combination of penalty factors C and γ with the best cross-validation accuracy is selected. Among them, the grid or Bayesian optimization algorithm is a relatively mature existing technology, which will not be described one by one here. By selecting the best combination of penalty factors C and γ, the distance L between the positive hyperplane and the negative hyperplane is maximized, that is, the decision hyperplane is located as much as possible in the middle of the positive hyperplane and the negative hyperplane. If the distance L between the positive hyperplane and the negative hyperplane is small, the decision hyperplane may be biased to one side of the positive hyperplane and the negative hyperplane, resulting in misclassification of the data. By selecting the best combination of penalty factors C and γ, the classification ability of the input data is improved, while ensuring the minimum classification error, thereby improving the prediction reliability and prediction accuracy of the capsizing warning model.
[0114] Further, in step S300, the predicted capsizing value f * The conversion into probability output value P includes:
[0115] S301: Use IsotonicRegression function to predict the overturning value f * Converted into probability output value P:
[0116] P(y=1|x)=h(f(x)) (15)
[0117] h is a monotonically increasing function obtained by fitting the SVM algorithm model using a number of training data.
[0118] When training the SVM algorithm model, f(x) is used as the X-axis and its corresponding probability output value P is used as the Y-axis for linear fitting. Exemplarily, the probability output value P corresponding to f(x)=0 is a given value, such as 0.5-0.6. For example, the probability output value P corresponding to f(x)=0 is 0.6. When the f(x) calculated from a certain input data is greater than 0, the Y-axis coordinate corresponding to the point is greater than 0.6. The larger the f(x), the larger the corresponding Y-axis coordinate. When the f(x) calculated from the input data is less than 0, the Y-axis coordinate corresponding to the point is less than 0.6. The smaller the f(x), the smaller the corresponding Y-axis coordinate. By fitting, a monotonically increasing function h about f(x) and the probability output value P is obtained. When predicting the overturning state of an aerial work vehicle, the observed data is brought into the overturning warning model, and then the f output by the overturning warning model is converted into the Y-axis coordinate. * Substituting into formula (15) we can get the corresponding probability output value P.
[0119] Furthermore, the preset threshold R is determined based on a cross-validation method, specifically including:
[0120] A number of input data are divided into n subsets; for example, 100 input data are divided into 200 subsets, each subset includes one or more input data, for example, one subset includes 3 or 5 input data. The number of input data is not limited to 100, but can be other, and the specific number of subsets is not limited to 200, which is set according to actual needs and is not limited here.
[0121] Each time, np subsets are used as training subsets, and the remaining p subsets are used as validation subsets to train the SVM model, and the true positive rate and false positive rate corresponding to each training are calculated. The training is repeated n times to obtain the ROC curve, and the probability output value P under the optimal balance between the true positive rate and the false positive rate is selected according to the ROC curve as the preset threshold R. Among them, p can be selected but not limited to 1, for example, p is 2, 3, 4, etc. Exemplarily, p is 1, and 199 subsets are used as training subsets each time, and the remaining one subset is used as the validation subset. When 199 training subsets are used for training each time, the corresponding true positive rate and false positive rate are obtained, and the training is repeated 200 times to obtain the ROC curve about the true positive rate and the false positive rate. When the best balance between the true positive rate and the false positive rate is selected, the corresponding probability output value P is the preset threshold R. The misjudgment rate of the capsizing warning model finally obtained is within the tolerance range, that is, the judgment accuracy is within the preset range.
[0122] In this embodiment, the sum of the true positive rate and the false positive rate is 1, and when the true positive rate and the false positive rate are optimally balanced, the true positive rate ranges from 0.95 to 0.99. For example, when the true positive rate and the false positive rate are optimally balanced, the true positive rate is 0.95, 0.96, 0.97, 0.98 or 0.99, etc., and the corresponding false positive rates are 0.05, 0.04, 0.03, 0.02, 0.01, etc.
[0123] By determining the optimal hyperplane, the monotonic linear relationship between the probability output value P and the model output value f(x), and the preset threshold R when training the SVM algorithm model, the overturning state of the aerial work vehicle can be predicted with high prediction accuracy.
[0124] like Figure 2 As shown, this embodiment also provides an aerial work vehicle overturning warning system, which includes a signal acquisition module, a processing module and an alarm module. The signal acquisition module is used to collect the pitch angle θ, the rotation angle α and the telescopic arm length l of the aerial work vehicle; the processing module is used to receive the pitch angle θ, the rotation angle α and the telescopic arm length l data collected by the signal acquisition module, and calculate the probability output value P through a preset algorithm to determine whether the aerial work vehicle will overturn; the alarm module is connected to the processing module, and the alarm module is used to issue an alarm based on the prediction result of whether the overturning will occur. Among them, the preset algorithm includes two parts, one part is the overturning warning model outputs the predicted overturning value f according to the observed data * The other part is to predict the capsizing value f * Converted into a probability output value P. Compare the probability output value P with the preset threshold R. If P ≥ R, it is considered that the aerial work vehicle will overturn under the observed data, and the alarm module issues an alarm. If P < R, it is considered that the aerial work vehicle will not overturn under the observed data.
[0125] Specifically, the signal acquisition module includes a pitch angle sensor, a swivel angle sensor, and a telescopic arm length sensor. The pitch angle sensor, the swivel angle sensor, and the telescopic arm length sensor are all connected to the processing module. The processing module receives the pitch angle, swivel angle, and telescopic arm length detected by the pitch angle sensor, the swivel angle sensor, and the telescopic arm length sensor.
[0126] More specifically, the alarm module includes a visual alarm module and an audio alarm module. The visual alarm module may be, but is not limited to, an alarm light, and the audio alarm module may be, but is not limited to, an alarm horn. By using a combination of visual and audio signals, a strong warning effect is provided, which can immediately attract the attention of personnel and prompt them to take corresponding measures. Preferably, the alarm module also includes a controller, which is connected to the processing module, the visual alarm module, and the audio alarm module, and the controller is used to control the activation of the visual alarm module and the audio alarm module according to the judgment result of the processing module.
[0127] Preferably, the aerial work vehicle overturning warning system also includes a storage module, which is used to store data received and calculated by the processing module. The storage module is a non-volatile storage medium, and the storage module can keep the data from being lost after power failure.
[0128] The processing module calculates the probability output value P based on the received pitch angle θ, rotation angle α and telescopic arm length l of the aerial work vehicle, and then determines whether the aerial work vehicle will overturn. This can greatly improve the efficiency and reliability of overturning response, and has better early warning accuracy, higher intelligence and automation.
[0129] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims
1. A method for early warning of overturning of an aerial work vehicle, characterized in that: include: Establish a capsizing warning model, which outputs a predicted capsizing value f according to the input observation data * ; Collect the observation data, which includes the pitch angle θ, the rotation angle α, and the telescopic arm length l of the aerial work vehicle, and bring the observation data into the overturning warning model, which outputs the predicted overturning value f * ; The predicted overturning value f * It is converted into a probability output value P, and the probability output value P is compared with a preset threshold value R. If P≥R, it is considered that the aerial work vehicle will overturn under the observed data, and the aerial work vehicle will issue an alarm. If P<R, it is considered that the aerial work vehicle will not overturn under the observed data.
2. The method for early warning of overturning of an aerial work vehicle according to claim 1, characterized in that: The establishment of a capsize warning model includes: Construct SVM algorithm model; The SVM algorithm model is trained by a number of training data, wherein the training data includes input data and a capsizing label y of the input data, the input data includes a pitch angle θ, a rotation angle α, and a telescopic arm length l of the aerial work vehicle, a capsizing label y of 1 indicates that the input data is a capsizing sample, and a capsizing label y of -1 indicates that the input data is a non-capsulating sample. The SVM algorithm model obtains the capsizing warning model through training learning, and the capsizing warning model distinguishes whether the input data is a capsizing sample or a non-capsulating sample according to the input data, and can output a predicted capsizing value f * .
3. The method for early warning of overturning of an aerial work vehicle according to claim 2, characterized in that: The SVM algorithm model is: f=ω T x i -b (1) Among them, x i is the three-dimensional feature vector of the input data, x i =[θ i ,α i ,l i ], ω is the normal vector of the hyperplane in the SVM algorithm model, and b is the bias; The SVM algorithm model is trained by the plurality of training data to obtain the optimal hyperplane of the SVM algorithm model and the ω of the optimal hyperplane. * and b * , and obtain the capsizing warning model: f * =ω *T x i -b * (2)。 4. The method for early warning of overturning of an aerial work vehicle according to claim 3, characterized in that: The SVM algorithm model uses a soft-margin hyperplane selection method and uses two parallel hyperplanes to separate the two categories of data: w T x-b=1 (3) Any sample on or above this boundary is of one class, with label 1, i.e., a positive hyperplane; w T x-b=-1 (4) Any sample on or above this boundary is of one class and has label -1, i.e., a negative hyperplane; The distance between a positive hyperplane and a negative hyperplane is: Introducing the hinge loss function: ε i =max(0,1-y i (w T x i -b)) (6) where y i is the i-th target category, w T x i -b is the i-th output value, and the optimization goal is: Among them, C is the penalty factor, which controls the degree of penalty for misclassification, C ≥ 0; The constraints are: y i (w T x i -b)≥1-ε i (8) Construct the Lagrangian function: In order to save computing resources and obtain higher solution efficiency, it is transformed into a dual problem: Where α is the Lagrange multiplier vector; The constraints that need to be satisfied by the dual problem are: The optimal Lagrange multiplier α* is calculated and obtained from equations (10) and (11): The optimal hyperplane ω is calculated by equations (12) and (13): * and b * .
5. The method for early warning of overturning of an aerial work vehicle according to claim 4, characterized in that: Training the SVM algorithm model through a number of training data also includes: Perform Gaussian kernel function transformation on the input data: Converting the three-dimensional feature vector of the input data into a feature vector higher than three dimensions; The best combination of C and γ is selected by searching through a grid or Bayesian optimization algorithm, and each combination of C and γ is tested by cross-validation, and the combination of C and γ with the best cross-validation accuracy is selected.
6. The method for early warning of overturning of an aerial work vehicle according to claim 2, characterized in that: The predicted capsizing value f * The conversion into probability output value P includes: The predicted overturning value f is converted into * Converted into the probability output value P: P(y=1|x)=h(f(x)) (15) h is a monotonically increasing function obtained by fitting when the SVM algorithm model is trained using the plurality of training data.
7. The method for early warning of overturning of an aerial work vehicle according to claim 6, characterized in that: The preset threshold R is determined based on a cross-validation method, including: Dividing the plurality of input data into n subsets; Each time, np subsets are used as training subsets and the remaining p subsets are used as validation subsets to train the SVM model, and the true positive rate and false positive rate corresponding to each training are calculated. The training is repeated n times to obtain the ROC curve, and the probability output value P that achieves the best balance between the true positive rate and the false positive rate is selected according to the ROC curve as the preset threshold R.
8. The method for early warning of overturning of an aerial work vehicle according to claim 7, characterized in that: The sum of the true positive rate and the false positive rate is 1. When the best balance is achieved between the true positive rate and the false positive rate, the true positive rate ranges from 0.95 to 0.
99.
9. A tipping warning system for aerial work vehicles, characterized in that: include: The signal acquisition module is used to collect the pitch angle θ, rotation angle α and telescopic arm length l of the aerial work vehicle; A processing module is used to receive the pitch angle θ, the rotation angle α and the telescopic arm length l data collected by the signal collection module, and calculate the probability output value P through a preset algorithm to determine whether the aerial work vehicle will overturn; An alarm module is connected to the processing module, and is used to issue an alarm according to the prediction result of whether capsizing occurs.
10. The overturning warning system for aerial work vehicles according to claim 9, characterized in that: Also includes: The storage module is used to store the data received and calculated by the processing module.
Citation Information
Cited By
Dynamic stable overturn-preventing device for intelligent inspection robot dog
CN119690116A
Safety early warning method and system for overhead working truck and storage medium
CN121158706A
