Intelligent start-stop control system and method for automobile engine

Abstract

The invention relates to the technical field of engine start-stop control, and particularly discloses an intelligent start-stop control system and method for an automobile engine, which adopt a deep learning technology to carry out conjoint analysis on current time and current position and direction of an automobile, extract current driving state characteristics of the automobile, and simultaneously combine with information of a lane where the automobile is located to obtain an intelligent start-stop control system for the automobile engine. Compensation type interactive fusion coding is carried out on the current driving state of an automobile and the information of a lane where the automobile is located, so that the driving environment of the automobile is comprehensively understood, and intelligent control over the automatic start-stop function of an engine is achieved. Therefore, the problem of engine abrasion caused by frequent starting and stopping can be effectively avoided, the driving experience of a user is improved, the fuel efficiency is further improved, emission is reduced, and the service life of the engine is prolonged.

Description

Technical Field

[0001] The present application relates to the technical field of engine start-stop control, and more specifically, to an intelligent start-stop control system and method for an automobile engine. Background Art

[0002] With the development of automobile technology and the improvement of environmental awareness, energy conservation and emission reduction have become one of the important development directions of the automobile industry. As an effective energy-saving measure, automobile engine automatic start-stop technology (Stop-Start System) has been widely used. Automatic start-stop technology automatically shuts down the engine when the vehicle stops temporarily and quickly restarts it when it needs to continue driving, thereby effectively saving fuel and reducing emissions. However, the existing automatic start-stop technology still has some shortcomings in practical applications. For example, when waiting for a short time at a traffic light intersection, frequent start-stop will increase engine wear and start-up time, affecting the driving experience. At this time, if the user does not want frequent automatic start-stop, the automatic start-stop function needs to be manually turned off, which affects the user's operating experience.

[0003] In this regard, the invention patent with publication number CN112977390A discloses an intelligent control system and method for automatic start and stop of automobile engines. On the basis of the existing automatic start and stop control function, it adds an intelligent control function, which automatically determines whether to turn on the automatic start and stop function according to the status of traffic lights, the current position, direction, lane and current time of the automobile, avoiding the inconvenience caused by the user turning on and off the automatic start and stop function.

[0004] However, in the intelligent mode, the above-mentioned automobile engine automatic start-stop intelligent control system mainly realizes intelligent analysis and judgment of the automobile status by obtaining the artificial neural network parameters in the common areas set by the user from the server, which has certain limitations. For example, the update frequency of the artificial neural network parameters is relatively low, and it is difficult to adapt to the dynamic changes of traffic conditions in real time. Moreover, when the user has not selected a specific area or is driving in a new area, it may not be possible to accurately obtain the optimal parameter configuration, thereby affecting the intelligent judgment of the automatic start-stop function.

[0005] Therefore, an optimized intelligent start-stop control method for automobile engines is expected. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an intelligent start-stop control system and method for an automobile engine, which uses deep learning technology to jointly analyze the current time, the current position and direction of the automobile, extract the current driving state characteristics of the automobile, and at the same time combine the lane information where the automobile is located, and perform compensatory interactive fusion encoding of the current driving state of the automobile and the lane information, so as to fully understand the driving environment of the automobile, thereby realizing intelligent control of the automatic start-stop function of the engine. In this way, the problem of engine wear caused by frequent start-stop can be effectively avoided, and the user's driving experience can be improved, thereby further improving fuel efficiency, reducing emissions, and extending the service life of the engine.

[0007] Accordingly, according to one aspect of the present application, a method for intelligent start-stop control of an automobile engine is provided, comprising: obtaining the position information of the automobile at regular intervals, and collecting an image of the intersection when the distance from the intersection is less than a preset distance threshold; judging through the image of the intersection whether the traffic light at the intersection has a countdown, and turning on the automatic start-stop function when the countdown is greater than the preset time threshold and the vehicle speed is zero; judging through the image of the intersection whether there is a countdown of the traffic light at the intersection or there is no traffic light, determining whether to turn on or off the automatic start-stop function according to the current time, the current position of the automobile, the current direction of the automobile, and the lane where the automobile is located, wherein determining whether to turn on or off the automatic start-stop function according to the current time, the current position of the automobile, the current direction of the automobile, and the lane where the automobile is located comprises:

[0008] Extract the vehicle's current state features and lane information semantic features from the current time, the vehicle's current position, the vehicle's current direction, and the lane where the vehicle is located to obtain a vehicle's current state cascade coding vector and a vehicle's lane information semantic embedding coding vector;

[0009] Performing a compensatory interactive fusion on the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located, using common features as a mask, to obtain a multimodal interactive compensatory fusion representation vector of the vehicle state;

[0010] Based on the multimodal interactive compensation fusion representation vector of the vehicle state, it is determined whether to turn on or off the automatic start-stop function.

[0011] According to another aspect of the present application, there is provided an intelligent start-stop control system for an automobile engine, comprising:

[0012] A vehicle information encoding module is used to extract the vehicle current state features and lane information semantic features from the current time, the vehicle current position, the vehicle current direction and the lane where the vehicle is located to obtain a cascade encoding vector of the vehicle current state and a semantic embedding encoding vector of the lane where the vehicle is located;

[0013] An information interaction fusion module, used for performing a compensatory interactive fusion of the cascade coding vector of the current state of the vehicle and the semantic embedded coding vector of the lane information where the vehicle is located, using common features as a mask, to obtain a multimodal interactive compensation fusion representation vector of the vehicle state;

[0014] The automatic start-stop control module is used to determine whether to turn on or off the automatic start-stop function based on the multi-modal interactive compensation fusion representation vector of the vehicle state.

[0015] Compared with the prior art, the automobile engine intelligent start-stop control system and method provided by the present application adopts deep learning technology to jointly analyze the current time, the current position and direction of the automobile, extract the current driving state characteristics of the automobile, and at the same time, combine the lane information where the automobile is located, and perform compensatory interactive fusion coding on the current driving state of the automobile and the lane information where the automobile is located, so as to fully understand the driving environment of the automobile, thereby realizing intelligent control of the automatic start-stop function of the engine. In this way, the engine wear problem caused by frequent start-stop can be effectively avoided, and the user's driving experience can be improved, thereby further improving fuel efficiency, reducing emissions, and extending the service life of the engine. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 This is a flow chart of an automobile engine intelligent start-stop control method according to an embodiment of the present application.

[0018] Figure 2 This is a data flow diagram of the automobile engine intelligent start-stop control method according to an embodiment of the present application.

[0019] Figure 3 This is a flowchart of step S110 in the automobile engine intelligent start-stop control method according to an embodiment of the present application.

[0020] Figure 4 This is a flowchart of step S120 in the automobile engine intelligent start-stop control method according to an embodiment of the present application.

[0021] Figure 5 This is a block diagram of an automobile engine intelligent start-stop control system according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0023] As mentioned in the background technology above, patent CN112977390A proposes an intelligent control method for automatic start and stop of automobile engines, including: obtaining the position information of the automobile at regular intervals, and collecting an intersection image when the distance to the intersection is less than a preset distance threshold; judging through the intersection image whether the traffic light at the intersection has a countdown, and turning on the automatic start and stop function when the countdown is greater than the preset time threshold and the vehicle speed is zero; judging through the intersection image whether there is no traffic light countdown or no traffic light at the intersection, determining whether to turn on or off the automatic start and stop function based on the current time, the current position of the automobile, the current direction of the automobile, and the lane in which the automobile is located.

[0024] Since the above-mentioned automobile engine automatic start-stop intelligent control system mainly realizes intelligent analysis and judgment of the automobile state by obtaining the artificial neural network parameters in the common areas set by the user from the server, there are certain limitations. For example, the update frequency of the artificial neural network parameters is relatively low, and it is difficult to adapt to the dynamic changes of traffic conditions in real time. When the user does not select a specific area or drives in a new area, it may not be possible to accurately obtain the optimal parameter configuration, thereby affecting the intelligent judgment of the automatic start-stop function. In response to this technical problem, the present application proposes an optimized automobile engine intelligent start-stop control method, which uses deep learning technology to jointly analyze the current time, the current position and direction of the car, extract the current driving state characteristics of the car, and combine the lane information where the car is located. By compensating the interactive fusion coding of the current driving state of the car and the lane information, the driving environment of the car is fully understood, thereby realizing the intelligent control of the automatic start-stop function of the engine. In this way, the engine wear problem caused by frequent start-stop can be effectively avoided, and the user's driving experience can be improved, thereby further improving fuel efficiency, reducing emissions, and extending the service life of the engine.

[0025] Figure 1 This is a flow chart of an automobile engine intelligent start-stop control method according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the intelligent start-stop control method for an automobile engine according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the intelligent start-stop control method for an automobile engine includes the following steps: S110, extracting the automobile current state features and the lane information semantic features from the current time, the current position of the automobile, the current direction of the automobile and the lane where the automobile is located to obtain the automobile current state cascade coding vector and the automobile lane information semantic embedded coding vector; S120, performing compensatory interactive fusion of the automobile current state cascade coding vector and the automobile lane information semantic embedded coding vector using common features as a mask to obtain a multimodal interactive compensated fusion representation vector of the automobile state; S130, determining whether to turn on or off the automatic start-stop function based on the automobile state multimodal interactive compensated fusion representation vector.

[0026] It should be understood that in the case of long waiting time and traffic congestion, activating the automatic start-stop function can effectively reduce fuel consumption and emissions. However, in the case of short waiting time and smooth traffic, frequent start-stop may affect driving safety and comfort. Therefore, in order to optimize the service life of the engine and driving experience, it is necessary to intelligently control the opening and closing of the automatic start-stop function according to the actual driving environment and vehicle status.

[0027] Traffic conditions on roads may vary greatly at different times. For example, traffic congestion is more likely during peak hours in the morning and evening, and frequent operation of the automatic start-stop function may increase the inconvenience of the driver, while automatic start-stop can work more effectively during off-peak hours. At the same time, the location of the car is also an important factor affecting traffic conditions. For example, traffic congestion is more common in the central area of ​​the city, while vehicles travel more smoothly in the suburbs or on highways. The direction of the car's travel indicates whether the vehicle is entering or leaving the central area of ​​the city, which helps to further analyze the traffic conditions on the road. Generally, when a vehicle enters the central area of ​​the city, the possibility of traffic congestion increases, while when leaving, the traffic conditions are relatively smooth. In addition, the traffic flow speed of different lanes is also different. For example, the waiting time at the traffic light is different for the straight lane and the left turn lane. Therefore, the present application obtains the current time (for example, morning rush hour, evening rush hour, noon time, night time, etc.), the current location of the car (for example, city center, suburbs, highway, etc.), the current direction of the car (east, west, south, north, etc.), and the lane information of the car (for example, straight lane, left turn lane, dedicated lane, lane line color and style, etc.), so as to more accurately analyze and predict the traffic conditions and driving environment, thereby intelligently controlling the automatic start-stop function of the engine.

[0028] The following is a specific method for obtaining the current time, the current position of the car, the current direction of the car, and the lane information in which the car is located.

[0029] First of all, modern cars are generally equipped with hardware clock devices to provide current time information. Therefore, the current time is mainly obtained by the car's built-in clock system and the on-board computer system. These systems generally perform regular time synchronization operations to ensure the accuracy of time information. In addition, in addition to providing accurate location information, the car's GPS module can also provide accurate time information, because the GPS satellite itself is a high-precision time reference source. And if the car is connected to the Internet, time synchronization can be further achieved through the Network Time Protocol (NTP) server to ensure the accuracy of time information.

[0030] Specifically, the car's hardware clock is initialized every time it starts and maintains time continuity through an internal oscillator. To prevent time drift, the hardware clock is regularly calibrated with a GPS module or other time source. The GPS module calculates precise time information by receiving signals from multiple satellites and transmits it to the on-board computer system. If the car is connected to the Internet, the on-board computer system can synchronize time with a remote time server through the NTP protocol to ensure the accuracy and consistency of time.

[0031] The current location information of the car can be obtained through a variety of sensors and systems, mainly including GPS modules, inertial navigation systems (INS) and map matching algorithms. Among them, the GPS module is the most commonly used method, which receives signals from multiple satellites and calculates the longitude and latitude coordinates of the car to determine the current location of the car. However, in some cases, such as tunnels, high-rise dense areas or underground parking lots, GPS signals may be interfered with or disappear completely. In this case, INS can provide auxiliary positioning, measure the movement state of the car through sensors such as accelerometers and gyroscopes, and deduce the current position.

[0032] The working principle of INS is to calculate the speed and direction changes of the car by integrating the data of the accelerometer and gyroscope, and then deduce the current position of the car. Although the error of INS will accumulate over time, it can still provide relatively accurate positioning information in a short time. In order to further improve the positioning accuracy, GPS data and INS data can be fused, and algorithms such as Kalman filtering can be used to eliminate noise and errors to obtain more accurate position information.

[0033] In addition, the positioning accuracy can be improved through the map matching algorithm. The map matching algorithm matches the rough location information obtained by GPS with the digital map, and corrects the location error according to the topological structure and geographical features of the road network, thereby improving the accuracy of positioning. Digital maps usually contain detailed road network information, such as road name, direction, number of lanes, etc. This information can help the system determine the current location of the car more accurately.

[0034] The current direction of the car can be obtained through a variety of methods, such as an electronic compass, GPS module, and inertial navigation system (INS). The electronic compass is a sensor built into the car that can provide direction information, usually with magnetic north as the reference point. The electronic compass measures the direction of the earth's magnetic field and calculates the heading angle of the car to determine the direction of the car. However, the electronic compass may be interfered by surrounding metal objects, resulting in errors in direction measurement. In order to reduce this error, the electronic compass can be fused with the GPS module and INS data to improve the accuracy of direction measurement.

[0035] The GPS module can also be used to obtain the direction information of the car. By continuously obtaining the position information of two time points, the direction of the car can be calculated. Specifically, the GPS module obtains the latitude and longitude coordinates of the car at two time points respectively, and obtains the direction of the car by calculating the vector between the two coordinates. This method is suitable for measuring the direction of the car in a moving state, but for a stationary or slow-moving car, the accuracy of the direction measurement may be affected.

[0036] The gyroscope in INS can provide the vehicle's heading angle information, thereby determining the direction of the car. The gyroscope measures the rotation rate of the car, calculates the change in the heading angle, and then infers the current direction of the car. The gyroscope has a high degree of direction measurement accuracy, but it will also produce cumulative errors over time. In order to reduce this error, the gyroscope data can be fused with the GPS data and corrected using algorithms such as Kalman filtering to improve the accuracy of direction measurement.

[0037] The lane information of the car can be obtained through a variety of methods, such as camera, LIDAR, millimeter wave radar, high-precision map data, and vehicle-to-vehicle (V2V) or vehicle-to-infrastructure (V2I) communication. Among them, the camera is one of the most commonly used sensors. The camera installed on the front of the car can capture road images and identify lane lines, guide marks and obstacle information through image processing technology.

[0038] The process of obtaining lane information by the camera usually includes the following steps: First, the camera captures the road image and transmits it to the image processing module. The image processing module identifies the position and shape of the lane line through algorithms such as edge detection, line detection, and Hough transform. Next, the lane the car is currently in is determined based on the position and shape information of the lane line. In order to improve the accuracy of recognition, machine learning algorithms can also be combined to train the model to recognize different types of lane lines and guide marks.

[0039] Laser radar (LIDAR) can provide high-precision 3D point cloud data to help identify lane boundaries and the location of other vehicles. LIDAR obtains 3D point cloud data of the surrounding environment by emitting laser beams and measuring the reflection time. This data can be used to build a high-precision environmental model to identify lane boundaries, curbs and other obstacles. LIDAR data is highly accurate, but the cost is relatively high, and it is usually used in high-end models or self-driving vehicles.

[0040] Millimeter-wave radar can also be used to detect the position of vehicles and obstacles ahead and assist in determining lane information. Millimeter-wave radar obtains the distance and relative position of objects ahead by emitting millimeter-wave signals and measuring the reflection time. Compared with LIDAR, millimeter-wave radar has lower cost and stronger penetration, and can work normally in adverse weather conditions such as rain and snow. However, millimeter-wave radar has lower resolution and is mainly used to detect the position of larger objects.

[0041] High-precision map data can provide information such as the number of lanes and lane width on the current road section to assist in determining which lane the car is in. High-precision maps usually contain detailed road network information, such as road name, direction, number of lanes, lane line type, etc. This information can help the system determine the lane the car is in more accurately. In order to obtain high-precision map data, the car needs to be connected to the Internet or a device with built-in storage for high-precision maps.

[0042] Vehicle-to-vehicle (V2V) or vehicle-to-infrastructure (V2I) communication can also be used to obtain lane information. Through V2V communication, cars can exchange location information and lane information with other vehicles to assist in determining the current lane. Through V2I communication, cars can receive information from infrastructure such as traffic lights and traffic signs to further improve the accuracy of lane determination. V2V and V2I communication technologies can provide real-time traffic information to help the system better adapt to dynamically changing traffic environments.

[0043] In summary, through hardware clocks, GPS modules, electronic compasses, cameras, laser radars, millimeter-wave radars, high-precision map data, and V2V / V2I communications, etc., multiple sensors and systems can accurately obtain the current time, the current position, direction, and lane information of the car. Then, deep learning technology can be further used for data fusion and analysis to achieve intelligent control of the engine automatic start-stop function.

[0044] In the above-mentioned intelligent start-stop control method for automobile engines, the step S110 extracts the current state features of the automobile and the semantic features of the lane information from the current time, the current position of the automobile, the current direction of the automobile and the lane where the automobile is located to obtain the cascade coding vector of the current state of the automobile and the semantic embedded coding vector of the lane information where the automobile is located. Among them, Figure 3FIG. 1 is a flow chart of step S110 in the method for intelligent start-stop control of an automobile engine according to an embodiment of the present application. Figure 3 As shown, the step S110 includes: S111, performing one-hot encoding on the current time, the current position of the car and the current direction of the car to obtain a current time one-hot encoding vector, a current position one-hot encoding vector and a current direction one-hot encoding vector; S112, cascading the current time one-hot encoding vector, the current position one-hot encoding vector and the current direction one-hot encoding vector to obtain a cascade encoding vector of the current state of the car; S113, performing semantic embedding encoding on the lane where the car is located to obtain a semantic embedding encoding vector of the lane information where the car is located.

[0045] Specifically, the step S111 performs unique hot encoding on the current time, the current position of the car and the current direction of the car to obtain the current time unique hot encoding vector, the current position unique hot encoding vector and the current direction unique hot encoding vector. It should be understood that the current time, the current position and direction of the car are all category data. Therefore, in order to more effectively extract and analyze data features, the present application first performs unique hot encoding on the current time, the current position of the car and the current direction of the car to convert the category data into a binary vector, wherein each category value corresponds to a dimension, and if a sample belongs to this category, the value of the dimension is 1, otherwise it is 0, thereby obtaining the current time unique hot encoding vector, the current position unique hot encoding vector and the current direction unique hot encoding vector. In this way, converting non-numerical category data into numerical form can better adapt to the input requirements of the deep learning model, thereby improving the processing efficiency and accuracy of the data.

[0046] In a specific example of the present application, the one-hot encoding process of the current time is as follows: First, divide the day into different time periods, such as "morning peak period", "evening peak period", "noon period" and "night period". In order to convert the current time into a one-hot encoding vector, you first need to define the categories of these time periods. For example: morning peak period: 06:00-09:00; evening peak period: 17:00-20:00; noon period: 12:00-14:00; night period: 22:00-06:00; other periods: other times. After defining the time period categories, you can create a one-hot encoding vector for each time period. For example: morning peak period: [1,0,0,0,0]; evening peak period: [0,1,0,0,0]; noon period: [0,0,1,0,0]; night period: [0,0,0,1,0]; other periods: [0,0,0,0,1].

[0047] Next, the current time is obtained through the car's built-in clock system or GPS module. For example, the current time is 17:30, which is during the evening rush hour. According to the time period to which the current time belongs, the corresponding one-hot encoding vector is generated. For example, for 17:30, the generated one-hot encoding vector is [0,1,0,0,0].

[0048] Similarly, following similar logic, the above-mentioned one-hot encoding is also performed for the current position and direction of the car. For example, the latitude and longitude coordinates of the car are obtained through the GPS module, and combined with the digital map data to determine the area where the car is located. For example, the current car is located in the city center. According to the area where the car is located, the corresponding one-hot encoding vector is generated. For the city center, the generated one-hot encoding vector is [1,0,0,0]. The heading angle of the car is obtained through the electronic compass, and the heading angle is converted into a direction category. For example, the current heading angle is 90 degrees, and the corresponding direction is east. According to the current direction of the car, the corresponding one-hot encoding vector is generated. For east, the generated one-hot encoding vector is [1,0,0,0].

[0049] Specifically, in step S112, the current time unique hot encoding vector, the current position unique hot encoding vector and the current direction unique hot encoding vector are cascaded to obtain the cascaded encoding vector of the current state of the car. It should be understood that since the current time, the current position and direction of the car are important factors affecting the driving state and traffic conditions of the car, in order to comprehensively consider the impact of these factors on the driving state of the car, the present application further fuses the current time unique hot encoding vector, the current position unique hot encoding vector and the current direction unique hot encoding vector by cascade encoding to form a comprehensive representation containing multi-dimensional related features, ensuring that the multi-dimensional features can be fully utilized, thereby improving the model's recognition and prediction capabilities for complex driving environments.

[0050] Specifically, the step S113 performs semantic embedding coding on the lane where the car is located to obtain a semantic embedding coding vector of the lane information where the car is located. It should be understood that considering that lane information is not just a simple physical location, it also contains rich traffic rules. For example, there are potential differences in traffic flow speed and waiting time between the straight lane and the left turn lane. In the straight lane waiting in front of the traffic light, it may be decided whether to turn off the engine according to the state of the vehicle in front and the traffic flow; while at a complex intersection with frequent lane changes, it may be inclined to keep the engine running in order to quickly respond to traffic changes. Therefore, in order to fully explore the deep semantic meaning of lane information, this application uses semantic embedding technology to encode the lane where the car is located, converting the lane information into a numerical vector with semantic meaning, fully capturing the semantic features of the lane, such as lane type, lane line color and style, etc., so as to further enhance the model's understanding of the driving environment. In an embodiment of the present application, a pre-trained word embedding model such as Word2Vec or GloVe can be used to implement semantic embedding coding of lane information, thereby generating a semantic embedding coding vector of the lane information where the car is located.

[0051] In the above-mentioned automobile engine intelligent start-stop control method, the step S120 performs a compensatory interactive fusion of the cascade coding vector of the current state of the automobile and the semantic embedded coding vector of the lane information where the automobile is located with common features as a mask to obtain a multimodal interactive compensated fusion representation vector of the automobile state. That is, through the feature interactive fusion processing, the driving state and lane information of the automobile are integrated to generate a more comprehensive and accurate feature representation. In particular, since the cascade coding vector of the current state of the automobile and the semantic embedded coding vector of the lane information where the automobile is located have undergone different feature encoding processing and have different feature dimensions and semantic depths, direct simple splicing and fusion may not be able to fully utilize the complementary information of the two. Therefore, in order to achieve effective information interaction between the two, the present application proposes a compensatory interactive fusion method using common features as a mask to process the cascade coding vector of the current state of the automobile and the semantic embedded coding vector of the lane information where the automobile is located.

[0052] Figure 4 FIG. 1 is a flow chart of step S120 in the method for intelligent start-stop control of an automobile engine according to an embodiment of the present application. Figure 4As shown, the step S120 includes: S121, extracting the common features between the cascade coding vector of the current state of the vehicle and the semantic embedded coding vector of the lane information where the vehicle is located to obtain a multimodal common feature representation vector of the vehicle state; S122, based on the specific features of the cascade coding vector of the current state of the vehicle and the semantic embedded coding vector of the lane information where the vehicle is located relative to the multimodal common feature representation vector of the vehicle state, the cascade coding vector of the current state of the vehicle and the semantic embedded coding vector of the lane information where the vehicle is located perform feature compensation modulation to obtain a feature compensation vector of the current state of the vehicle and a semantic feature compensation vector of the lane information where the vehicle is located; S123, fusing the feature compensation vector of the current state of the vehicle, the semantic feature compensation vector of the lane information where the vehicle is located and the multimodal common feature representation vector of the vehicle state to obtain the multimodal interactive compensation fusion representation vector of the vehicle state.

[0053] In a specific example of the present application, the step S121 includes: inputting the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located into the common feature extraction network to obtain the multimodal common feature representation vector of the vehicle state, which can be expressed by the formula:

[0054]

[0055] Wherein, V1 represents the cascade coding vector of the current state of the car, V2 represents the semantic embedding coding vector of the lane information where the car is located, V J Represents the multi-modal common feature representation vector of the vehicle state, W VT and b VT denote the weight parameter matrix and bias term respectively, represents addition by position point, and tanh represents the hyperbolic tangent function. That is, a common feature extraction network is constructed using a neural network architecture based on the tanh function to learn the common pattern and intrinsic connection between the cascade encoding vector of the current state of the car and the semantic embedding encoding vector of the lane information where the car is located.

[0056] In a specific example of the present application, the step S122 includes: inputting the cascade coding vector of the current state of the car, the semantic embedding coding vector of the lane information where the car is located, and the multimodal common feature representation vector of the car state into a vector probabilization unit based on the Sigmoid function to obtain a probabilistic cascade coding vector of the current state of the car, a probabilistic semantic embedding coding vector of the lane information where the car is located, and a probabilistic multimodal common feature representation vector of the car state, which can be expressed by the formula:

[0057] V a =sigmoid(V1)

[0058] Vb =sigmoid(V2)

[0059] V c =sigmoid(V J )

[0060] Among them, V a Represents the probabilistic cascade coding vector of the current state of the car, V b represents the semantic embedding encoding vector of the probabilistic lane information of the car, V c represents the multi-modal common feature representation vector of the probabilistic vehicle state, and sigmoid represents the sigmoid function. That is, the value range of the feature vector is adjusted through normalization processing, the feature distribution is smoothed, and the subsequent feature compensation modulation process is ensured to be more stable and effective.

[0061] Next, the specific features of the probabilistic cascade coding vector of the current state of the vehicle relative to the probabilistic multimodal common feature representation vector of the vehicle state are calculated to obtain the specific feature compensation representation vector of the current state of the vehicle; at the same time, the specific features of the semantic embedding coding vector of the probabilistic lane information of the vehicle relative to the probabilistic multimodal common feature representation vector of the vehicle state are calculated to obtain the specific feature compensation representation vector of the lane information of the vehicle, which can be expressed as:

[0062]

[0063] in, and denote the ith eigenvalue of the probabilistic cascade coding vector of the current state of the vehicle, the ith eigenvalue of the probabilistic semantic embedding coding vector of the lane information where the vehicle is located, and the ith eigenvalue of the probabilistic multimodal common feature representation vector of the vehicle state, log(·) denotes a logarithmic function with base 2, exp(·) denotes a natural exponential function, softmax denotes a normalized exponential function, and v a ' and v b 'respectively represent the specific feature compensation representation vector of the current state of the car and the specific feature compensation representation vector of the lane information where the car is located.

[0064] That is, the dot division vector between the probabilistic car current state cascade coding vector and the probabilistic car state multimodal common feature representation vector is calculated, and the base-two logarithmic value of the absolute value of each eigenvalue in the dot division vector is calculated to obtain the car current state specific feature vector; then, the dot product vector between the car current state specific feature vector and the probabilistic car current state cascade coding vector is calculated, and the exponential function value with base e and each eigenvalue of the dot product vector as the exponent is calculated to obtain the indexed car current state specific feature representation vector; then, the indexed car current state specific feature representation vector is input into the softmax function for normalization processing to obtain the car current state specific feature compensation representation vector.

[0065] Here, by measuring the feature differences between the probabilistic cascade coding vector of the current state of the car and the probabilistic semantic embedding coding vector of the lane information where the car is located relative to the probabilistic multimodal common feature representation vector of the car state, we can capture the unique feature information of the current state features of the car and the lane information features where the car is located in addition to the common features, strengthen the complementary information between the two, and thus generate the specific feature compensation representation vector of the current state of the car and the specific feature compensation representation vector of the lane information where the car is located.

[0066] Then, the specific feature compensation representation vector of the current state of the vehicle and the cascade coding vector of the current state of the vehicle are input into the fine-grained feature compensation module to obtain the feature compensation vector of the current state of the vehicle; at the same time, the specific feature compensation representation vector of the lane information where the vehicle is located and the semantic embedding coding vector of the lane information where the vehicle is located are input into the fine-grained feature compensation module to obtain the semantic feature compensation vector of the lane information where the vehicle is located, which can be expressed as follows:

[0067] v 1a =V1⊙v a '

[0068] v 2b =V2⊙v b '

[0069] Among them, ⊙ represents the point product by position, v 1a and v 2b They respectively represent the feature compensation vector of the current state of the car and the semantic feature compensation vector of the lane information that the car is in. That is, the feature compensation vector of the current state of the car is obtained by calculating the point-by-point multiplication between the specific feature compensation representation vector of the current state of the car and the cascade coding vector of the current state of the car.

[0070] Here, the original cascade coding vector of the car's current state and the semantic embedding coding vector of the car's lane information are compensated and strengthened through weighted operations to ensure that during the fusion process, as much unique information of the car's current state and the car's lane information features as possible is retained, thereby improving the richness of the features.

[0071] In a specific example of the present application, the step S123 includes: cascading the feature compensation vector of the current state of the vehicle, the semantic feature compensation vector of the lane information where the vehicle is located, and the multimodal common feature representation vector of the vehicle state to obtain the multimodal interactive compensation fusion representation vector of the vehicle state, which can be expressed by the formula:

[0072] v f =concat[V J ;v 1a ;v 2b ]

[0073] Among them, concat[·;·;·] represents the cascade operation, v f The multimodal interactive compensation fusion representation vector of the vehicle state is represented. That is, feature fusion is performed through cascade operation, and the common information of the vehicle's current state features and the lane information features of the vehicle are considered comprehensively, as well as the unique information in each feature source, so as to achieve a comprehensive understanding of the vehicle's driving state, which helps to more accurately predict and respond to complex traffic conditions.

[0074] In the above-mentioned intelligent start-stop control method for automobile engines, the step S130 determines whether to turn on or off the automatic start-stop function based on the multimodal interactive compensation fusion representation vector of the automobile state. In a specific example of the present application, the step S130 includes: inputting the multimodal interactive compensation fusion representation vector of the automobile state into an intelligent start-stop control module based on a classifier to determine whether to turn on or off the automatic start-stop function. Specifically, the classifier adopts a multi-layer perceptron structure, and analyzes whether the current driving environment is suitable for the activation of the automatic start-stop function by learning the characteristic patterns in the multimodal interactive compensation fusion representation vector of the automobile state, so as to output a binary classification result to indicate whether the automatic start-stop function should be activated, thereby realizing intelligent start-stop control of the automobile engine. In this way, it is possible to respond to changes in traffic conditions and driving environment in real time, optimize fuel efficiency, reduce emissions, and ensure driving comfort and safety.

[0075] In summary, the intelligent start-stop control method of the automobile engine according to the embodiment of the present application is explained, which uses deep learning technology to jointly analyze the current time, the current position and direction of the automobile, extract the current driving state characteristics of the automobile, and at the same time combine the lane information where the automobile is located, and perform compensatory interactive fusion encoding of the current driving state of the automobile and the lane information where the automobile is located, so as to fully understand the driving environment of the automobile, thereby realizing intelligent control of the automatic start-stop function of the engine. In this way, the engine wear problem caused by frequent start-stop can be effectively avoided, and the user's driving experience can be improved, thereby further improving fuel efficiency, reducing emissions, and extending the service life of the engine.

[0076] Figure 5 FIG. 1 is a block diagram of an intelligent start-stop control system for an automobile engine according to an embodiment of the present application. Figure 5 As shown, according to an embodiment of the present application, the intelligent start-stop control system 100 of the automobile engine includes: an automobile information encoding module 110, which is used to extract the current state characteristics of the automobile and the semantic characteristics of the lane information from the current time, the current position of the automobile, the current direction of the automobile and the lane where the automobile is located to obtain a cascade coding vector of the current state of the automobile and a semantic embedded coding vector of the lane information where the automobile is located; an information interaction fusion module 120, which is used to perform a compensatory interactive fusion of the cascade coding vector of the current state of the automobile and the semantic embedded coding vector of the lane information where the automobile is located with common features as a mask to obtain a multi-modal interactive compensation fusion representation vector of the automobile state; and an automatic start-stop control module 130, which is used to determine whether to turn on or off the automatic start-stop function based on the multi-modal interactive compensation fusion representation vector of the automobile state.

[0077] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned automobile engine intelligent start-stop control system have been referred to above. Figures 1 to 4 The invention has been introduced in detail in the description of the intelligent start-stop control method of an automobile engine, and therefore, its repeated description will be omitted.

[0078] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

Claims

1. An intelligent start-stop control system and method for an automobile engine, comprising: Obtain the location information of the car at regular intervals, and when the distance to the intersection is less than a preset distance threshold, collect the intersection image; It is determined through the image of the intersection that there is a countdown for the traffic light at the intersection, and when the countdown is greater than a preset time threshold and the vehicle speed is zero, the automatic start-stop function is turned on or off; when it is determined through the image of the intersection that there is no countdown for the traffic light at the intersection or there is no traffic light, the automatic start-stop function is turned on or off according to the current time, the current position of the car, the current direction of the car, and the lane where the car is located, and it is characterized in that the automatic start-stop function is turned on or off according to the current time, the current position of the car, the current direction of the car, and the lane where the car is located, including: Extract the vehicle's current state features and lane information semantic features from the current time, the vehicle's current position, the vehicle's current direction, and the lane where the vehicle is located to obtain a vehicle's current state cascade coding vector and a vehicle's lane information semantic embedding coding vector; Performing a compensatory interactive fusion of the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located with common features as a mask to obtain a multimodal interactive compensatory fusion representation vector of the vehicle state; Based on the multimodal interactive compensation fusion representation vector of the vehicle state, it is determined whether to turn on or off the automatic start-stop function.

2. The method for intelligent start-stop control of an automobile engine according to claim 1, characterized in that: Extracting the vehicle current state feature and the lane information semantic feature from the current time, the current position of the vehicle, the current direction of the vehicle and the lane where the vehicle is located to obtain the vehicle current state cascade coding vector and the vehicle lane information semantic embedding coding vector, including: One-hot encoding the current time, the current position of the car, and the current direction of the car to obtain a current time one-hot encoding vector, a current position one-hot encoding vector, and a current direction one-hot encoding vector; Cascading the current time one-hot encoding vector, the current position one-hot encoding vector and the current direction one-hot encoding vector to obtain a cascade encoding vector of the current state of the vehicle; The lane where the car is located is semantically embedded and encoded to obtain a semantic embedded encoding vector of the lane information where the car is located.

3. The automobile engine intelligent start-stop control method according to claim 2, characterized in that: The cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located are subjected to a compensatory interactive fusion with common features as a mask to obtain a multimodal interactive compensation fusion representation vector of the vehicle state, including: Extracting common features between the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located to obtain a multimodal common feature representation vector of the vehicle state; Based on the specific features of the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the information of the lane where the vehicle is located relative to the multimodal common feature representation vector of the vehicle state, the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the information of the lane where the vehicle is located perform feature compensation modulation on the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the information of the lane where the vehicle is located to obtain a feature compensation vector of the current state of the vehicle and a semantic feature compensation vector of the information of the lane where the vehicle is located; The vehicle current state feature compensation vector, the vehicle lane information semantic feature compensation vector and the vehicle state multimodal common feature representation vector are fused to obtain the vehicle state multimodal interactive compensation fusion representation vector.

4. The method for intelligent start-stop control of an automobile engine according to claim 3, characterized in that: Extracting common features between the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located to obtain a multimodal common feature representation vector of the vehicle state, including: The cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located are input into a common feature extraction network to obtain a multimodal common feature representation vector of the vehicle state.

5. The automobile engine intelligent start-stop control method according to claim 4, characterized in that: Based on the specific features of the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located relative to the multimodal common feature representation vector of the vehicle state, the cascade coding vector of the current state of the vehicle and the semantic embedding coding vector of the lane information where the vehicle is located are subjected to feature compensation modulation to obtain a feature compensation vector of the current state of the vehicle and a semantic feature compensation vector of the lane information where the vehicle is located, including: Inputting the cascade coding vector of the current state of the vehicle, the semantic embedding coding vector of the lane information where the vehicle is located, and the multimodal common feature representation vector of the vehicle state into a vector probabilization unit based on a Sigmoid function to obtain a probabilistic cascade coding vector of the current state of the vehicle, a probabilistic semantic embedding coding vector of the lane information where the vehicle is located, and a probabilistic multimodal common feature representation vector of the vehicle state; Calculating the specific features of the probabilistic automobile current state cascade coding vector relative to the probabilistic automobile state multimodal common feature representation vector to obtain a compensation representation vector of the specific features of the automobile current state; Calculating the specific features of the semantic embedding encoding vector of the probabilistic vehicle lane information relative to the probabilistic vehicle state multimodal common feature representation vector to obtain a compensation representation vector of the specific features of the vehicle lane information; Inputting the vehicle current state specific feature compensation representation vector and the vehicle current state cascade coding vector into a fine-grained feature compensation module to obtain the vehicle current state feature compensation vector; The specific feature compensation representation vector of the lane information where the vehicle is located and the semantic embedding coding vector of the lane information where the vehicle is located are input into the fine-grained feature compensation module to obtain the semantic feature compensation vector of the lane information where the vehicle is located.

6. The automobile engine intelligent start-stop control method according to claim 5, characterized in that: Calculating the specific features of the probabilistic vehicle current state cascade coding vector relative to the probabilistic vehicle state multimodal common feature representation vector to obtain a vehicle current state specific feature compensation representation vector, including: Calculate the dotted division vector between the probabilistic cascade coding vector of the current state of the vehicle and the probabilistic multimodal common feature representation vector of the state of the vehicle, and calculate the base-two logarithm of the absolute value of each eigenvalue in the dotted division vector to obtain a specific feature vector of the current state of the vehicle; Calculate the dot product vector between the specific feature vector of the current state of the vehicle and the probabilistic cascade coding vector of the current state of the vehicle, and calculate the exponential function value with e as the base and each eigenvalue of the dot product vector as the exponent to obtain the indexed specific feature representation vector of the current state of the vehicle; The indexed vehicle current state specific feature representation vector is input into a softmax function for normalization processing to obtain the vehicle current state specific feature compensation representation vector.

7. The method for intelligent start-stop control of an automobile engine according to claim 6, characterized in that: Inputting the vehicle current state specific feature compensation representation vector and the vehicle current state cascade coding vector into a fine-grained compensation module to obtain the vehicle current state feature compensation vector, including: The point-by-point multiplication between the vehicle current state specific feature compensation representation vector and the vehicle current state cascade coding vector is calculated to obtain the vehicle current state feature compensation vector.

8. The automobile engine intelligent start-stop control method according to claim 7, characterized in that: The method of fusing the vehicle current state feature compensation vector, the vehicle lane information semantic feature compensation vector and the vehicle state multimodal common feature representation vector to obtain the vehicle state multimodal interactive compensation fusion representation vector includes: The vehicle current state feature compensation vector, the vehicle lane information semantic feature compensation vector and the vehicle state multimodal common feature representation vector are cascaded to obtain the vehicle state multimodal interactive compensation fusion representation vector.

9. The automobile engine intelligent start-stop control method according to claim 8, characterized in that: Based on the multi-modal interactive compensation fusion representation vector of the vehicle state, determining whether to turn on or off the automatic start-stop function includes: The multimodal interactive compensation fusion representation vector of the vehicle state is input into a classifier-based intelligent start-stop control module to determine whether to turn on or off the automatic start-stop function.

10. An intelligent start-stop control system for an automobile engine, characterized in that: include: A vehicle information encoding module is used to extract the vehicle current state features and lane information semantic features from the current time, the vehicle current position, the vehicle current direction and the lane where the vehicle is located to obtain a cascade encoding vector of the vehicle current state and a semantic embedding encoding vector of the lane where the vehicle is located; An information interaction fusion module, used for performing a compensatory interactive fusion of the cascade coding vector of the current state of the vehicle and the semantic embedded coding vector of the lane information where the vehicle is located, using common features as a mask, to obtain a multimodal interactive compensation fusion representation vector of the vehicle state; The automatic start-stop control module is used to determine whether to turn on or off the automatic start-stop function based on the multi-modal interactive compensation fusion representation vector of the vehicle state.

Images