An intelligent automobile control system and method based on artificial intelligence

Abstract

The application discloses an intelligent automobile control system and method based on artificial intelligence, and relates to the technical field of intelligent control of new energy vehicles; brain electrical signals of a target driver are acquired, and intention feature extraction is performed to obtain driving intention instructions; multi-modal data are acquired, and environment perception feature extraction is performed to obtain environment perception parameters; the driving intention instructions and the environment perception parameters are input into a multi-modal large model for training to obtain semantic feature instructions, the semantic feature instructions are input into an intelligent agent decision model to obtain new energy vehicle control instructions; and the new energy vehicle is controlled according to a control instruction evaluation result. The brain-computer interface accurately analyzes the intention of the driver, multi-modal environment perception data are fused to realize high-precision scene understanding, and the control instructions are generated in combination with intelligent agent decision, the dynamic optimization and safety redundancy mechanism ensure the reliability of the instructions, the closed-loop feedback continuously optimizes the model, and the control intuitiveness, decision safety and system adaptability are significantly improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of intelligent control of new energy vehicles, and particularly relates to an intelligent vehicle control system and method based on artificial intelligence. BACKGROUND

[0002] With the development of intelligent driving technology, traditional physical control and interaction modes such as voice and gestures have problems of limb dependence or environmental interference. A brain-computer interface (BCI) realizes "mind driving control" by reading brain waves, providing new possibilities for the disabled, but it still faces challenges such as noise interference, low recognition accuracy and insufficient real-time performance in new energy vehicle control.

[0003] Patent No. CN119271050A discloses a brain-computer interface device, a control method thereof, a medium and an electronic device; brain signals generated by a subject in any stimulation state are acquired; feature recognition is performed on the brain signals to obtain brain response features contained in the brain signals; instruction recognition under the brain response features is performed on the brain signals to obtain target instructions represented by the brain signals; a preset execution probability is configured based on the target instructions, and the target instructions are output based on the execution probability to control external controlled devices. The application can identify target instructions from brain signals in real time through response feature recognition and instruction recognition under the response features, and output target instructions to control external controlled devices in real time, which can meet the needs of application fields with high real-time requirements.

[0004] In the prior art, new energy vehicles are controlled through a brain-computer interface, which has low instruction recognition accuracy, low real-time performance, and poor control effect of brain-computer signals on new energy vehicles. SUMMARY

[0005] The application aims to solve the problems of low instruction recognition accuracy, low real-time performance, and poor control effect of brain-computer signals on new energy vehicles in controlling new energy vehicles through a brain-computer interface, and proposes an intelligent vehicle control system and method based on artificial intelligence.

[0006] The object of the application can be achieved by the following technical solutions:

[0007] First, an intelligent vehicle control system based on artificial intelligence is proposed, which comprises an electroencephalogram signal acquisition module, an environment perception module, a new energy vehicle control instruction generation module and a new energy vehicle control module:

[0008] The brain electrical signal acquisition module is configured to acquire brain electrical signals of a target driver, perform a preset operation on the brain electrical signals to obtain a target brain electrical signal set, fuse brain electrical signals in the target brain electrical signal set to obtain a final brain electrical signal, and perform intention feature extraction on the final brain electrical signal to obtain a driving intention instruction; the preset operation is a first preset operation and a second preset operation; and the target brain electrical signal set includes a first target brain electrical signal and a second target brain electrical signal.

[0009] The environment perception module is configured to acquire multi-modal data and perform environment perception feature extraction on the multi-modal data to obtain an environment perception parameter; the multi-modal data includes image data, radar data, ultrasonic wave data, and driving data.

[0010] The new energy vehicle control instruction generation module is configured to input the driving intention instruction and the environment perception parameter into a multi-modal large model for training to obtain a semantic feature instruction, and input the semantic feature instruction into an agent decision model to obtain a new energy vehicle control instruction.

[0011] The new energy vehicle control module is configured to perform correctness verification on the new energy vehicle control instruction to obtain an evaluation result, and control the new energy vehicle according to the evaluation result.

[0012] Optionally, the new energy vehicle control instruction generation module includes a real-time state analysis module, a membership degree determination module, a fuzzy rule generation module, and a driving instruction generation module.

[0013] The real-time state analysis module is configured to acquire a real-time state of a new energy vehicle, and use the real-time state and the semantic feature instruction as input variables of a fuzzy controller to obtain an output variable; the output variable is a steering wheel rotation angle, and the real-time state includes lateral deviation and deviation rate.

[0014] The membership degree determination module is configured to generate an input fuzzy set according to the input variable, generate an output fuzzy set according to the output variable, and determine a membership function for the input fuzzy set and the output fuzzy set; the membership function is used to quantify the membership degrees of the input variable and the output variable in each fuzzy set.

[0015] The fuzzy rule generation module is configured to determine fuzzy rules, perform fuzzy rule matching according to the membership degrees of the input variables to obtain a plurality of matched fuzzy rules, and calculate the output variable according to each matched fuzzy rule to obtain a plurality of fuzzy values.

[0016] The driving instruction generation module is configured to perform a synthesis operation according to each fuzzy value to obtain a target fuzzy output, and perform de-fuzzification on the target fuzzy output to obtain a new energy vehicle control instruction; the new energy vehicle control instruction is used to adjust the rotation angle of the new energy vehicle.

[0017] Optionally, the new energy vehicle control module comprises a transfer function determination module and an evaluation result calculation module:

[0018] The transfer function determination module is configured to establish a new energy vehicle state and construct an event set, and determine a transfer function according to the new energy vehicle state and the event set; the new energy vehicle state comprises a left turn degree, a straight driving degree and a right turn degree, and the event set comprises a controllable event and an uncontrollable event.

[0019] The evaluation result calculation module is configured to determine a new energy vehicle state according to the new energy vehicle control instruction and define a constraint term, calculate an evaluation result according to the constraint term and the controllable event, and determine a control logic according to the evaluation result; the evaluation result comprises a correct, an error and a serious error.

[0020] Optionally, the evaluation result calculation module further comprises a first judgment module, a second judgment module and a third judgment module:

[0021] The first judgment module is configured to determine that the evaluation result is correct if the evaluation result is greater than or equal to a first threshold value, and then execute a driving intention instruction.

[0022] The second judgment module is configured to determine that the evaluation result is an error if the first threshold value is greater than the evaluation result and the evaluation result is greater than or equal to a second threshold value, and then reduce the weight of automatic decision.

[0023] The third judgment module is configured to determine that the evaluation result is a serious error if the second threshold value is greater than the evaluation result, and then determine the instruction through a feedback interface.

[0024] Optionally, the system further comprises a data uploading module and a data updating module:

[0025] The data uploading module is configured to periodically acquire local running data, encrypt the local running data to obtain an encrypted package, and upload the encrypted package to the cloud.

[0026] The data updating module is configured to periodically receive a cloud update package and update a multi-modal large model.

[0027] An intelligent vehicle control method based on artificial intelligence is proposed, which comprises:

[0028] The brain electrical signals of the target driver are acquired, preset operations are performed on the brain electrical signals to obtain a target brain electrical signal set, the brain electrical signals in the target brain electrical signal set are fused to obtain a final brain electrical signal, and intention feature extraction is performed on the final brain electrical signal to obtain a driving intention instruction; the preset operations are a first preset operation and a second preset operation; the target brain electrical signal set includes a first target brain electrical signal and a second target brain electrical signal;

[0029] Multi-modal data are acquired, and environment perception feature extraction is performed on the multi-modal data to obtain an environment perception parameter; the multi-modal data include image data, radar data, ultrasonic wave data and driving data;

[0030] The driving intention instruction and the environment perception parameter are input into a multi-modal large model for training to obtain a semantic feature instruction, and the semantic feature instruction is input into an agent decision model to obtain a new energy vehicle control instruction;

[0031] The new energy vehicle control instruction is subjected to correctness verification to obtain an evaluation result, and the new energy vehicle is controlled according to the evaluation result.

[0032] Optionally, inputting the semantic feature instruction into the agent decision model to obtain the new energy vehicle control instruction includes:

[0033] The real-time state of the new energy vehicle is acquired, and the real-time state and the semantic feature instruction are taken as input variables of a fuzzy controller to obtain an output variable; the output variable is a steering wheel turning angle, and the real-time state includes lateral deviation and deviation rate;

[0034] An input fuzzy set is generated according to the input variable, an output fuzzy set is generated according to the output variable, and a membership function is determined for the input fuzzy set and the output fuzzy set; the membership function is used to quantify the membership degrees of the input variable and the output variable in each fuzzy set;

[0035] Fuzzy rules are determined, a plurality of matched fuzzy rules are obtained by fuzzy rule matching according to the membership degrees of the input variable, and a plurality of fuzzy values are obtained by calculating the output variable according to each matched fuzzy rule;

[0036] Synthesis operations are performed according to the fuzzy values to obtain a target fuzzy output, and the new energy vehicle control instruction is obtained by de-fuzzification of the target fuzzy output; the new energy vehicle control instruction is used to adjust the turning angle of the new energy vehicle.

[0037] Optionally, correctness verification is performed on the new energy vehicle control instruction to obtain an evaluation result, including:

[0038] Establish a new energy vehicle state and build an event set, determine a transition function according to the new energy vehicle state and the event set; the new energy vehicle state includes: left turning degree, straight going degree and right turning degree, and the event set includes: controllable events and uncontrollable events;

[0039] According to the new energy vehicle control instruction, determine the new energy vehicle state and define a constraint term, calculate an evaluation result according to the constraint term and the controllable events, and determine the control logic according to the evaluation result; the evaluation result includes: correct, error and serious error.

[0040] Optionally, determining the control logic according to the evaluation result comprises:

[0041] If the evaluation result is greater than or equal to a first threshold value, it is determined that the evaluation result is correct, and the driving intention instruction is executed;

[0042] If the first threshold value is greater than the evaluation result and greater than or equal to a second threshold value, it is determined that the evaluation result is error, and the weight of automatic decision is reduced;

[0043] If the second threshold value is greater than the evaluation result, it is determined that the evaluation result is a serious error, and the instruction is determined through a feedback interface.

[0044] Optionally, after controlling the new energy vehicle according to the evaluation result, the method further comprises:

[0045] Periodically acquire local running data, encrypt the local running data to obtain an encrypted package, and upload the encrypted package to the cloud;

[0046] Periodically receive a cloud update package and update the multi-modal large model.

[0047] The beneficial effects of the present application are:

[0048] The present application provides an intelligent automobile control method based on artificial intelligence, which obtains the brain electrical signal of the target driver and extracts the intention feature to obtain the driving intention instruction; obtains multi-modal data and extracts the environment perception feature to obtain the environment perception parameter; inputs the driving intention instruction and the environment perception parameter into the multi-modal large model for training to obtain the semantic feature instruction, inputs the semantic feature instruction into the agent decision model to obtain the new energy vehicle control instruction, and controls the new energy vehicle according to the evaluation result. The brain-computer interface accurately analyzes the intention of the driver, fuses multi-modal environment perception data to realize high-precision scene understanding, combines the intelligent agent decision to generate the control instruction, dynamically optimizes and ensures the reliability of the instruction with the safety redundancy mechanism, continuously optimizes the model with the closed-loop feedback, and significantly improves the control intuitiveness, decision safety and system adaptability. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 A structural schematic diagram of an intelligent automobile control system based on artificial intelligence is provided for an embodiment of the present application.

[0050] Figure 2 A flowchart of an intelligent automobile control method based on artificial intelligence is provided for an embodiment of the present application.

[0051] Figure 3 A brain-computer interface signal processing flowchart of an intelligent automobile control method based on artificial intelligence is provided for an embodiment of the present application.

[0052] Figure 4 A multi-modal feature fusion flowchart of an intelligent automobile control method based on artificial intelligence is provided for an embodiment of the present application.

[0053] Figure 5 A control mode flowchart of an intelligent automobile control method based on artificial intelligence is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0055] Based on the same inventive concept, an intelligent automobile control system based on artificial intelligence is also provided in the embodiments of the present application. Referring to Figure 1 , Figure 1 A structural schematic diagram of an intelligent automobile control system based on artificial intelligence is provided for an embodiment of the present application, which comprises an electroencephalogram signal acquisition module, an environment perception module, a new energy vehicle control instruction generation module and a new energy vehicle control module:

[0056] The electroencephalogram signal acquisition module is configured to acquire electroencephalogram signals of a target driver, perform a preset operation on the electroencephalogram signals to obtain a target electroencephalogram signal set, fuse the electroencephalogram signals in the target electroencephalogram signal set to obtain a final electroencephalogram signal, and perform intention feature extraction on the final electroencephalogram signal to obtain a driving intention instruction; the preset operation is a first preset operation and a second preset operation; the target electroencephalogram signal set comprises a first target electroencephalogram signal and a second target electroencephalogram signal;

[0057] The environment perception module is configured to acquire multi-modal data and perform environment perception feature extraction on the multi-modal data to obtain environment perception parameters.

[0058] The new energy vehicle control instruction generation module is configured to input the driving intention instruction and the environmental perception parameter into a multi-modal large model to obtain a semantic feature instruction, and input the semantic feature instruction into an agent decision model to obtain a new energy vehicle control instruction.

[0059] The new energy vehicle control module is configured to verify the correctness of the new energy vehicle control instruction to obtain an evaluation result, and control the new energy vehicle according to the evaluation result.

[0060] The multi-modal data includes image data, radar data, ultrasonic data and driving data.

[0061] The intelligent automobile control system based on artificial intelligence provided by the embodiment of the application can accurately analyze the intention of the driver through a brain-computer interface, realize high-precision scene understanding by fusing multi-modal environmental perception data, and generate a control instruction in combination with an agent decision. The dynamic optimization and safety redundancy mechanism ensures the reliability of the instruction, the closed-loop feedback continuously optimizes the model, and the control intuitiveness, decision safety and system adaptability are significantly improved.

[0062] In one embodiment, the new energy vehicle control instruction generation module includes a real-time state analysis module, a membership determination module, a fuzzy rule generation module and a driving instruction generation module.

[0063] The real-time state analysis module is configured to obtain a real-time state of the new energy vehicle, and use the real-time state and the semantic feature instruction as input variables of a fuzzy controller to obtain an output variable. The output variable is a steering wheel angle, and the real-time state includes lateral deviation and deviation rate.

[0064] The membership determination module is configured to generate an input fuzzy set according to the input variable, generate an output fuzzy set according to the output variable, and determine a membership function for the input fuzzy set and the output fuzzy set. The membership function is used to quantify the membership of the input variable and the output variable in each fuzzy set.

[0065] The fuzzy rule generation module is configured to determine fuzzy rules, match the fuzzy rules according to the membership of the input variable to obtain a plurality of matched fuzzy rules, and calculate the output variable according to each matched fuzzy rule to obtain a plurality of fuzzy values.

[0066] The driving instruction generation module is configured to perform a synthesis operation according to each fuzzy value to obtain a target fuzzy output, and de-fuzzify the target fuzzy output to obtain a new energy vehicle control instruction. The new energy vehicle control instruction is used to adjust the steering angle of the new energy vehicle.

[0067] In one embodiment, the new energy vehicle control module includes a transfer function determination module and an evaluation result calculation module.

[0068] The transfer function determination module is configured to establish a new energy vehicle state and construct an event set, and determine a transfer function according to the new energy vehicle state and the event set; the new energy vehicle state includes a left turning degree, a straight going degree and a right turning degree, and the event set includes controllable events and uncontrollable events.

[0069] The evaluation result calculation module is configured to determine a new energy vehicle state according to a new energy vehicle control instruction and define a constraint term, calculate an evaluation result according to the constraint term and the controllable events, and determine a control logic according to the evaluation result; the evaluation result includes correct, error and serious error.

[0070] In one embodiment, the evaluation result calculation module further includes a first judgment module, a second judgment module and a third judgment module.

[0071] The first judgment module is configured to determine that the evaluation result is correct if the evaluation result is greater than or equal to a first threshold value, and then execute a driving intention instruction.

[0072] The second judgment module is configured to determine that the evaluation result is error if the first threshold value is greater than the evaluation result and the evaluation result is greater than or equal to a second threshold value, and then reduce the weight of automatic decision.

[0073] The third judgment module is configured to determine that the evaluation result is serious error if the second threshold value is greater than the evaluation result, and then determine the instruction through a feedback interface.

[0074] In one embodiment, the system further includes a data uploading module and a data updating module.

[0075] The data uploading module is configured to periodically acquire local running data, encrypt the local running data to obtain an encrypted package, and upload the encrypted package to the cloud.

[0076] The data updating module is configured to periodically receive a cloud update package and update the multi-modal large model.

[0077] The embodiment of the present application provides an intelligent automobile control method based on artificial intelligence. Referring to Figure 2 , Figure 2 The embodiment of the present application provides a flow chart of an intelligent automobile control method based on artificial intelligence. The method includes the following steps:

[0078] S101, acquiring an electroencephalogram signal of a target driver, performing a preset operation on the electroencephalogram signal to obtain a target electroencephalogram signal set, fusing the electroencephalogram signals in the target electroencephalogram signal set to obtain a final electroencephalogram signal, and performing intention feature extraction on the final electroencephalogram signal to obtain a driving intention instruction;

[0079] S102, acquiring multi-modal data, and performing environment perception feature extraction on the multi-modal data to obtain an environment perception parameter;

[0080] S103, input the driving intention instruction and the environment perception parameter into the multi-modal big model for training to obtain a semantic feature instruction, and input the semantic feature instruction into an agent decision model to obtain a new energy vehicle control instruction;

[0081] S104, performing correctness verification on the new energy vehicle control instruction to obtain an evaluation result, and controlling the new energy vehicle according to the evaluation result.

[0082] The preset operation is a first preset operation and a second preset operation; the target electroencephalogram set includes a first target electroencephalogram and a second target electroencephalogram; and the multi-modal data includes image data, radar data, ultrasonic data and driving data.

[0083] The intelligent automobile control method based on artificial intelligence provided in the embodiment of the application can accurately analyze the intention of the driver through the brain-computer interface, realize high-precision scene understanding by fusing multi-modal environment perception data, and generate a control instruction in combination with an agent decision, so that the reliability of the instruction is ensured by dynamic optimization and a safety redundancy mechanism, the model is continuously optimized through closed-loop feedback, and the intuitive operation, decision safety and system adaptability are significantly improved.

[0084] In an implementation manner, a preset operation is performed on the electroencephalogram to obtain a target electroencephalogram set, for example, discrete wavelet transform enhancement and empirical mode decomposition, to respectively obtain a first target electroencephalogram and a second target electroencephalogram; wherein the discrete wavelet transform enhancement is to decompose the electroencephalogram to obtain sub-signals of different frequency bands, that is, to decompose into approximation coefficients (low-frequency components, reflecting the overall trend of the signal) and detail coefficients (high-frequency components, reflecting the local fluctuation of the signal), and to combine the electroencephalograms of the same category to obtain target coefficients (the database contains standard approximation coefficients and standard detail coefficients, and the standard approximation coefficients and the detail coefficients of the target user (driver) are combined, or the standard detail coefficients and the approximation coefficients of the target user (driver) are combined), so as to cope with individual differences by fusing the signal characteristics of different subjects; inverse discrete wavelet transform is performed on the recombined target coefficients to generate enhanced source subject signals and target subject signals, and finally the first target electroencephalogram enhanced in the time domain is obtained.

[0085] In an implementation manner, the electroencephalogram is decomposed through empirical mode decomposition to obtain a group of intrinsic mode functions and a residual error. The intrinsic mode function is a sub-signal of different frequency bands (reflecting the local characteristics of the signal in different frequency ranges), and the residual error is the overall trend of the signal; for electroencephalograms of the same category, the standard intrinsic mode function and the intrinsic mode function of the target user (driver) are combined to fuse the time-frequency information of the subjects to process individual differences; the recombined intrinsic mode function and the corresponding residual error are summed to generate the second electroencephalogram enhanced in the time domain, and the first target electroencephalogram and the second electroencephalogram are fused to obtain the final electroencephalogram.

[0086] In an implementation, the brain-computer interface module is used to collect the brain signals of the driver in real time (for example, a dry electrode EEG headset or an implanted brain-computer interface system is used to collect the brain activity signals of the driver in real time, see Figure 3 , Figure 3 The brain-computer interface signal processing flowchart of the intelligent automobile control method based on artificial intelligence provided by the embodiment of the application; wherein butterworth is a Butterworth filter), and an adaptive feature extraction technology (for example, wavelet transform or a machine learning algorithm) is used to generate a driving intention instruction. The driver can directly participate in the control of the new energy vehicle through brain signals, which provides the possibility of barrier-free driving for people with physical disabilities, and reduces the dependence on limb movements in regular driving, significantly improving the intuitiveness of control.

[0087] In an implementation, see Figure 4 , Figure 4 The multi-modal feature fusion flowchart of the intelligent automobile control method based on artificial intelligence provided by the embodiment of the application; wherein Cross-Attenion represents cross-attention, and Perceiver is a general framework suitable for multi-modal learning; the multi-modal deep learning model (i.e., a multi-modal large model) can use a neural network architecture such as Transformer, and is trained in advance on a large amount of driving scene data and brain intention signal data. In the running process, the multi-modal fusion module takes the brain-computer interface features and the environment perception features as inputs, and performs feature association and semantic fusion through the multi-layer attention mechanism of the model, and outputs a comprehensive semantic representation of the current scene. For example, at a certain moment, the brain-computer interface features indicate that the driver has a lane change intention, and the environment perception features indicate that there is a nearby new energy vehicle on the left lane, then the multi-modal large model can fuse the two pieces of information and output a high-level semantic representation of “driver intention left lane change but left side has a vehicle”. This representation can include the current decision intention of the driver, the urgency assessment, and the key factors related to the intention in the environment (such as new energy vehicles, pedestrians, road signs, etc.) and risk prediction.

[0088] In one implementation, by integrating multi-modal data such as vehicle-mounted cameras, radars, ultrasonic sensors, etc. (multiple vehicle-mounted cameras for acquiring images of the front and surroundings of the new energy vehicle; millimeter wave radar and laser radar for detecting distance and speed information of obstacles around the new energy vehicle; ultrasonic sensor for close-range ranging (such as parking assistance); and internal state sensors of the new energy vehicle, such as steering wheel angle sensor, pedal position sensor, etc., to acquire the current operation behavior of the driver, that is, driving data), combined with a deep learning model (such as a convolutional neural network) to extract environmental perception feature instructions (instructions, that is, feature vectors), the technology realizes all-around dynamic perception of complex driving scenes, including obstacle distance, lane line recognition, traffic sign detection, etc., ensuring that the system can still accurately understand the environmental state under complex road conditions, providing a reliable data basis for subsequent decision-making.

[0089] In one implementation, based on a multi-modal large model (such as a Transformer architecture), the driving intention instructions and environmental perception parameters are semantically fused to generate a comprehensive representation containing intention priority, environmental risk, and scene semantics. The agent decision-making model (such as a reinforcement learning agent) generates control instructions based on this, which can both respond to the driver's intention and dynamically adjust the strategy in combination with safety rules. This technology realizes the deep cooperation of human-machine intention, both retaining the subjective decision-making ability of the driver and ensuring the safety and adaptability of the autonomous driving system.

[0090] In one implementation, the correctness of the control instructions is verified through an evaluation mechanism, and the corrected control instructions are generated based on the scoring results. This technology combines a redundant control architecture (such as a dual-channel switching mechanism) and a safety monitoring module, automatically triggers a safety mode or requests manual takeover when the brain-computer interface is abnormal or the instructions conflict, ensuring that the new energy vehicle is always in a controlled state, significantly improving the fault tolerance of the system. The control system is for assisted driving, and all operations need to be verified by the driver through voice, display, etc. before execution.

[0091] In one implementation, the multi-modal large model and the agent decision-making model are continuously optimized through a federated learning mechanism, cross-new energy vehicle data sharing is achieved through edge computing and cloud aggregation, the adaptability of the model to different driver brain signal features and diversified road scenes is gradually improved, the system has self-evolution ability, and long-term technical advancement and personalized service ability are maintained.

[0092] In one embodiment, inputting the semantic feature instructions into the agent decision-making model to obtain the new energy vehicle control instructions includes:

[0093] Obtain the real-time state of the new energy vehicle, and take the real-time state and semantic feature instruction as input variables of the fuzzy controller to obtain output variables; the output variables are steering wheel turning angles, and the real-time state includes lateral deviation and deviation rate;

[0094] Generate input fuzzy sets according to the input variables, generate output fuzzy sets according to the output variables, and determine membership functions of the input fuzzy sets and the output fuzzy sets; the membership functions are used to quantify the membership degrees of the input variables and the output variables in the respective fuzzy sets;

[0095] Determine fuzzy rules, perform fuzzy rule matching according to the membership degrees of the input variables to obtain a plurality of matching fuzzy rules, and calculate the output variables according to the respective matching fuzzy rules to obtain a plurality of fuzzy values;

[0096] Perform a synthesis operation according to the respective fuzzy values to obtain a target fuzzy output, and de-fuzzify the target fuzzy output to obtain a new energy vehicle control instruction; the new energy vehicle control instruction is used to adjust the turning angle of the new energy vehicle.

[0097] In an implementation manner, the distance of the new energy vehicle from the center line of the road (lateral deviation), the deviation rate and the semantic feature instruction are selected as the input variables, and the steering angle is selected as the output variable; the lateral position and the deviation rate of the new energy vehicle are key factors affecting the driving stability, and the driving state of the new energy vehicle can be monitored in real time through the two variables; the steering angle is a direct parameter for controlling the driving direction of the new energy vehicle, and provides necessary input information for the fuzzy controller, so that the fuzzy controller can generate a corresponding control decision according to the current state of the new energy vehicle.

[0098] In an implementation manner, input fuzzy sets are generated according to the input variables, and output fuzzy sets are generated according to the output variables; the fuzzy sets of the input variables f (lateral deviation) and ef (deviation rate) and the fuzzy set of the output variable v (steering angle) are defined; for example, the fuzzy set of f is {MC, MO, MT, AP, QT, QO, QC}, which respectively represent “large left deviation”, “medium left deviation”, “small left deviation”, “zero”, “small right deviation”, “medium right deviation” and “large right deviation”; the fuzzy set of ef is {MC, MO, MT, AP, QT, QO, QC}, which respectively represent “large left deviation rate”, “medium left deviation rate”, “small left deviation rate”, “zero rate”, “small right deviation rate”, “medium right deviation rate” and “large right deviation rate”; the fuzzy set of v is {MC, MO, MT, AP, QT, QO, QC}, which respectively represent “large left turn”, “medium left turn”, “small left turn”, “zero turning angle”, “small right turn”, “medium right turn” and “large right turn”; the fuzzy sets are used to convert precise numerical inputs into fuzzy states, so that the fuzzy controller can process uncertainty and fuzziness, provide a basis for fuzzy reasoning, and enable the controller to generate reasonable control decisions according to the fuzzy rules.

[0099] In one implementation, membership functions are determined for the input fuzzy set and the output fuzzy set; the membership functions are used to quantify the degree of membership of the input and output variables in each fuzzy set and are the basis of fuzzy inference; the precise input values ​​are converted into fuzzy states so that the fuzzy controller can perform inference according to fuzzy rules.

[0100] The membership function of f (lateral deviation) is, for example:

[0101] Membership MC MO MT AP QT QO QC 0 1 0 0 0 0 0 0 0.2 0 0 0 0 0 1 1 0.4 -2 0 0 2.5 2 5 2

[0102] In one implementation, fuzzy rules are determined, for example:

[0103] f\ef MC MO MT AP QT QO QC MC QC QC QO QO QT QT AP MO QC QO QO QT QT AP MT MT QO QO QT QT AP MT MT AP QO QT QT AP MT MT MO QT QT QT AP MT MT MO MO QO QT AP MT MT MO MO MC QC AP MT MT MO MO MC MC

[0104] In one implementation, the fuzzy value of the output variable v is calculated based on the matched fuzzy rules. For example, if the input variables f and ef belong to fuzzy sets B and C respectively, then the fuzzy value D of the output variable v is, for example: D=(B*C)∘S, where ∘ represents the composition operation and S represents the fuzzy rule table. A synthesis operation is performed based on each fuzzy value to obtain the target fuzzy output. The max-min method is then used to perform a synthesis operation on all matched fuzzy rules to obtain the final fuzzy output.

[0105] μV(A)=max{min[μB(x),μC(y)]}

[0106] Fuzzy inference is the core component of a fuzzy controller. It is used to generate fuzzy values ​​of output variables based on fuzzy rules and the fuzzy states of input variables. Through fuzzy inference, the controller can generate reasonable control decisions based on the states of input variables, thereby achieving dynamic control of new energy vehicles.

[0107] In one implementation, the target fuzzy output is defuzzified to obtain the control decision value, and the fuzzy output value obtained from fuzzy inference is converted into a precise control decision value.

[0108]

[0109] Where v represents the defuzzified precise control decision value, μ V (An) is the output variable v in the fuzzy set A. n Membership degree in A n N represents an element in the fuzzy set of the output variable v, and N represents the total number of elements in the fuzzy set of the output variable v.

[0110] The result of fuzzy reasoning is a fuzzy value, while the actual control system needs accurate numerical values as control instructions. Converting the fuzzy output value into an accurate control decision value enables the controller to output specific control instructions, thereby achieving accurate control of the new energy vehicle. According to the accurate control decision value v after defuzzification, the steering angle of the new energy vehicle is adjusted. The steering angle is a key parameter for lateral control of the new energy vehicle, and by adjusting the steering angle, accurate control of the driving trajectory of the new energy vehicle can be achieved. The output of the controller directly affects the driving state of the new energy vehicle, and through accurate adjustment of the steering angle, accurate control of the new energy vehicle can be achieved to make it travel according to the predetermined trajectory.

[0111] In one embodiment, the correctness verification of the new energy vehicle control instruction obtains an evaluation result, which includes:

[0112] The new energy vehicle state is established and an event set is constructed, and a transition function is determined according to the new energy vehicle state and the event set. The new energy vehicle state includes: left turning degree, straight going degree and right turning degree, and the event set includes: controllable event and uncontrollable event.

[0113] The new energy vehicle state is determined according to the new energy vehicle control instruction, and a constraint term is defined. The evaluation result is calculated according to the constraint term and the controllable event, and the control logic is determined according to the evaluation result. The evaluation result includes: correct, error and serious error.

[0114] In one implementation, the new energy vehicle state R is established to describe the lateral motion of the brain-controlled new energy vehicle system, R={r L ,r D ,r R}, the new energy vehicle state includes: left turning degree r L , straight going degree r D and right turning degree r R ; the event set is defined as Q, including controllable event Q kk and uncontrollable event Q bk , the controllable event Q kk includes receiving the brain control instruction of the driver (left turn, keep direction, right turn), the uncontrollable event Q bk represents the interval without receiving the brain control instruction; Q=Q kk ∪Q bk ,Q kk ∩Q bk =∅, Q kk ={q L ,q D ,q R}, wherein q L ,q D ,q Rrespectively represent the brain control instructions for receiving left turn, straight, and right turn; a transition function is determined according to the new energy vehicle state and the event set, the transition function: R*Q→R', through fuzzy reasoning, the system can dynamically adjust the state according to the current state and the input event, so as to realize real-time supervision of the brain control instruction.

[0115] In an implementation manner, F(q)=μr(q), wherein F(q) is a constraint term, μr(q) is a product of a membership function value of the deviation f and the deviation rate ef obtained through fuzzy reasoning, and the value is between [0, 1]; the evaluation result is greater than or equal to 0.7 (the first threshold value), and it is determined that the evaluation result is correct; 0.3 is less than or equal to the evaluation result and less than 0.7, and it is determined that the evaluation result is wrong; and the evaluation result is less than 0.3, and it is determined that the evaluation result is a serious error.

[0116] In an implementation manner, the evaluation result is calculated according to the constraint term and the controllable event, for example: R(q)=F(q)∩Q kk (q), wherein R(q) represents the updated new energy vehicle state, F(q) represents the instruction correctness constraint term, Q kk (q) represents the current event (instruction); the transition function combines the driver instruction with the real-time state of the new energy vehicle, and determines whether to accept the instruction or trigger the correction mechanism. For example, if the instruction causes the new energy vehicle to continuously deviate from the lane, the transition function can ignore the instruction and switch to the automatic control mode.

[0117] In an implementation manner, by establishing the new energy vehicle state and the event set, and determining the transition function according to the new energy vehicle state and the event set, the system can dynamically adjust the control state of the new energy vehicle according to the current state and the input event. This dynamic adjustment capability enables the system to quickly respond when receiving the brain control instruction, and to remain stable when not receiving the instruction. If the event set is an uncontrollable event (when the intention information is uncertain or the signal is lost), a voice prompt is sent to the target driver, and if no reply is received within a preset time period, the vehicle speed is gradually reduced to zero;

[0118] In an implementation manner, refer to Figure 5 , Figure 5This invention provides a flowchart of the control mode of an artificial intelligence-based intelligent vehicle control method, where SNR represents the signal-to-noise ratio. The brain-computer interface control channel and the traditional manual control channel coexist in the control of new energy vehicles, and are always ready to receive control signals directly input by the driver via the steering wheel and pedals. Under normal circumstances, the intelligent decision-making unit controls the vehicle based on the brain-computer interface and the autonomous driving strategy. Once an abnormality in the brain-computer interface signal is detected (e.g., driver inattention leading to invalid EEG features) or the system detects that the agent's decision may lead to an unsafe situation, the system will automatically switch control weights: prompting the driver to intervene and allowing the new energy vehicle control unit to temporarily operate in an autonomous driving safety mode (e.g., deceleration and stopping). This dual-channel redundancy ensures that even if the new interaction method fails, the new energy vehicle can still be controlled by conventional means, preventing loss of control.

[0119] In one embodiment, determining control logic based on the evaluation results includes:

[0120] If the evaluation result is greater than or equal to the first threshold, the evaluation result is deemed correct, and the driving intention command is executed.

[0121] If the first threshold > the evaluation result ≥ the second threshold, the evaluation result is determined to be incorrect, and the weight of the automatic decision is reduced.

[0122] If the second threshold is greater than the evaluation result, the evaluation result is determined to be a serious error, and instructions are given through the feedback interface.

[0123] In one implementation, an intelligent decision-making unit executes control logic, generating specific control decisions and interaction strategies for the new energy vehicle based on the fused semantic representation. By combining the driver's intent priority information, the intelligent decision-making unit can quickly generate corresponding instructions, such as a strong braking intent, when it detects a clear control command intent from the driver. This mechanism not only improves the system's response speed and accuracy to driver intent but also enhances the human-machine collaborative decision-making capability, enabling the control behavior of the new energy vehicle to better reflect the driver's subjective intent.

[0124] In one implementation, the intelligent decision-making unit considers both driving task objectives and safety rules when generating control decisions. When there is a conflict between human and machine intentions, the intelligent agent can initiate a negotiation strategy, such as asking the driver through a feedback interface whether to confirm a certain action, or partially executing the driver's instructions while ensuring safety. This safety-first decision-making mechanism ensures that new energy vehicles adhere to safe driving principles under all circumstances. Even when the driver's intentions are inconsistent with safety rules, the system can still ensure driving safety by verifying or adjusting instructions, thus improving the system's reliability and safety.

[0125] In one implementation, the intelligent decision-making unit can output low-level control commands, including steering angle, target vehicle speed, and braking intensity, or output strategic decisions at a higher level (such as suggesting lane changes or maintaining a safe following distance). This flexibility allows the system to dynamically adjust its control strategy based on a comprehensive representation of the current scenario and the driving task objective. For example, in complex traffic environments, the system can suggest lane changes to avoid potential hazards or adjust vehicle speed to maintain a safe following distance. This strategy adjustment capability not only improves the system's adaptability but also enhances its performance and reliability in different driving scenarios.

[0126] In one embodiment, after controlling the new energy vehicle based on the evaluation results, the method further includes:

[0127] Periodically retrieve local running data, encrypt the local running data to obtain an encrypted package, and upload the encrypted package to the cloud;

[0128] It periodically receives update packages from the cloud and updates the multimodal large model.

[0129] In one implementation, each new energy vehicle periodically uses locally collected brain-computer interface data and driving data to train the model and obtain local model update parameters; the local update parameters are uploaded to the server for aggregation calculation to obtain global model updates; the updated model is then distributed to each new energy vehicle for application in subsequent control, thereby gradually improving the model's adaptability to different drivers and diverse road environments.

[0130] In one implementation, federated learning continuously updates the multimodal large model and the agent decision-making model to adapt to the brain-computer interface signal characteristics of different drivers and the road environment in different regions. Each vehicle acts as an edge node, periodically collecting local operating data for model fine-tuning training, and uploading the encrypted model updates (gradients or parameters) to a cloud-based federated server for aggregation. The federated server aggregates the update information from multiple vehicles, calculates global model improvements, and distributes the updated model parameters to each new energy vehicle, enhancing the system's robustness.

[0131] It should be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0132] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. An intelligent vehicle control system based on artificial intelligence, characterized in that, The system includes: an EEG signal acquisition module, an environmental perception module, a new energy vehicle control command generation module, and a new energy vehicle control module. The EEG signal acquisition module is used to acquire the EEG signal of the target driver, perform preset operations on the EEG signal to obtain a target EEG signal set, fuse the EEG signals in the target EEG signal set to obtain a final EEG signal, and extract intention features from the final EEG signal to obtain a driving intention command; the preset operations are subject to a first preset operation and a second preset operation; the target EEG signal set includes a first target EEG signal and a second target EEG signal; The EEG signal is decomposed to obtain approximation coefficients and detail coefficients. Standard approximation coefficients and standard detail coefficients are obtained from the database. The standard approximation coefficients are combined with the approximation coefficients to obtain the target approximation coefficients. The standard detail coefficients are combined with the detail coefficients to obtain the target detail coefficients. The target approximation coefficients and target detail coefficients are used as the target coefficients. The inverse discrete wavelet transform is performed on the target coefficients to obtain the first target EEG signal. EEG signals are decomposed using empirical mode decomposition to obtain intrinsic mode functions and residuals. Standard intrinsic mode functions are then obtained. The intrinsic mode functions are combined with the standard intrinsic mode functions to obtain reconstructed intrinsic mode functions. Finally, the reconstructed intrinsic mode functions are summed with the residuals to obtain the second EEG signal. The environmental perception module is used to acquire multimodal data and extract environmental perception parameters from the multimodal data; the multimodal data includes: image data, radar data, ultrasonic data, and driving data; The new energy vehicle control command generation module is used to input the driving intention command and the environmental perception parameters into a multimodal large model for training to obtain semantic feature commands, and input the semantic feature commands into an intelligent agent decision model to obtain new energy vehicle control commands. The new energy vehicle control module is used to verify the correctness of the new energy vehicle control commands, obtain an evaluation result, and control the new energy vehicle based on the evaluation result.

2. The intelligent vehicle control system based on artificial intelligence according to claim 1, characterized in that, The new energy vehicle control command generation module includes: a real-time status analysis module, a membership determination module, a fuzzy rule generation module, and a driving command generation module. The real-time status analysis module is used to acquire the real-time status of the new energy vehicle, and use the real-time status and the semantic feature command as input variables of the fuzzy controller to obtain the output variable; the output variable is the steering wheel angle, and the real-time status includes: lateral deviation and deviation rate; The membership determination module is used to generate an input fuzzy set based on the input variable, generate an output fuzzy set based on the output variable, and determine a membership function for the input fuzzy set and the output fuzzy set; the membership function is used to quantify the membership degree of the input variable and the output variable in each fuzzy set. The fuzzy rule generation module is used to determine fuzzy rules, perform fuzzy rule matching based on the membership degree of the input variable to obtain multiple matching fuzzy rules, and calculate multiple fuzzy values ​​for the output variable based on each matching fuzzy rule; The driving instruction generation module is used to perform a synthesis operation based on each fuzzy value to obtain a target fuzzy output, and to defuzzify the target fuzzy output to obtain a new energy vehicle control instruction; the new energy vehicle control instruction is used to adjust the steering angle of the new energy vehicle.

3. The intelligent vehicle control system based on artificial intelligence according to claim 1, characterized in that, The new energy vehicle control module includes: a transfer function determination module and an evaluation result calculation module. The transfer function determination module is used to establish the state of the new energy vehicle and construct an event set, and determine the transfer function based on the new energy vehicle state and the event set; the new energy vehicle state includes: left turn degree, straight-ahead degree, and right turn degree, and the event set includes: controllable events and uncontrollable events; The evaluation result calculation module is used to determine the state of the new energy vehicle and define constraints according to the new energy vehicle control command, calculate the evaluation result according to the constraints and the controllable event, and determine the control logic according to the evaluation result; the evaluation result includes: correct, incorrect and serious error.

4. The intelligent vehicle control system based on artificial intelligence according to claim 3, characterized in that, The evaluation result calculation module includes: a first judgment module, a second judgment module, and a third judgment module. The first judgment module is used to determine that the evaluation result is correct if the evaluation result is ≥ a first threshold, and then execute the driving intention command; The second judgment module is used to determine that the evaluation result is wrong if the first threshold > the evaluation result ≥ the second threshold, and then reduce the weight of the automatic decision. The third judgment module is used to determine that the evaluation result is a serious error if the second threshold is greater than the evaluation result, and then confirm the instruction through the feedback interface.

5. The intelligent vehicle control system based on artificial intelligence according to claim 1, characterized in that, The system also includes: a data upload module and a data update module. The data upload module is used to periodically acquire local running data, encrypt the local running data to obtain an encrypted package, and upload the encrypted package to the cloud. The data update module is used to periodically receive update packages from the cloud and update the multimodal large model.

6. An intelligent vehicle control method based on artificial intelligence, characterized in that, The method includes: The system acquires the brainwave signals of a target driver, performs preset operations on the brainwave signals to obtain a target brainwave signal set, fuses the brainwave signals in the target brainwave signal set to obtain a final brainwave signal, and extracts intention features from the final brainwave signal to obtain a driving intention command; the preset operations are subject to a first preset operation and a second preset operation; the target brainwave signal set includes a first target brainwave signal and a second target brainwave signal. The EEG signal is decomposed to obtain approximation coefficients and detail coefficients. Standard approximation coefficients and standard detail coefficients are obtained from the database. The standard approximation coefficients are combined with the approximation coefficients to obtain the target approximation coefficients. The standard detail coefficients are combined with the detail coefficients to obtain the target detail coefficients. The target approximation coefficients and target detail coefficients are used as the target coefficients. The inverse discrete wavelet transform is performed on the target coefficients to obtain the first target EEG signal. EEG signals are decomposed using empirical mode decomposition to obtain intrinsic mode functions and residuals. Standard intrinsic mode functions are then obtained. The intrinsic mode functions are combined with the standard intrinsic mode functions to obtain reconstructed intrinsic mode functions. Finally, the reconstructed intrinsic mode functions are summed with the residuals to obtain the second EEG signal. Acquire multimodal data, and extract environmental perception parameters from the multimodal data by performing environmental perception feature extraction; the multimodal data includes: image data, radar data, ultrasonic data, and driving data; The driving intention command and the environmental perception parameters are input into a multimodal large model for training to obtain semantic feature commands, and the semantic feature commands are input into an intelligent agent decision model to obtain new energy vehicle control commands; The correctness of the control commands for the new energy vehicles is verified to obtain an evaluation result, and the new energy vehicles are controlled according to the evaluation result.

7. The intelligent vehicle control method based on artificial intelligence according to claim 6, characterized in that, The semantic feature instructions are input into the intelligent agent decision model to obtain new energy vehicle control instructions, including: The real-time status of the new energy vehicle is acquired, and the real-time status and the semantic feature instruction are used as input variables of the fuzzy controller to obtain the output variable; the output variable is the steering wheel angle, and the real-time status includes: lateral deviation and deviation rate; An input fuzzy set is generated based on the input variables, and an output fuzzy set is generated based on the output variables. A membership function is determined for the input fuzzy set and the output fuzzy set. The membership function is used to quantify the membership degree of the input variables and the output variables in each fuzzy set. Determine fuzzy rules, perform fuzzy rule matching based on the membership degree of the input variable to obtain multiple matching fuzzy rules, and calculate multiple fuzzy values ​​for the output variable based on each matching fuzzy rule; A synthesis operation is performed based on each fuzzy value to obtain a target fuzzy output. The target fuzzy output is then defuzzified to obtain a new energy vehicle control command. The new energy vehicle control command is used to adjust the steering angle of the new energy vehicle.

8. The intelligent vehicle control method based on artificial intelligence according to claim 6, characterized in that, The correctness verification of the control commands for the new energy vehicles yielded evaluation results, including: Establish the status of new energy vehicles and construct an event set, and determine the transfer function based on the new energy vehicle status and the event set; the new energy vehicle status includes: left turn degree, straight-going degree, and right turn degree, and the event set includes: controllable events and uncontrollable events; The new energy vehicle status is determined and constraint terms are defined according to the new energy vehicle control command. An evaluation result is calculated based on the constraint terms and the controllable event. The control logic is determined based on the evaluation result. The evaluation result includes: correct, incorrect, and critical error.

9. The intelligent vehicle control method based on artificial intelligence according to claim 8, characterized in that, The control logic is determined based on the evaluation results, including: If the evaluation result is greater than or equal to the first threshold, the evaluation result is determined to be correct, and the driving intention command is executed. If the first threshold > the evaluation result ≥ the second threshold, then the evaluation result is determined to be incorrect, and the weight of the automatic decision is reduced. If the second threshold is greater than the evaluation result, the evaluation result is determined to be a serious error, and instructions are given through the feedback interface.

10. The intelligent vehicle control method based on artificial intelligence according to claim 6, characterized in that, After controlling the new energy vehicles based on the evaluation results, the method further includes: Periodically acquire local operating data, encrypt the local operating data to obtain an encrypted package, and upload the encrypted package to the cloud; It periodically receives update packages from the cloud and updates the multimodal large model.

Images