Multifunctional rescue vehicle based on automation and intelligent control system thereof

Abstract

The invention relates to the field of rescue vehicle control, in particular to an automation-based multifunctional rescue vehicle and an intelligent control system thereof, and the system comprises a vibration feature extraction module which is used for enabling vibration data to pass through a vibration feature extraction model based on multi-frequency domain scale feature extraction and an extrusion excitation network to generate vibration features; the temperature feature extraction module is used for processing the temperature data through a temperature feature extraction model based on a time sequence convolutional network to generate temperature features; and the fusion control module is used for enabling the vibration characteristics and the temperature characteristics to pass through a fusion model based on a full connection layer so as to generate a target rotating speed for controlling the execution of the cooling fan group. According to the method, the environment noise and characteristic vibration representing poor interface connection can be effectively distinguished, key characteristics of vibration and temperature are synthesized, a safer fan rotating speed instruction is generated, and the interface safety of the rescue vehicle for high-power transmission rescue operation is improved.

Description

Technical Field

[0001] This invention relates to the field of rescue vehicle control, and more particularly to an automated multi-functional rescue vehicle and its intelligent control system. Background Technology

[0002] With the popularization of new energy vehicles and the continuous growth of traditional fuel vehicles, the demand for vehicle rescue is becoming increasingly diversified and complex. Among them, multi-functional rescue vehicles, as the core equipment for on-site emergency rescue, need to have multiple rescue capabilities such as rapid jump-start and charging of new energy vehicles.

[0003] However, when multi-functional rescue vehicles perform high-current output tasks, such as charging new energy vehicles or jump-starting fuel vehicles or new energy vehicles, the rescue equipment interfaces are prone to generating a lot of heat due to continuous high-load operation. If heat dissipation is not timely, it may lead to interface overheating, insulation aging, or even connection failure, causing safety accidents.

[0004] Furthermore, due to the limited internal space of the multi-functional rescue vehicle, the vibration of the integrated cooling fan can easily be transmitted to the interface, and the connection of the interface is easily affected by vibration, causing poor contact or momentary power failure, which further increases the operational risk.

[0005] Currently, most common rescue vehicle control systems employ simple fan control strategies based on temperature thresholds, such as initiating cooling when the interface temperature exceeds a fixed value. While this method is simple in structure, it has significant limitations: firstly, threshold control has a slow response time, making it difficult to handle sudden temperature rises or intermittent contact problems; secondly, the interface current fluctuates greatly and heat accumulates rapidly during charging of new energy vehicles, making traditional control modes prone to insufficient cooling or continuous high-speed operation leading to energy waste.

[0006] In recent years, some studies have attempted to improve fan control accuracy through multi-sensor fusion, such as simultaneously collecting temperature and current data to adjust cooling strategies. However, current data can only be read directly from the battery's BMS (Battery Management System) data frames, resulting in unavoidable system response and communication delays, thus failing to reflect true instantaneous faults. Furthermore, the failure to deeply mine the temporal characteristics and frequency domain patterns in sensor data leads to insufficient system adaptability in dynamic environments.

[0007] Therefore, how to integrate vibration and temperature detection data to achieve precise real-time control of the cooling fan of the multi-functional rescue vehicle interface, thereby ensuring the interface safety of the multi-functional rescue vehicle during high-load plug-in charging rescue missions, is a technical problem that needs to be solved. Summary of the Invention

[0008] To address this, the present invention provides an automated multi-functional rescue vehicle and its intelligent control system. Through dilated causal convolutional blocks, it can efficiently process temperature time-series data and accurately capture temperature rise trends. By analyzing vibration signals through multi-frequency domain scale feature extraction and excitation networks, it can effectively distinguish between environmental noise and characteristic vibrations representing poor interface connections. By integrating key features of vibration and temperature, it can then predict the interface connection status through convolutional neural networks, generating safer interface fan speeds and improving the interface safety of the rescue vehicle during high-power power transmission rescue operations.

[0009] To achieve the above objectives, this invention proposes an intelligent control system for a multi-functional rescue vehicle based on automation, comprising:

[0010] The battery pack is located inside the multi-functional rescue vehicle;

[0011] An interface bracket is provided, which is equipped with a cooling fan assembly, a jump-start interface, and a charging interface. The cooling fan assembly is used to dissipate heat from the jump-start interface and the charging interface. The jump-start interface and the charging interface are connected to the battery pack for vehicle jump-starting and charging of new energy vehicles.

[0012] A vibration sensor array is arranged on the interface bracket around the jumper interface and the charging interface;

[0013] A temperature sensor array is located at the female contacts of the jumper and charging interfaces;

[0014] The controller communicates with the cooling fan assembly, vibration sensor assembly, and temperature sensor assembly. The controller is equipped with:

[0015] The vibration feature extraction module is used to generate vibration features from the vibration data collected by the vibration sensor group through a vibration feature extraction model based on multi-frequency domain scale feature extraction and compression excitation network.

[0016] The temperature feature extraction module is used to generate temperature features by processing the temperature data collected by the temperature sensor group through a temperature feature extraction model based on a temporal convolutional network.

[0017] The fusion control module is used to generate a target rotational speed for controlling the cooling fan assembly by using a fusion model based on a fully connected layer to combine the vibration characteristics and the temperature characteristics.

[0018] Furthermore, the vibration feature extraction model includes multi-scale convolutional layers and a squeezing excitation network, and the vibration feature extraction module includes:

[0019] A multi-frequency domain scale feature extraction unit is used to pass the vibration data through the multi-scale convolutional layer with the expansion rate and receptive field set based on the sampling frequency to generate a multi-scale fused feature tensor.

[0020] The squeeze excitation network unit is used to generate a calibration feature vector by passing the multi-scale fused feature tensor through a squeeze excitation network that calculates a compressed global feature vector based on the target frequency;

[0021] A residual connection unit is used to perform a residual connection between the calibration feature vector and the vibration data to generate the vibration feature.

[0022] Furthermore, the multi-frequency domain scale feature extraction unit includes:

[0023] The dilation rate calculation subunit calculates the dilation rate based on the ratio of the kernel size of the multi-scale convolutional layer to the sampling frequency.

[0024] The receptive field computation subunit is used to generate the receptive field based on the product of the convolution kernel size and the dilation rate.

[0025] In particular, by designing multi-scale convolutional layers with multi-branch convolutional structures, each branch targets a specific feature frequency, and each branch is a parallel dilated convolutional channel using the same kernel size but different dilation rates. In this way, each channel of the multi-scale fused feature tensor corresponds to a feature map tuned to a specific frequency. Compared to setting a fixed dilation rate and receptive field, this allows for customization for specific frequency feature extraction tasks.

[0026] Furthermore, the squeeze excitation network unit includes:

[0027] The compressed global feature vector calculation subunit is used to perform a fast Fourier transform based on the target frequency on the multi-scale fused feature tensor in the channel summation dimension to generate frequency components, and to perform squaring operation on the frequency components and frequency summation within the frequency tolerance window to generate a compressed global feature vector.

[0028] The mapping computation subunit is used to pass the compressed global feature vector through the double fully connected layer of the squeeze excitation network to generate a channel weight vector;

[0029] The calibration calculation subunit is used to calculate the product of the channel weight vector and the multi-scale fusion feature tensor to generate the calibration feature vector.

[0030] Furthermore, the temperature feature extraction module includes:

[0031] The difference calculation unit is used to calculate the first-order and second-order differences in the time dimension of the temperature data to generate a difference sequence;

[0032] A temperature feature extraction unit is used to pass the difference sequence through the temperature feature extraction model to generate temperature features.

[0033] Furthermore, the fusion control module includes:

[0034] The feature fusion unit is used to perform feature splicing on the vibration feature and the temperature feature to generate a fused feature;

[0035] A velocity mapping unit is used to pass the fused features through the fusion model to generate the target rotational speed;

[0036] The control unit is used to control the speed of the cooling fan assembly using a PID algorithm based on the target speed.

[0037] Furthermore, the velocity mapping unit includes:

[0038] A velocity transformation mapping subunit is used to pass the fused features through the fusion model to generate velocity transformation quantities;

[0039] A speed control subunit is used to determine the target rotational speed based on the sum of the speed change and the historical target rotational speed.

[0040] Furthermore, the control unit includes:

[0041] The differential term calculation subunit is used to calculate the differential of the difference between the current speed change and the current target speed to generate a differential term;

[0042] The PID control subunit is used to control the speed of the cooling fan assembly based on the differential term using a PID algorithm.

[0043] Furthermore, the controller also includes:

[0044] The collaborative training module is used to construct a loss function based on the mean square error of the target rotational speed and the optimal rotational speed of the sample, and to collaboratively train the vibration feature extraction model, the temperature feature extraction model and the fusion model through the loss function.

[0045] This invention also proposes an automated multi-functional rescue vehicle, which applies the aforementioned automated multi-functional rescue vehicle intelligent control system, and further includes:

[0046] The battery pack is placed on the bottom wall inside the compartment of the multi-functional rescue vehicle;

[0047] The interface bracket includes a vertical bracket and a horizontal bracket. The vertical bracket is equipped with a cooling fan assembly, a jumper connector, and a charging connector. The horizontal bracket is equipped with the battery pack.

[0048] The cooling fan assembly is connected to the jumper interface and the charging interface via an air duct.

[0049] In particular, the fully connected layer establishes a mapping relationship between different risk combinations and corresponding heat dissipation requirements by learning from a large number of samples of fusion features—optimal speed. Compared with traditional threshold-triggered control, it can identify the trend of risk changes. Speed ​​adjustment is no longer a passive response to the current risk, but an adaptation to the direction of risk change in advance, thus extending the lead time for risk response.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention can efficiently process temperature time series data and accurately capture temperature rise trends through dilated causal convolutional blocks; it can analyze vibration signals through multi-frequency domain scale feature extraction and excitation network compression, effectively distinguishing environmental noise from characteristic vibrations that characterize poor interface connections; it can integrate key features of vibration and temperature, and then predict the connection status of the interface through convolutional neural network, generating safer and more scientific fan speed commands for the interface, thus improving the interface safety of rescue vehicles performing high-power power transmission rescue operations.

[0051] In particular, this invention designs a multi-scale convolutional layer with a multi-branch convolutional structure, where each branch targets a specific feature frequency and each branch is a parallel dilated convolutional channel using the same kernel size and different dilation rates. In this way, each channel of the multi-scale fused feature tensor corresponds to a feature map tuned to a specific frequency. Compared to setting a fixed dilation rate and receptive field, this invention achieves customization for specific frequency feature extraction tasks.

[0052] In particular, the fully connected layer of the present invention establishes a mapping relationship between different risk combinations and corresponding heat dissipation requirements by learning from a large number of samples of fusion features—optimal speed. Compared with traditional threshold-triggered control, it can identify the trend of risk changes. Speed ​​adjustment is no longer a passive response to the current risk, but an adaptation to the direction of risk change in advance, thus extending the lead time for risk response. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the controller structure of the automated multi-functional rescue vehicle and its intelligent control system according to an embodiment of the present invention;

[0054] Figure 2 This is a side view of the automated multi-functional rescue vehicle and its intelligent control system according to an embodiment of the present invention.

[0055] Figure 3 This is a schematic diagram of the bottom structure of the automated multi-functional rescue vehicle and its intelligent control system according to an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the front structure of the automated multi-functional rescue vehicle and its intelligent control system according to an embodiment of the present invention;

[0057] Figure 5 This invention relates to an automated multi-functional rescue vehicle and its intelligent control system. Figure 4 A partial structural diagram of part A in the middle;

[0058] Figure 6 This is a flowchart illustrating an automated multi-functional rescue vehicle and its intelligent control system according to an embodiment of the present invention.

[0059] The main components shown in the diagram are: 1. Interface bracket; 11. Vertical bracket; 12. Horizontal bracket; 2. Battery pack; 3. Interface box; 4. Cooling fan assembly; 5. Jump start interface; 6. Charging interface; 7. Vibration sensor assembly; 8. Temperature sensor assembly. Detailed Implementation

[0060] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0061] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0062] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0063] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0064] like Figures 1 to 6As shown, this invention provides an automated multi-functional rescue vehicle and its intelligent control system. Through dilated causal convolutional blocks, it can efficiently process temperature time-series data and accurately capture temperature rise trends. By analyzing vibration signals through multi-frequency domain scale feature extraction and excitation network, it can effectively distinguish between environmental noise and characteristic vibrations representing poor interface connections. By integrating key features of vibration and temperature, it can then predict the interface connection status through convolutional neural networks, generating safer and more scientific fan speed commands for the interface, thus improving the interface safety of the rescue vehicle during high-power power transmission rescue operations.

[0065] like Figures 1 to 6 As shown, this embodiment proposes an intelligent control system for a multi-functional rescue vehicle based on automation, including:

[0066] Battery pack 2 is located inside the multi-functional rescue vehicle;

[0067] The interface bracket 1 has a frame that is equipped with a cooling fan group 4, a jump-start interface 5 and a charging interface 6. The cooling fan group 4 is used to dissipate heat from the jump-start interface 5 and the charging interface 6. The jump-start interface 5 and the charging interface 6 are connected to the battery pack 2 for vehicle jump-starting and charging of new energy vehicles.

[0068] Vibration sensor group 7 is arranged in a semi-circular manner around the grounding interface 5 and the charging interface 6 on the interface bracket 1;

[0069] Temperature sensor group 8 is disposed at the female contacts of the jumper interface 5 and the charging interface 6;

[0070] The controller communicates with the cooling fan group 4, the vibration sensor group 7, and the temperature sensor group 8. The controller is equipped with:

[0071] The vibration feature extraction module is used to generate vibration features from the vibration data collected by the vibration sensor group 7 through a vibration feature extraction model based on multi-frequency domain scale feature extraction and compression excitation network.

[0072] The temperature feature extraction module is used to generate temperature features by passing the temperature data collected by the temperature sensor group 8 through a temperature feature extraction model based on a temporal convolutional network.

[0073] The fusion control module is used to generate a target rotational speed for controlling the cooling fan group 4 by using the vibration characteristics and temperature characteristics through a fusion model based on a fully connected layer.

[0074] See Figure 4 and 5Since the cooling fan group 4 used for cooling the interior of the multi-functional rescue vehicle, the jump-start interface 5 and the charging interface 6 are integrated with the interface bracket 1, the vibration of the fan will be transmitted to the jump-start interface 5 and the charging interface 6 through the interface bracket 1. Multiple large fan vibrations can easily accumulate at the interface, causing poor contact between the interface and the charging cable.

[0075] Therefore, in this embodiment, the connection status of the interface is predicted by integrating the data collected by the vibration sensor group 7 and the temperature sensor group 8 through a model based on a convolutional neural network, and an appropriate fan speed is set to ensure the connection safety of the jump-start interface 5 and the charging interface 6.

[0076] In this embodiment, the vibration feature extraction model includes multi-scale convolutional layers and a squeezing excitation network, and the vibration feature extraction module includes:

[0077] A multi-frequency domain scale feature extraction unit is used to pass the vibration data through the multi-scale convolutional layer with the expansion rate and receptive field set based on the sampling frequency to generate a multi-scale fused feature tensor.

[0078] The squeeze excitation network unit is used to generate a calibration feature vector by passing the multi-scale fused feature tensor through a squeeze excitation network that calculates a compressed global feature vector based on the target frequency;

[0079] A residual connection unit is used to perform a residual connection between the calibration feature vector and the vibration data to generate the vibration feature.

[0080] Specifically, the process by which residual connection elements generate vibration characteristics can be represented as follows:

[0081]

[0082] In the formula, Indicates vibration characteristics, This represents the activation function, preferably ReLU. Represents the calibration feature vector. This represents a 1x1 convolution operation used to extract features and adjust channel dimensions on vibration data X. The vibration data format is as follows: Where B represents the batch size, preferably 32, T represents the time step, and C represents the number of input channels, preferably 3, corresponding to the triaxial data of the vibration sensor group.

[0083] In particular, by setting the convolutional layer expansion rate and receptive field based on the vibration period of the rescue interface, low-frequency turbulence vibration and high-frequency contact vibration of the interface can be captured simultaneously, avoiding the problem of missed or misjudged detection under a single scale.

[0084] In this embodiment, the multi-frequency domain scale feature extraction unit includes:

[0085] The dilation rate calculation subunit calculates the dilation rate based on the ratio of the kernel size of the multi-scale convolutional layer to the sampling frequency.

[0086] The receptive field computation subunit is used to generate the receptive field based on the product of the convolution kernel size and the dilation rate.

[0087] Specifically, the process of generating the expansion rate and the receptive field can be represented as follows:

[0088]

[0089] In the formula, d represents the expansion rate. This represents the scaling factor, preferably 0.5, so that the receptive field covers half a sampling period, suitable for capturing peaks and troughs. The target period is represented by K, which represents the kernel size of the multi-scale convolutional layer, preferably 3. This represents the receptive field of the i-th branch of a multi-scale convolutional layer.

[0090] Among them, through Calculate the number of sampling points within the target period, where These represent the sampling frequency and the target frequency of the vibration signal, respectively. The sampling frequency is defined as the number of times the vibration signal repeats per second, and the target frequency is defined as the number of sampling points collected per second. The target frequency includes the basic rotational frequency of the cooling fan assembly, the nth harmonic frequency, the resonant frequency of the interface bracket, the contact vibration frequency during power-on (20-30Hz), the vibration period during bumps (1-5Hz), and the vibration period during interface loosening (40-50Hz). Therefore... This indicates how many sampling points a complete target frequency period contains.

[0091] Specifically, the process of the multi-frequency domain scale feature extraction unit generating the multi-scale fused feature tensor can be represented as:

[0092]

[0093] In the formula, This represents the feature tensor output by the dilated convolution of the i-th branch of a multi-scale convolutional layer. This represents the activation function, preferably Sigmoid. This represents the set expansion rate of the i-th branch. , feel the wild The convolution operation, where X represents the vibration data, The multi-scale fusion feature tensor is represented by a format including the dimension of the time step T and the dimension of the total number of channels (sum of the input channels of all branches). , This indicates a concatenation operation on the feature tensors from the i-th branch to the n-th branch, where n is preferably 6.

[0094] In particular, vibration signals are time-domain signals, but their characteristics are mainly reflected in the frequency domain, such as the fundamental frequency and harmonics corresponding to fan speed. Dilated convolution performs pattern matching in the time domain, and the dilation rate determines the frequency range that the convolution kernel can match, while the receptive field determines the length of the input sequence for the dilated convolution. Therefore, this embodiment designs a multi-scale convolutional layer with a multi-branch convolutional structure. Each branch targets a specific feature frequency, and each branch is a parallel dilated convolution channel using the same kernel size and different dilation rates. In this way, each channel of the multi-scale fused feature tensor corresponds to a feature map tuned to a specific frequency. Compared to setting a fixed dilation rate and receptive field, this embodiment can determine the features of the main frequency components, achieving customization for specific frequency feature extraction tasks.

[0095] In this embodiment, the extrusion excitation network unit includes:

[0096] The compressed global feature vector calculation subunit is used to perform a fast Fourier transform based on the target frequency on the multi-scale fused feature tensor in the channel summation dimension to generate frequency components, and to perform squaring operation on the frequency components and frequency summation within the frequency tolerance window to generate a compressed global feature vector.

[0097] The mapping computation subunit is used to pass the compressed global feature vector through the double fully connected layer of the squeeze excitation network to generate a channel weight vector;

[0098] The calibration calculation subunit is used to calculate the product of the channel weight vector and the multi-scale fusion feature tensor to generate the calibration feature vector.

[0099] Specifically, the process of generating compressed global feature vectors can be represented as:

[0100]

[0101] In the formula, M represents the compressed global feature vector of channel c, and M represents the target frequency. The number of This represents the summation of the results over all target frequencies. The half-width of the frequency tolerance window determines the range around each target frequency. What size neighborhood range will be included in the calculation? This represents the summation of frequencies within the frequency tolerance window. This indicates that a Fast Fourier Transform is performed on the frequency tolerance window B of data A. Represents frequency components, A vector representing the channel c of the multi-scale fused feature tensor's channel sum dimension. Indicates the target frequency This is converted to the corresponding index position in the Fast Fourier Transform (FFT) result array. This represents the square of the modulus of the complex number calculated using the Fast Fourier Transform with respect to the index position.

[0102] In particular, the total energy of the c-th feature channel is calculated across all target frequencies and their neighborhoods, and this energy value is used as a compressed global feature vector representing the importance of the channel. This compressed global feature vector serves as the input for subsequent excitation operations, directly reflecting the correlation strength between each channel and the target physical frequency, which greatly enhances the interpretability and physical consistency of the model.

[0103] Specifically, the process of generating channel weight vectors and calibrating feature vectors can be represented as follows:

[0104]

[0105] In the formula, s represents the channel weight vector. This represents the activation function, preferably Sigmoid. Let these represent the learnable matrix and bias term of the second fully connected layer in the dual fully connected layer, respectively. This represents the activation function, preferably ReLU. Let represent the learnable matrix and bias term of the first fully connected layer of the dual fully connected layer, respectively, and z represent the compressed global feature vector of all channels c. The concatenated vector, Represents the calibration feature vector. This indicates element-wise multiplication. This represents the multi-scale fusion feature tensor.

[0106] In particular, the nonlinear relationship between channels is learned through the channel weight vector generated by the dual fully connected layer. The value of the channel weight vector is directly guided by the energy of the channel at the target frequency. The weight is applied to the multi-scale fusion feature tensor, so that the calibration feature vector contains multi-scale frequency information and is weighted according to physical prior knowledge.

[0107] In this embodiment, the temperature feature extraction module includes:

[0108] The difference calculation unit is used to calculate the first-order and second-order differences in the time dimension of the temperature data to generate a difference sequence;

[0109] A temperature feature extraction unit is used to pass the difference sequence through the temperature feature extraction model to generate temperature features.

[0110] Specifically, the process of generating the difference sequence can be represented as:

[0111]

[0112] In the formula, This represents the set of all temperature data. Represents the first-order difference. , Indicates the second-order difference. Indicates the first to the last Temperature data collected at each time point, i.e., the first-order difference directly describes the instantaneous rate of temperature change and is the most direct indicator of abnormal heat generation caused by poor contact. The second-order difference describes the change in the rate of temperature change and is a strong signal of poor contact. This represents a difference sequence, which is a weighted concatenation vector of temperature data, first-order differences, and second-order differences. The format specifies the number of data acquisition points and the dimension. These represent scaling factors, preferably 0.7 and 0.3.

[0113] Specifically, the process of generating fused features can be represented as:

[0114]

[0115] In the formula, This represents the features of dilated causal convolution. This represents a dilated causal convolution operation with a dilation factor of d, where d is preferably 3. This represents the bias term of dilated causal convolution.

[0116] In particular, after receiving highly recognizable temperature features and precise vibration features, the fusion control module can more accurately determine the actual state of the rescue interface. When the vibration features reflect low-frequency vibration anomalies, and the temperature features reflect slight heating of the first-order differential and no acceleration of the second-order differential, the module can determine it as a low-risk anomaly.

[0117] In this embodiment, the fusion control module includes:

[0118] The feature fusion unit is used to perform feature splicing on the vibration feature and the temperature feature to generate a fused feature;

[0119] A velocity mapping unit is used to pass the fused features through the fusion model to generate the target rotational speed;

[0120] The control unit is used to control the speed of the cooling fan assembly using a PID algorithm based on the target speed.

[0121] Specifically, the process of generating fused features can be represented as:

[0122]

[0123] In the formula, Indicates fusion features, Indicates vibration characteristics, Indicates temperature characteristics.

[0124] In particular, the fully connected layer establishes a mapping relationship between different risk combinations and corresponding heat dissipation requirements by learning from a large number of samples of fusion features—optimal speed. Compared with traditional threshold-triggered control, the fully connected layer can output differentiated target speeds.

[0125] In this embodiment, the velocity mapping unit includes:

[0126] A velocity transformation mapping subunit is used to pass the fused features through the fusion model to generate velocity transformation quantities;

[0127] A speed control subunit is used to determine the target rotational speed based on the sum of the speed change and the historical target rotational speed.

[0128] Specifically, the process of generating the target rotational speed can be represented as:

[0129]

[0130] In the formula, Indicates preliminary mapping features, express Activation function These represent the learnable matrix and bias term of the third fully connected layer, respectively. Indicates fusion features, Indicates the change in velocity. Indicates the target rotational speed. This indicates the historical target rotational speed.

[0131] In particular, by identifying the trend of risk changes, speed adjustment is no longer a passive response to the current risk, but an adaptation to the direction of risk changes in advance, thus extending the lead time for risk response.

[0132] In this embodiment, the control unit includes:

[0133] The differential term calculation subunit is used to calculate the differential of the difference between the current speed change and the current target speed to generate a differential term;

[0134] The PID control subunit is used to control the speed of the cooling fan assembly based on the differential term using a PID algorithm.

[0135] Specifically, the calculation process of the differential term can be expressed as follows:

[0136]

[0137] In the formula, Represents the differential term. Denotes the differential coefficient. Indicates the sampling period. This indicates a smaller value, preferably 0.6. These represent the target rotational speeds at the previous and current moments, respectively. These represent the changes in rotational speed at the previous and current moments, respectively.

[0138] Specifically, the proportional and integral terms of the PID algorithm are conventional calculations of the speed change, and the values ​​of the proportional coefficient, derivative coefficient, and integral coefficient of the PID algorithm are 0.3, 0.6, and 0.1, respectively.

[0139] In particular, the differential term only differentiates the feedback value and not the setpoint, which can greatly reduce the impact of sudden changes in the setpoint on the system.

[0140] In this embodiment, the controller is also equipped with:

[0141] The collaborative training module is used to construct a loss function based on the mean square error of the target rotational speed and the optimal rotational speed of the sample, and to collaboratively train the vibration feature extraction model, the temperature feature extraction model and the fusion model through the loss function.

[0142] Specifically, the loss function can be expressed as:

[0143]

[0144] In the formula, Let N represent the loss function, and N represent the number of samples. Indicates the i-th target rotational speed. This represents the optimal rotational speed for the i-th sample.

[0145] In particular, by optimizing multiple models together and sharing backpropagation loss, the problems of poor model adaptability and low overall control accuracy caused by independent training are solved.

[0146] In this embodiment, an automated multi-functional rescue vehicle is also proposed. The multi-functional rescue vehicle utilizes the aforementioned automated multi-functional rescue vehicle intelligent control system. The multi-functional rescue vehicle further includes:

[0147] The battery pack 2 is placed on the bottom wall inside the compartment of the multi-functional rescue vehicle;

[0148] The interface bracket 1 includes a vertical bracket 11 and a horizontal bracket 12. The vertical bracket 11 is equipped with a cooling fan assembly 4, a jump-start interface 5, and a charging interface 6. The horizontal bracket 12 is equipped with the battery pack 2.

[0149] The cooling fan assembly 4 is connected to the jumper interface 5 and the charging interface 6 via an air duct.

[0150] See Figure 2 The interface bracket 1 includes three sets of spaced-apart vertical brackets 11 and four sets of spaced-apart horizontal brackets 12. See also... Figure 3 It can be clearly seen that battery pack 2 almost completely occupies the bottom wall of the vehicle compartment. Therefore, battery pack 2 can be a large-volume, high-capacity battery, enabling high-power rapid charging and discharging to achieve the purpose of rapid rescue.

[0151] See Figure 4 and 5 It can be clearly concluded that the vertical bracket 11 integrates at least a cooling fan group 4 with 12 fans, 4 charging ports 6 and 2 jump-start ports 5.

[0152] In particular, it has made the interface operation of the multi-functional rescue vehicle more convenient and increased the storage space for tools.

[0153] In this embodiment, dilated causal convolutional blocks efficiently process temperature time-series data and accurately capture temperature rise trends. Multi-frequency domain scale feature extraction and compression excitation networks analyze vibration signals, effectively distinguishing between environmental noise and characteristic vibrations representing poor interface connections. By integrating key features of vibration and temperature, convolutional neural networks predict interface connectivity, generating safer and more scientific fan speed commands, thus improving interface safety for high-power power transmission rescue operations. By designing multi-scale convolutional layers with multi-branch convolutional structures, each branch targets a specific feature frequency. Each branch is a parallel dilated convolutional channel using the same kernel size but different dilation rates. This allows each channel of the multi-scale fused feature tensor to correspond to a feature map tuned to a specific frequency, achieving customization for specific frequency feature extraction tasks compared to setting fixed dilation rates and receptive fields. By learning from a large number of samples of fusion features—optimal speeds—the fully connected layer establishes a mapping relationship between different risk combinations and corresponding heat dissipation requirements. Compared with traditional threshold-triggered control, it can identify risk change trends. Speed ​​adjustment is no longer a passive response to the current risk, but rather an adaptation to the direction of risk change in advance, thus extending the lead time for risk response.

[0154] Those skilled in the art will recognize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0155] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An automated multi-functional rescue vehicle intelligent control system, characterized in that, The utility model relates to a kind of new energy vehicle rescue device, including: Battery pack (2) is arranged in the inside of multifunctional rescue vehicle; Interface support (1), its frame body installs heat dissipation fan group (4), connects electricity interface (5) and charging interface (6), the heat dissipation fan group (4) is used to heat dissipation to connect electricity interface (5) and charging interface (6), connect electricity interface (5) and charging interface (6) with the battery pack (2), to carry out the vehicle rescue of vehicle connecting electricity and new energy vehicle charging; Vibration sensor group (7) is arranged around connecting electricity interface (5) and charging interface (6) on the interface support (1); Temperature sensor group (8) is arranged at the female contact of connecting electricity interface (5) and charging interface (6); Controller, with the heat dissipation fan group (4), vibration sensor group (7) and temperature sensor group (8) communication, the controller carries: Vibration feature extraction module, to generate vibration feature by vibration feature extraction model based on multi-frequency domain scale feature extraction and extrusion excitation network to the vibration data collected by vibration sensor group (7); Temperature feature extraction module, to generate temperature feature by temperature feature extraction model based on time series convolution network to the temperature data collected by temperature sensor group (8); Fusion control module, to generate target rotating speed executed by the heat dissipation fan group (4) by fusion model based on full connection layer to the vibration feature and the temperature feature.

2. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 1, wherein, The vibration feature extraction model includes multi-scale convolution layer and extrusion excitation network, and the vibration feature extraction module includes: Multi-frequency domain scale feature extraction unit, to generate multi-scale fusion feature tensor by the multi-scale convolution layer based on sampling frequency setting inflation rate and receptive field to the vibration data; Extrusion excitation network unit, to generate calibration feature vector by extrusion excitation network based on target frequency calculation compression global feature vector to the multi-scale fusion feature tensor; Residual connection unit, to generate the vibration feature by residual connection to the calibration feature vector and the vibration data.

3. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 2, wherein, The multi-frequency domain scale feature extraction unit includes: Inflation rate calculation subunit, based on the proportion of the convolution kernel size of the multi-scale convolution layer and the sampling frequency, to generate the inflation rate; Receptive field calculation subunit, to generate the receptive field based on the product of the convolution kernel size and the inflation rate.

4. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 2, wherein, The extrusion excitation network unit includes: Compression global feature vector calculation subunit, to generate frequency component by fast fourier transform based on target frequency to the multi-scale fusion feature tensor in channel total dimension, to generate compression global feature vector by square operation, frequency summation in frequency tolerance window to the frequency component; Mapping calculation subunit, to generate channel weight vector by double full connection layer of the extrusion excitation network to the compression global feature vector; Calibration calculation subunit, to generate the calibration feature vector by calculating the product of the channel weight vector and the multi-scale fusion feature tensor.

5. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 1, wherein, The temperature feature extraction module includes: a difference calculation unit configured to calculate first-order difference and second-order difference of the temperature data in time dimension to generate a difference sequence; a temperature feature extraction unit configured to pass the difference sequence through the temperature feature extraction model to generate temperature features.

6. The automated based multi-functional rescue vehicle intelligent control system according to claim 1, wherein, The fusion control module comprises: a feature fusion unit configured to fuse the vibration features and the temperature features to generate fusion features; a speed mapping unit configured to pass the fusion features through the fusion model to generate the target rotating speed; a control unit configured to control the heat dissipation fan group (4) through a PID algorithm based on the target rotating speed.

7. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 6, wherein, The speed mapping unit comprises: a speed variable mapping subunit configured to pass the fusion features through the fusion model to generate a speed variable; a speed control subunit configured to determine the target rotating speed based on a sum of the speed variable and a historical target rotating speed.

8. The automated based multi-functional rescue vehicle intelligent control system as claimed in claim 6, wherein, The control unit comprises: a differential term calculation subunit configured to calculate a differential of a difference between a current rotating speed variable and the target rotating speed to generate a differential term; a PID control subunit configured to control the heat dissipation fan group (4) through a PID algorithm based on the differential term.

9. The automated based multi-functional rescue vehicle intelligent control system according to claim 1, wherein, The controller further comprises: a cooperative training module configured to construct a loss function based on a mean square error between the target rotating speed and a sample optimal rotating speed, and to cooperatively train the vibration feature extraction model, the temperature feature extraction model and the fusion model through the loss function.

10. An automated based multi-functional rescue vehicle characterized in that, The multifunctional rescue vehicle applies the intelligent control system based on automation according to any one of claims 1 to 9, and further comprises: The battery pack (2) is placed on an inner bottom wall of a vehicle compartment of the multifunctional rescue vehicle; The interface support (1) comprises a vertical support (11) and a horizontal support (12), the vertical support (11) is installed with a heat dissipation fan group (4), a power-on interface (5) and a charging interface (6), and the horizontal support (12) is assembled with the battery pack (2); The heat dissipation fan group (4) is connected with the power-on interface (5) and the charging interface (6) through an air duct.

Images