Safe driving system of electric tricycle
By integrating vehicle weight, turning angle and stability detection modules on electric tricycles, and combining pre-trained neural network models to generate vehicle speed thresholds, real-time monitoring and prevention of electric tricycles rollover risks is achieved, and the occurrence of electric tricycles rollover accidents is solved. There are technical problems, which improves driving safety and stability.
Patent Information
- Application Number
- CN202411011232.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Electric tricycles are prone to overturning during sharp turns, high-speed driving or heavy loading. The existing technical means cannot effectively prevent overturning accidents, and increasing the body stability control system will interfere with the driver's operation and reduce driving comfort.
The electric tricycle safety driving system adopts a vehicle weight detection module, a turning angle detection module, a stability detection module and a pre-trained tricycle anti-rolling neural network model. By monitoring vehicle speed, weight, turning angle and stability data in real time, the vehicle speed is generated. Threshold, intelligently controls the acceleration and deceleration of the tricycle to prevent rollover.
It improves the driving safety and stability of tricycles, enhances the ability to predict and prevent rollover risks, simplifies hardware design, reduces system costs, and improves user experience and driving safety.
Smart Images

Figure CN119002474A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of safe driving of tricycles, and in particular to a safe driving system for an electric tricycle. Background Art
[0002] As an energy-saving and environmentally friendly means of transportation, electric tricycles have been widely used in recent years. Especially in the field of urban short-distance transportation and logistics distribution, electric tricycles have been favored by the market due to their flexibility, convenience, low noise and zero emissions. However, with the increase in the number of electric tricycles, related traffic safety issues have become increasingly prominent. In particular, due to the high center of gravity, narrow body and simple structure of electric tricycles, rollover accidents are very likely to occur when turning, seriously threatening the safety of drivers and pedestrians. Therefore, it is of great practical significance to research and develop a safe driving system that can effectively prevent electric tricycles from rolling over.
[0003] The rollover of electric tricycles mainly occurs when making sharp turns, driving at high speeds or when heavily loaded. The main reasons include: electric tricycles are usually designed with a lightweight structure and a high center of gravity. When the center of gravity height exceeds a certain limit, it is easy to roll over when making sharp turns; when electric tricycles make sharp turns, the centrifugal force acts on them more, especially when driving at high speeds and with heavy loads. The centrifugal force causes the body to tilt, increasing the risk of rollover; when driving at high speeds, due to inertia, electric tricycles are more likely to lose balance when turning, leading to rollover accidents; the increase in load will not only increase the center of gravity of the body, but also increase the inertia force when turning, further exacerbating the risk of rollover; some drivers lack safe driving knowledge and fail to slow down when turning or turn too quickly, causing the vehicle to lose control and roll over.
[0004] At present, in order to prevent electric tricycles from rolling over, some technical means are adopted in the market, such as strengthening the frame structure, adding a body stability control system, etc. However, these technical means have the following limitations: Although strengthening the frame structure can improve the stability of the body to a certain extent, it cannot fundamentally solve the rollover problem, especially in sharp turns and heavy loads, the improvement effect of the body structure is limited; adding a body stability control system is a passive protection measure, which only works when the vehicle is tilted or has a tendency to roll over, and cannot prevent the occurrence of rollover accidents in advance; the use of high-strength materials and complex body structure design will significantly increase the vehicle manufacturing cost, which is not conducive to the popularization and promotion of electric tricycles;
[0005] Some stability control systems may interfere with the driver's normal operations when working, reducing driving comfort and user experience.
[0006] In summary, existing technical means have certain limitations in preventing electric tricycles from rolling over, and a new technical solution is urgently needed to improve the driving safety and stability of electric tricycles. Summary of the invention
[0007] The present invention provides a safe driving system for an electric tricycle, which can solve the above problems.
[0008] In order to solve the above problems, the technical solution adopted by the present invention is as follows:
[0009] The present invention provides a safe driving system for an electric tricycle, comprising a vehicle weight detection module, a turning angle detection module, a stability detection module, a vehicle speed control module and a pre-trained tricycle rollover prevention neural network model;
[0010] The vehicle weight detection module is used to detect the weight data of the tricycle when it is loaded;
[0011] The turning angle detection module is used to detect the real-time turning angle data of the tricycle;
[0012] The stability detection module is used to detect the stability data of the tricycle;
[0013] The pre-trained tricycle rollover prevention neural network model is used to generate a vehicle speed threshold based on vehicle speed, height of the tricycle's center of gravity, total weight of the tricycle when loaded, real-time turning angle data of the tricycle, and stability data of the tricycle, wherein the total weight of the tricycle when loaded is obtained based on the detected weight data and the tricycle's own weight;
[0014] The vehicle speed control module is used to limit the speed of the three-wheeled vehicle based on a vehicle speed threshold.
[0015] As a further description of the above technical solution: the vehicle weight detection module includes a weighing sensor installed on the bearing point of the tricycle chassis, which is used to detect the weight data of the tricycle when loaded; the turning angle detection module includes a gyroscope sensor installed on the frame of the tricycle, which is used to detect the real-time turning angle data of the tricycle; the stability detection module includes an accelerometer and a tilt sensor installed on the frame or chassis of the tricycle, which is used to detect the stability data of the tricycle; the pre-trained tricycle anti-rollover neural network model is deployed in the central processing unit of the electric tricycle; the detected weight data, real-time turning angle data and stability data are transmitted to the central processing unit of the electric tricycle.
[0016] As a further description of the above technical solution: the vehicle speed control module includes an electronic control unit and an actuator. When the three wheels turn, the vehicle speed control module can adjust the motor output of the three-wheeled vehicle based on the vehicle speed threshold, thereby controlling the vehicle speed.
[0017] As a further description of the above technical solution: the tricycle anti-rollover neural network model includes an input layer, a hidden layer and an output layer; the input layer is used to receive the vehicle speed, the height of the center of gravity of the tricycle, the total weight of the tricycle when loaded, the real-time turning angle of the tricycle and the stability of the tricycle; the hidden layer adopts a multi-layer perceptron structure, and each layer of the hidden layer uses a ReLU activation function; the output layer is a single neuron layer, which uses a linear activation function to output a vehicle speed threshold.
[0018] As a further description of the above technical solution: there are two hidden layers; the input vector is linearly transformed and bias adjusted, and then passes through the ReLU activation function to obtain the output f1(x) of the first hidden layer: the output of the first hidden layer is linearly transformed and bias adjusted, and then passes through the ReLU activation function to obtain the output f2(f1(x)) of the second hidden layer.
[0019] As a further description of the above technical solution: the vehicle speed threshold v max The calculation formula is:
[0020] v max =W3·f2(f1(x))+b3
[0021] Among them, W3 is the weight matrix of the output layer, and b3 is the bias of the output layer.
[0022] As a further description of the above technical solution: During the pre-training process of the tricycle anti-rollover neural network model, the gradient descent method is used to minimize the loss function and update the model parameters. The loss function used takes into account the rollover risk, overload risk and stability risk.
[0023] Compared with the prior art, the present invention has the following beneficial effects:
[0024] 1) Through real-time monitoring of the vehicle weight detection module, turning angle detection module and stability detection module, it can respond in time when the tricycle is at risk of rollover. The anti-rollover neural network model combines multiple parameters to generate a speed threshold in real time to ensure that the tricycle travels within a safe speed, thereby greatly improving the driving safety of the tricycle.
[0025] 2) The detection accuracy and reliability of the system are improved through precise sensor data and scientific neural network models.
[0026] 3) The system centralizes the data of each sensor module into a central processing unit for processing, which simplifies the hardware design of the system, improves data processing efficiency, centrally manages hardware and software, makes system debugging and maintenance more convenient, and reduces system costs.
[0027] 4) The anti-rollover neural network model takes into account multiple parameters such as vehicle speed, height of the tricycle's center of gravity, load, turning angle, and stability, and comprehensively evaluates the stability and rollover risk of the tricycle. This can more accurately predict and prevent rollover accidents and improve the safety and stability of the system.
[0028] 5) The installation position of each sensor module has been carefully designed to ensure accurate measurement of relevant data. At the same time, the installation is simple and the centralized data processing method makes the system debugging and maintenance easier and faster.
[0029] 6) The speed control module intelligently controls the acceleration and deceleration of the tricycle based on the speed threshold generated by the anti-rollover neural network model, so that the tricycle is always below the rollover speed when turning, avoiding rollover accidents caused by improper operation of the driver. Through intelligent control methods, the user's driving experience and sense of security are improved.
[0030] 7) The system can dynamically adjust the anti-rollover model parameters and vehicle speed thresholds according to different loads and driver conditions to adapt to different driving conditions and environments, and provide a stable and safe driving experience regardless of whether the vehicle is empty or loaded.
[0031] 8) The present invention adopts a tricycle anti-rollover neural network model with a multi-layer perceptron (MLP) structure. Through the combination of multiple hidden layers and activation functions, it can effectively handle complex nonlinear relationships. For the anti-rollover problem of electric tricycles, the multi-layer perceptron can maintain low computational complexity and resource consumption while ensuring model performance; the input layer receives multi-dimensional features such as vehicle speed, center of gravity height, load, turning angle and stability. Through the layer-by-layer transmission and processing of the hidden layer, these features can be automatically extracted and combined to form an effective representation of the anti-rollover risk. The ReLU (Rectified Linear Unit) activation function is used in the hidden layer. The ReLU activation function has good performance in handling nonlinear problems, can effectively avoid the gradient vanishing problem, and ensure the rapid convergence and stability of the network during the training process; the output layer is a single neuron layer, using a linear activation function to directly output the vehicle speed threshold. This design can ensure the continuity and stability of the output, making the vehicle speed control more accurate.
[0032] 9) The loss function used in the training model comprehensively considers the rollover risk, overload risk and stability risk. The loss function not only considers the traditional prediction error, but also introduces a variety of risk assessment factors to ensure that the model can comprehensively consider the different aspects of the rollover risk and improve the overall safety and reliability; by setting different weight parameters, the influence of each risk factor in the loss function can be flexibly adjusted to ensure that the model performs well under various driving conditions and avoid a single factor dominating the training process; the loss function focuses on reducing overfitting during the training process and improving the generalization ability of the model in different environments and conditions. This can ensure that the model has stable performance in practical applications and reduce performance degradation caused by insufficient data.
[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the embodiments of the present invention are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0035] Figure 1 is a schematic block diagram of the structure of the electric tricycle safe driving system according to an embodiment of the present invention;
[0036] Figure 2 It is a flow chart of the training method of the tricycle anti-rollover neural network model. DETAILED DESCRIPTION
[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.
[0038] Please refer to Figure 1 An embodiment of the present invention provides a safe driving system for an electric tricycle, comprising a vehicle weight detection module, a turning angle detection module, a stability detection module, a vehicle speed control module, and a pre-trained tricycle anti-rollover neural network model. The central processing unit of the electric tricycle is integrated with a microprocessor with a built-in analog-to-digital converter (ADC), and the pre-trained tricycle anti-rollover neural network model is deployed on a storage medium integrated in the central processing unit. The microprocessor can load the pre-trained tricycle anti-rollover neural network model from the storage medium.
[0039] The vehicle weight detection module is used to detect the weight data of the tricycle when it is loaded. The vehicle weight detection module includes a weighing sensor and a first signal conditioning module. In this embodiment, the weighing sensor is installed at the bearing point of the tricycle chassis. A strain gauge sensor is selected. The strain gauge is used to measure the mechanical deformation caused by weight, which has high precision and stability. The first signal conditioning module includes an amplifier and a filter. The amplifier is used to amplify the weak electrical signal output by the weighing sensor, and the filter is used to filter out the noise in the signal. The amplified and filtered signal is sent to the central processing unit of the driving system of the electric tricycle via a cable or wireless transmission.
[0040] The turning angle detection module includes a gyro sensor and a second signal conditioning module. The gyro sensor can measure the angle change of the tricycle when turning and is installed on the frame of the tricycle. The second signal conditioning module includes an amplifier and a filter. The amplifier is used to amplify the electrical signal output by the gyro sensor, and the filter is used to filter out the noise in the signal. The amplified and filtered signal is sent to the central processing unit of the driving system of the electric tricycle for processing and analysis.
[0041] The stability detection module includes an accelerometer, a tilt sensor and a third signal conditioning module. The accelerometer and the tilt sensor are used to detect the stability and tilt state of the vehicle and are installed on the frame or chassis of the tricycle to detect the dynamic state of the vehicle. The third signal conditioning module includes an amplifier and a filter, wherein the amplifier is used to amplify the electrical signal output by the accelerometer and the tilt sensor, and the filter is used to filter out the noise in the signal. The stability data signal after amplification and filtering is sent to the central processing unit of the driving system of the electric tricycle for processing and analysis.
[0042] The detected weight data, real-time turning angle data and stability data are transmitted to the microprocessor of the central processing unit. The microprocessor uses an analog-to-digital converter (ADC) to convert the analog signal of the sensor into a digital signal, and obtains the total weight of the three-wheeled vehicle when loaded based on the detected weight data and the weight of the three-wheeled vehicle.
[0043] The microprocessor loads a pre-trained tricycle anti-rollover neural network model from a storage medium. The tricycle anti-rollover neural network model is used to generate a vehicle speed threshold based on vehicle speed, the height of the center of gravity of the tricycle, the total weight of the tricycle when loaded, the real-time turning angle of the tricycle, and the stability of the tricycle.
[0044] The tricycle rollover prevention neural network model includes an input layer, a hidden layer, and an output layer; the input layer is used to receive the vehicle speed, the height of the tricycle's center of gravity, the total weight of the tricycle when loaded, the real-time turning angle of the tricycle, and the stability of the tricycle; the hidden layer adopts a multi-layer perceptron structure, and each layer of the hidden layer uses a ReLU activation function; the output layer is a single neuron layer, which uses a linear activation function to output the vehicle speed threshold. The central processing unit sends a speed control instruction to the speed control module according to the speed threshold.
[0045] In this embodiment, the hidden layer of the tricycle anti-rollover neural network model has two layers, and the number of neurons in each layer is N1 and N2 respectively. After the input vector x is linearly transformed and offset adjusted, it is activated by the ReLU function to obtain the output f1(x) of the first hidden layer:
[0046] f1(x)=ReLU(W1·x+b1)
[0047] The output f1(x) of the first hidden layer is linearly transformed and bias-adjusted, and then passes through the ReLU activation function to obtain the output f2(f1(x)) of the second hidden layer:
[0048] f2(f1(x))=ReLU(W2·f1(x)+b2)
[0049] Where x is the input vector, x = [v, h, W, θ, φ T , v is the current real-time speed of the tricycle, h is the height of the center of gravity of the tricycle, W is the total weight of the tricycle and its load, θ is the current real-time turning angle of the tricycle, and φ is the stability of the tricycle; W1 is the weight matrix of the first hidden layer, with a size of N1×5, b1 is the bias vector of the first hidden layer, with a size of N1; W2 is the weight matrix of the second hidden layer, with a size of N2×N1, and b2 is the bias vector of the second hidden layer, with a size of N2.
[0050] Speed threshold v max The calculation formula is:
[0051] v max =W3·f2(f1(x))+b3
[0052] Among them, W3 is the weight matrix of the output layer, the size is 1×N2, and b3 is the bias of the output layer, the size is 1.
[0053] Please refer to Figure 2 ,The training method of the tricycle anti-rollover neural network model includes the following steps:
[0054] S 1. Collect the turning driving history data of tricycles in several scenarios. The turning driving history data of tricycles in each scenario includes the vehicle speed v, the height of the center of gravity of the tricycle h, the total weight of the tricycle when loaded W, the real-time turning angle θ of the tricycle, the stability φ of the tricycle and the actual safe speed threshold v actual_max .
[0055] S2. Preprocess the collected data, including data cleaning, standardization, and splitting the training set and test set. Standardization is to standardize each input parameter so that its mean is 0 and its standard deviation is 1.
[0056] S3. Define the loss function
[0057] The loss function used in model training takes into account mean square error (MSE), weight decay (L2 regularization), stability penalty, and turning angle penalty. Mean square error is used to measure the error between the predicted speed threshold and the actual speed threshold to ensure the accuracy of model prediction. Weight decay is used to prevent overfitting. By penalizing the size of the weight, the model weight is forced to remain small and the generalization ability of the model is improved. The stability penalty term is used to consider the influence of stability during model training. When the stability is poor (i.e., φ is large), the value of this term is large, and the contribution to the total loss is also large, prompting the model to impose stricter restrictions on the speed threshold when the stability is poor. The turning angle penalty term is used to consider the influence of the turning angle during model training, providing higher safety in sharp turns. When the turning angle is large, the value of this term is large, and the contribution to the total loss is also large, prompting the model to impose stricter restrictions on the speed threshold when making sharp turns.
[0058] Rollover risk is considered through the turning angle penalty. The larger the turning angle, the higher the rollover risk, so it needs to be controlled through the penalty term. Overload risk is considered through mean square error and weight decay. Mean square error is used to ensure the accuracy of model prediction, thereby avoiding dangers caused by overload. Weight decay is used to prevent model overfitting and ensure the stability of the model under different load conditions. Stability risk is considered through the stability penalty term. The specific form of the loss function is as follows:
[0059]
[0060] in, is the total loss, n is the number of samples, v max,i is the predicted speed threshold of the i-th sample,
[0061] v actual_max,i is the actual vehicle speed threshold of the i-th sample, W j is the jth weight in the model,
[0062] m is the number of weights in the model, λ1, λ2, λ3 are the hyperparameters of the regularization term, which are used to adjust the weights of each loss, φ i is the stationarity of the i-th sample, θ i is the turning angle of the i-th sample.
[0063] S4, forward propagation, calculate the output of each layer.
[0064] S5. Use the gradient descent method to minimize the loss function and update the model parameters. The specific steps are as follows:
[0065] S51. Calculate the gradient of the loss function:
[0066]
[0067] S52, update weights and biases:
[0068]
[0069]
[0070] Among them, α is the learning rate.
[0071] S53, model training, inputting data into the model in batches, and performing multiple iterative training until the loss function converges or reaches the preset number of training rounds.
[0072] S54, model evaluation, use the test set to evaluate the model performance.
[0073] In this embodiment, in order to comprehensively evaluate the performance of the tricycle rollover prevention neural network model, the performance indicators are selected as follows:
[0074] 1) Mean Square Error (MSE)
[0075]
[0076] Measures the mean square error between the predicted speed threshold and the actual speed threshold. A lower MSE indicates a more accurate model prediction.
[0077] 2) Mean Absolute Error (MAE)
[0078]
[0079] Measures the mean absolute error between the predicted speed threshold and the actual speed threshold. A lower MAE indicates that the model prediction is more accurate and is insensitive to outliers.
[0080] 3) Coefficient of determination (R 2 )
[0081]
[0082] It indicates the proportion of the model explaining the variance of the variable, ranging from 0 to 1. 2 The closer it is to 1, the better the model is.
[0083] 4) Mean Squared Percent Error (MAPE)
[0084]
[0085] It measures the average percentage of error between the predicted value and the actual value. A lower MAPE indicates that the model prediction has a smaller relative error.
[0086] 5)Stability sensitivity analysis
[0087]
[0088] It is used to evaluate the sensitivity of the model to changes in stationarity, indicating the response of the model output to changes in the input stationarity. Lower sensitivity indicates better stability of the model under different stationarities.
[0089] 6) Turning angle sensitivity analysis
[0090]
[0091] It is used to evaluate the sensitivity of the model to the change of turning angle, indicating the response of the model output to the change of the input turning angle. The lower the sensitivity, the better the stability of the model under different turning angles.
[0092] After calculating and recording the above performance indicators, check the values of each indicator to find out where improvements are needed. For example, if the MSE and MAE are high, it means that the model prediction accuracy is insufficient; if the stability and turning angle sensitivity are high, it means that the model is unstable under different stability and turning angles. Based on various performance indicators, you can adjust the learning rate (if the model converges slowly or does not converge, you can increase or decrease the learning rate α appropriately. For example, if the current learning rate is 0.01, you can try to adjust it to 0.001 or 0.1 for comparison, and choose a learning rate that makes the model converge faster and has better results), regularization parameters (adjust the L2 regularization parameter λ1 to prevent overfitting, adjust the stability penalty parameter λ2 and the turning angle penalty parameter λ3 to improve the stability of the model under different stability and turning angles. For example, the current λ1 is 0.01, you can try to adjust it to 0.001 or 0.1 to observe whether the model is overfitting. The current λ2 and λ3 are 0. 01, you can try to increase it to 0.1 to observe whether the smoothness and turning angle sensitivity are reduced. ), batch size (adjusting the batch size can affect the stability and convergence speed of model training. A smaller batch size can update model parameters more frequently, but may introduce more noise; a larger batch size has fewer updates, but each update is more stable; for example, the current batch size is 32, you can try to adjust it to 64 or 128 to observe the impact on the convergence speed and stability of the model), model architecture (for example, the current network architecture has two hidden layers, each with 64 neurons, you can try to increase it to three hidden layers, each with 128 neurons), etc.
[0093] After each adjustment, retrain the model, calculate and record various performance indicators, and conduct comparative analysis. Repeat the above process until all indicators reach a satisfactory level, the model is stable and the performance is good.
[0094] By calculating the above performance indicators, we can fully understand the performance of the model in different scenarios. According to the results of these indicators, we can reasonably adjust the model parameters and architecture, and continuously optimize the model to ensure that it has good predictive ability and stability in practical applications.
[0095] The microprocessor sends a speed control instruction to the electronic control unit (ECU) of the speed control module according to the speed threshold, and the speed control module is used to limit the speed of the tricycle based on the speed control instruction. The speed control module includes an electronic control unit (ECU) and an actuator (such as a motor controller), which is responsible for adjusting the motor output of the tricycle, that is, the electronic control unit (ECU) adjusts the motor speed through the actuator according to the speed control instruction to achieve the speed control of the tricycle motor, thereby controlling the vehicle speed.
[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A safe driving system for an electric tricycle, characterized in that: It includes vehicle weight detection module, turning angle detection module, stability detection module, vehicle speed control module and pre-trained tricycle anti-rollover neural network model; The vehicle weight detection module is used to detect the weight data of the tricycle when it is loaded; The turning angle detection module is used to detect the real-time turning angle data of the tricycle; The stability detection module is used to detect the stability data of the tricycle; The pre-trained tricycle rollover prevention neural network model is used to generate a vehicle speed threshold based on vehicle speed, height of the tricycle's center of gravity, total weight of the tricycle when loaded, real-time turning angle data of the tricycle, and stability data of the tricycle, wherein the total weight of the tricycle when loaded is obtained based on the detected weight data and the tricycle's own weight; The vehicle speed control module is used to limit the speed of the three-wheeled vehicle based on a vehicle speed threshold.
2. The electric tricycle safety driving system according to claim 1, characterized in that: The vehicle weight detection module includes a weighing sensor installed on the bearing point of the tricycle chassis, which is used to detect the weight data of the tricycle when loaded; the turning angle detection module includes a gyroscope sensor installed on the frame of the tricycle, which is used to detect the real-time turning angle data of the tricycle; the stability detection module includes an accelerometer and a tilt sensor installed on the frame or chassis of the tricycle, which is used to detect the stability data of the tricycle; the pre-trained tricycle anti-rollover neural network model is deployed in the central processing unit of the electric tricycle; the detected weight data, real-time turning angle data and stability data are transmitted to the central processing unit of the electric tricycle.
3. The electric tricycle safety driving system according to claim 1, characterized in that: The vehicle speed control module includes an electronic control unit and an actuator. When the three wheels turn, the vehicle speed control module can adjust the motor output of the three-wheel vehicle based on the vehicle speed threshold, thereby controlling the vehicle speed.
4. The electric tricycle safety driving system according to claim 1, characterized in that: The tricycle anti-rollover neural network model includes an input layer, a hidden layer and an output layer; the input layer is used to receive the vehicle speed, the height of the tricycle's center of gravity, the total weight of the tricycle when loaded, the real-time turning angle of the tricycle and the stability of the tricycle; the hidden layer adopts a multi-layer perceptron structure, and each layer of the hidden layer uses a ReLU activation function; the output layer is a single neuron layer, which uses a linear activation function to output the vehicle speed threshold.
5. The electric tricycle safety driving system according to claim 2, characterized in that: There are two hidden layers; the input vector is linearly transformed and bias-adjusted, and then passes through the ReLU activation function to obtain the output f1(x) of the first hidden layer: the output of the first hidden layer is linearly transformed and bias-adjusted, and then passes through the ReLU activation function to obtain the output f2(f1(x)) of the second hidden layer.
6. The electric tricycle safety driving system according to claim 3, characterized in that: Speed threshold v max The calculation formula is: υ max =W3·f2(f1(x))+b3 Among them, W3 is the weight matrix of the output layer, and b3 is the bias of the output layer.
7. The electric tricycle safety driving system according to claim 4, characterized in that: During the pre-training process of the tricycle anti-rollover neural network model, the gradient descent method is used to minimize the loss function and update the model parameters. The loss function used takes into account the rollover risk, overload risk and stability risk.
Citation Information
Patent Citations
Method and system for preventing turning on one side during veering process of automobile
CN104590254A
Electric tricycle steering rollover prevention control method and tricycle adopting same
CN105946580A
Self-adaptive estimation method for rollover threshold of heavy-duty vehicle
CN111645698A
Anti-rollover method and device for new energy vehicle
CN112660113A
VR physical engine deployment system based on neural network
CN117472190A
