Method for implementing torque control using a neural network and system

Abstract

Method for implementing a torque control using a neural network (NN), comprising: in response to a torque request (225), executing, by a processor (220), a torque prediction model (210) of the NN to receive a set of measured vehicle operating inputs (205) associated with the torque prediction; replacing a set of multiple independent variables of the torque prediction model (210) implemented by the NN with a simplified mathematical expression in a pseudo-NN containing a reduced set of variables including at least one independent variable to replace a previous set of multiple independent variables of the NN;Processing the set of measured vehicle operating inputs by the pseudo-NN based on the torque prediction model (210) using at least one independent variable and one or more constants that replace a set of substituted independent variables in a simplified mathematical expression of the pseudo-NN; and solving at least one root of the simplified mathematical expression of the pseudo-NN by obtaining a root value without having to rely on an inversion operation of a mathematical expression consisting of a whole set of independent variables to obtain a requested torque value.

Description

The present description refers generally to vehicles and in particular to methods, systems and devices for adopting an artificial neural network (ANN) for torque prediction and subsequent application of torque control in a manner that avoids a complex mathematical inversion by reducing the number of variables of the ANN to a single variable(s), thereby creating a “pseudo ANN” that enables the solution of the variable root for extended process control variables. Engine control systems are designed to precisely control the engine's torque output to achieve the desired engine torque. Traditionally, torque prediction models, when implemented, require an "inversion" of the results for use in the control system. That is, only with a subsequent output inversion step can the results from a torque prediction model be used to enable responsive vehicle control. There are desirable prediction methods that do not require the inversion step when implementing a torque prediction model, as the need for an inversion process can prove to be unnecessarily cumbersome and difficult to implement. This is particularly true when using a prediction model that is highly nonlinear, such as an ANN. German patent DE 10 2007 020 355 A1 describes an engine control system that includes a torque demand control module to determine a first engine torque demand. A torque demand module for artificial neural networks (ANNs) determines a second engine torque demand using an ANN model. A torque safety test module optionally generates a malfunction signal based on the difference between the first and second engine torque demands. The task can be considered to be the specification of a method and a system for torque control that artificial neural networks solve for their roots via ignition (spark) or air-per-cylinder (APC) control variables. The torque control method aims to enable ANN-based torque prediction, which can then also be used for control. The problem is solved by a method according to claim 1 and a system according to claim 8. Furthermore, a vehicle device is described into which the system according to the invention can be integrated. An inventive method for implementing torque control using a neural network (NN) is provided. The method comprises, in response to a torque request, executing a torque prediction model of the NN by a processor to receive a set of measured vehicle operating inputs related to the torque prediction; replacing a set of multiple independent variables of the torque prediction model implemented by the NN with a simplified mathematical expression in a pseudo-NN containing a reduced set of variables, including at least one independent variable, to replace a previous set of multiple independent variables of the NN;Processing the set of measured vehicle operating inputs by the pseudo-NN based on the torque prediction model using at least one independent variable and one or more constants substituted for a set of substituted independent variables in a simplified mathematical expression of the pseudo-NN; and solving at least one root of the simplified mathematical expression of the pseudo-NN by obtaining a root value without relying on an inversion operation of a mathematical expression consisting of an entire set of independent variables to obtain a requested torque value. In one embodiment, the method further comprises the processor solving at least one root of the simplified mathematical expression of the pseudo-NN to obtain a root value in order to predict a change in an air-per-cylinder (APC) control to obtain the requested torque value. In at least one embodiment, the method further comprises solving at least one root of the simplified mathematical expression of the pseudo-NN by the processor to obtain a root value in order to predict a difference in an ignition control (spark / SPARK control) in order to change the engine timings to obtain the desired torque value. In at least one embodiment, the method further comprises solving at least one root of the simplified mathematical expression of the pseudo-NN without having to resort to an inversion operation despite a high degree of nonlinearity of the NN. In at least one embodiment, the method further comprises the application of the torque prediction model by the processor, including the composition of an activation function with multiple hidden layers, wherein a root resolution algorithm is used to solve the simplified mathematical expression for a single remaining variable. In at least one embodiment, the method further includes the set of measured vehicle operating inputs comprising intake camshafts (ICAM), exhaust camshafts (ECAM), revolutions per minute (RPM), air per cylinder (APC), and ignition timing of a vehicle in operation. In at least one embodiment, the method further comprises that the pseudo-NN includes a pseudo-artificial NN (ANN), wherein the pseudo-NN is a simplification of the trained prediction model. A system according to the invention is described. The system comprises a set of vehicle operating inputs received by a processor, relating to one or more measurements of vehicle operations, which are used to predict a torque requested by a vehicle; a vehicle control system instructed by a torque prediction model implemented by the processor using an artificial neural network (ANN) to solve a root of the prediction model used to obtain the torque request; the processor configured to execute the ANN's torque prediction model in response to the torque request in order to receive a set of measurements as vehicle operating inputs for torque prediction;the processor configured to replace a set of multiple independent variables of the torque prediction model implemented by the ANN in order to formulate a functional mathematical expression in a pseudo-ANN containing a reduced set of variables, which includes only one independent variable from a reduced set containing multiple independent variables used by the ANN; the processor configured to receive the set of measurements of a current operating vehicle as vehicle operating inputs into the pseudo-ANN in order to apply the torque prediction model with a single independent variable and one or more constants that have been replaced for a full set of independent variables;The processor is configured to process the single independent variable in a pseudo-ANN containing a simplified mathematical expression, and the processor is configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN to obtain a root value without relying on an inversion operation of a mathematical expression consisting of a full set of independent variables to predict a requested torque value. In one embodiment, the system also includes the processor configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN to obtain a root value to predict a change in an air-per-cylinder (APC) control to obtain the requested torque value. In at least one embodiment, the method further comprises the processor configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN in order to obtain a root value in order to predict a difference in an ignition control (SPARK control) in order to change the engine timings in order to obtain the requested torque value. In at least one embodiment, the system further comprises the processor, which is configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN despite a high degree of nonlinearity of the ANN, without having to resort to an inversion operation. In at least one embodiment, the system further comprises the processor configured to apply the torque prediction model, including the composition of a multi-hidden-layer activation function, using a root resolution algorithm to solve the expression for a single remaining variable. In at least one embodiment, the system also comprises the set of measured vehicle operating inputs, including intake camshaft (ICAM), exhaust camshaft (ECAM), revolutions per minute (RPM), air-per-cylinder (APC), and ignition timing of a vehicle in operation. In at least one embodiment, the system also includes the pseudo-ANN as a trained prediction model. Furthermore, a vehicle device is described, wherein the system according to the invention can be integrated into the vehicle device. The embodiments are described below in conjunction with the following figures, where identical reference numerals denote identical elements, and where: Fig. 1 is a functional block diagram representing an autonomous or semi-autonomous vehicle with a control system that controls vehicle actions based on the use of a neural network to predict torque characteristics in a vehicle control system; Fig. 2 is a diagram illustrating a pseudo-neural network that can be implemented with an APC or SPARK controller that uses the neural network to predict the torque characteristics of a vehicle control system; Fig. 3 is a functional block diagram illustrating a pseudo-artificial neural network (ANN) implemented to predict APC and SPARK changes for a requested torque shown in Figs. 1-2; and Fig.Figure 4 shows an exemplary flowchart illustrating the steps for use in the torque control method shown in Fig. 1-2. As used herein, the term “module” refers to any hardware, software, firmware, electronic control component, processing logic and / or processor device, individually or in any combination, including but not limited to: application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), electronic circuit, processor (common, dedicated or group) and memory executing one or more software or firmware programs, combinational logic circuit and / or other suitable components providing the described functionality. Embodiments can be described here in the form of functional and / or logical block components and various processing steps. It should be noted that such block components can be implemented by any number of hardware, software, and / or firmware components configured to perform the specified functions. For example, an embodiment of the present disclosure may employ various integrated circuit components, such as memory elements, digital signal processing elements, logic elements, lookup tables, or the like, which can perform a variety of functions under the control of one or more microprocessors or other control devices. For the sake of brevity, conventional techniques related to signal processing, data transmission, signaling, control, machine learning, image analysis, and other functional aspects of the systems (and the individual operating components of the systems) may not be described in detail here. Furthermore, the connecting lines shown in the various figures contained herein are intended to represent exemplary functional relationships and / or physical couplings between the various elements. It should be noted that many alternative or additional functional relationships or physical connections may exist in an embodiment. The vehicle system control system incorporates driver and vehicle operating inputs, including vehicle sensor data and torque requests, and the transmission of these inputs to an engine control module (ECM) and a transmission control module (TCM). The ECM can calculate a requested axle torque from the driver and vehicle operating inputs. This requested axle torque can then be transmitted to the engine and the ECM. The ECM can receive vehicle operating inputs such as intake camshaft adjuster (ICAM), exhaust camshaft adjuster (ECAM), vehicle speed or revolutions per minute (RPM), air per cylinder (APC), and engine control data or electronic ignition control (SPARK) data. Changes in APC and SPARK respond to a requested torque demand from the driver. The present description provides methods, systems and devices that enable torque control which avoids complex mathematical inversion by simplifying an artificial neural network (ANN) to a “pseudo-ANN” with a single variable to solve its roots via SPARK or APC control variables (i.e. changes in APC or SPARK). With reference to Fig. 1, a control system 100 is connected to a vehicle 10 (here also referred to as the “host vehicle”) in accordance with various embodiments. In general, the control system (or simply “system”) 100 provides control of various actions of the vehicle 10 (e.g., torque control) based on a trained pseudo-ANN model that controls the operation in response to data from vehicle operating inputs, as described in more detail below in conjunction with Figs. 2-4. In various exemplary embodiments, System 100 is able to reduce the dimensionality and complexity during the control step, despite a high degree of nonlinearity in ANNs, which is desirable in a prediction process. System 100 provides a torque prediction model that simplifies solving a one-dimensional nonlinear function for its roots, instead of attempting to compute a mathematical inverse of a non-invertible, high-degree-of-freedom ANN expression. System 100 offers a novel control method that can be paired with ANNs as prediction models. This process avoids the need for a full mathematical function inversion of an ANN. This is made possible by first creating a simplified "pseudo-ANN" with one variable, which is then solved for its roots (zeros). In various exemplary embodiments, System 100 provides a process that uses an algorithm to control the torque in the embedded control software of a vehicle 10 of System 10 and enables the use of ANNs for a torque prediction model. The process avoids the need for computations of a full mathematical inversion of a highly nonlinear (complex) ANN and instead first reduces the dimensionality of the ANN and then uses a simple square root solution algorithm to control System 100 to a torque requested by the driver. As shown in Fig. 1, the vehicle 10 generally comprises a chassis 12, a body 14, front wheels 16, and rear wheels 18. The body 14 is arranged on the chassis 12 and essentially encloses components of the vehicle 10. The body 14 and the chassis 12 can together form a frame. The wheels 16-18 are each rotatably connected to the chassis 12 near a corner of the body 14. In various embodiments, the wheels 16, 18 comprise a wheel assembly that also includes associated tires. In various embodiments, the vehicle 10 is an autonomous or semi-autonomous vehicle, and the control system 100 and / or components thereof are integrated into the vehicle 10. For example, the vehicle 10 is a vehicle that is automatically controlled to transport passengers from one place to another. In the embodiment shown, the vehicle 10 is depicted as a passenger car, but it should be recognized that any other vehicle, including motorcycles, trucks, sport utility vehicles (SUVs), recreational vehicles (RVs), watercraft, aircraft, and the like, can also be used. As shown, the vehicle 10 generally comprises a drive system 20, a transmission system 22, a steering system 24, a braking system 26, a canister rinsing system 31, one or more user input devices 27, a sensor system 28, an actuator system 30, at least one data storage device 32, at least one control unit 34, and a communication system 36. The drive system 20 may, in various embodiments, comprise an internal combustion engine, an electric machine such as a traction motor, and / or a fuel cell drive system. The transmission system 22 is configured to transmit power from the drive system 20 to the vehicle wheels 16 and 18 according to selectable speed ratios. According to various embodiments, the transmission system 22 may comprise a continuously variable automatic transmission, a continuously variable transmission, or other suitable transmissions. The braking system 26 is configured to exert a braking torque on the vehicle wheels 16 and 18. The braking system 26 can, in various embodiments, include friction brakes, a wire brake, a regenerative braking system such as an electric motor, and / or other suitable braking systems. The steering system 24 influences the position of the vehicle wheels 16 and / or 18. Although a steering wheel is shown for illustrative purposes, the steering system 24 may not include a steering wheel in some embodiments. The controller 34 comprises at least one processor 44 (and a neural network 33) and a computer-readable storage device or medium 46. As mentioned above, in various embodiments, the controller 34 (e.g., its processor 44) provides data relating to a projected future path of the vehicle 10, including projected future steering instructions, in advance to the steering control system 84 for use in controlling the steering for a limited period of time in the event that communication with the steering control system 84 is no longer available. In various embodiments, the controller 34 also provides communication with the steering control system 84 via the communication system 36 described below, for example, via a communication bus and / or a transmitter (not shown in Fig. 1). In various embodiments, the controller 34 comprises at least one processor 44 and a computer-readable storage device or medium 46. The processor 44 can be any custom or commercially available processor, a central processing unit (CPU), a graphics processing unit (GPU), an auxiliary processor among multiple processors associated with the controller 34, a semiconductor-based microprocessor (in the form of a microchip or chipset), any combination thereof, or, more generally, any instruction-executing device. The computer-readable storage device or medium 46 can include volatile and non-volatile memory, such as read-only memory (ROM), random-access memory (RAM), and keep-alive memory (KAM). KAM is a persistent or non-volatile memory that can be used to store multiple neural networks along with various operating variables while the processor 44 is powered off.The computer-readable storage device(s) 46 can be implemented using any number of known storage devices such as PROMs (programmable read-only memory), EPROMs (electrically erasable PROMs), EEPROMs (electrically erasable PROMs), flash memory, or other electrical, magnetic, optical, or combined storage devices capable of storing data, some of which constitute executable instructions used by the controller 34 in controlling the vehicle 10. The instructions can comprise one or more separate programs, each containing an ordered list of executable instructions for implementing logical functions. When executed by the processor 44, the instructions receive and process signals from the sensor system 28, perform logic, calculations, methods, and / or algorithms for the automatic control of the vehicle 10 components, and generate control signals that are transmitted to the actuator system 30 to automatically control the vehicle 10 components based on the logic, calculations, methods, and / or algorithms. Although in Fig.1 where only one control unit 34 is shown, embodiments of the vehicle 10 may include any number of control units 34 which communicate via any suitable communication medium or combination of communication media and which cooperate to process the sensor signals, perform logic, calculations, methods and / or algorithms and generate control signals to automatically control features of the vehicle 10. As shown in Fig. 1, the vehicle 10 generally comprises, in addition to the steering system 24 mentioned above and the control unit 34, a chassis 12, a body 14, front wheels 16, and rear wheels 18. The body 14 is arranged on the chassis 12 and essentially encloses components of the vehicle 10. The body 14 and the chassis 12 can together form a frame. The wheels 16-18 are each rotatably connected to the chassis 12 near a corner of the body 14. In various embodiments, the wheels 16, 18 comprise a wheel assembly that also includes the respective tires. In various embodiments, the vehicle 10 is an autonomous vehicle, and the control system 100 and / or components thereof are integrated into the vehicle 10. For example, the vehicle 10 is a vehicle that is automatically controlled to transport passengers from one place to another. In the embodiment shown, the vehicle 10 is depicted as a passenger car, but it should be recognized that any other vehicle, including motorcycles, trucks, sport utility vehicles (SUVs), recreational vehicles (RVs), watercraft, aircraft, and the like, can also be used. As shown, the vehicle 10 generally also includes a drive system 20, a transmission system 22, a braking system 26, one or more user input devices 27, a sensor system 28, an actuator system 30, at least one data storage device 32, and a communication system 36. The drive system 20 may, in various embodiments, comprise an internal combustion engine, an electric machine such as a traction motor, and / or a fuel cell drive system. The transmission system 22 is configured to transmit power from the drive system 20 to the vehicle wheels 16 and 18 according to selectable speed ratios. According to various embodiments, the transmission system 22 may comprise a continuously variable automatic transmission, a continuously variable transmission, or other suitable transmissions. The braking system 26 is configured to exert a braking torque on the vehicle wheels 16 and 18. The braking system 26 can, in various embodiments, include friction brakes, a wire brake, a regenerative braking system such as an electric motor, and / or other suitable braking systems. The steering system 24 influences the position of the vehicle wheels 16 and / or 18. Although a steering wheel is shown for illustrative purposes, the steering system 24 may not include a steering wheel in some embodiments. The control system 34 comprises a vehicle control system that is directly influenced based on the output of the models of the neural network 33. In an exemplary embodiment, a feed-forward operation can be applied to an adjustment factor, which is the continuous output of the models of the neural network 33, to generate a control action for the desired torque or another similar action (in the case of a continuous model of the neural network 33, for example, the continuous APC / SPARK prediction values ​​are outputs). In various embodiments, one or more user input devices 27 receive vehicle operating inputs from one or more passengers (and the driver 11) of the vehicle 10. In various embodiments, the vehicle operating inputs include a desired destination for the vehicle 10. In certain embodiments, one or more input devices 27 include an interactive touchscreen in the vehicle 10. In certain embodiments, one or more input devices 27 include a loudspeaker for receiving audio information from the passengers. In certain other embodiments, one or more input devices 27 may include one or more other types of devices and / or be paired with a user device (e.g., a smartphone and / or other electronic devices) of the passengers. The sensor system 28 comprises one or more sensors 40a-40n that detect observable conditions of the external environment and / or the internal environment of the vehicle 10. The sensors 40a-40n include, among others, radars, lidar, global positioning systems, optical cameras, thermal cameras, ultrasonic sensors, inertial measurement units and / or other sensors. The actuator system 30 comprises one or more actuators 42a-42n that control one or more vehicle functions, such as, but not limited to, the canister rinsing system 31, the intake system 38, the drive system 20, the transmission system 22, the steering system 24, and the braking system 26. In various embodiments, the vehicle 10 may also have internal and / or external vehicle features not shown in Fig. 1, such as different doors, a trunk, and cabin features such as air conditioning, music, lighting, touchscreen display components (such as those used in conjunction with navigation systems), and the like. The data storage device 32 stores data for use in the automatic control of the vehicle 10, including data from a pseudo-ANN used to predict a driver-requested torque for vehicle control. In various embodiments, the data storage device 32 stores a machine learning model of a pseudo-ANN and other data models, such as defined maps of the navigable environment. The pseudo-ANN itself is not trained. Model training is performed for an ANN prediction model (see Fig. 2, ANN prediction model (210)). No separate training is required for the pseudo-ANN; instead, the ANN prediction model is implemented with a set of values. In various embodiments, the pseudo-ANN can be predefined based on the ANN prediction model, which is configured with a set of values ​​by and received from a remote system.For example, the neural network 33 (i.e., the ANN prediction model) can be trained by a supervised learning methodology from a remote system and transferred to the vehicle 10 (wirelessly and / or wired) or deployed and stored in the data storage device 32. The ANN prediction model can also be trained by supervised or unsupervised learning based on vehicle data. The data storage device 32 is not limited to control data, as other data can also be stored in the data storage device 32. For example, route information can also be stored in the data storage device 32—that is, a set of road segments (geographically linked to one or more of the defined maps) that together define a route the user can take to get from a starting point (e.g., the user's current location) to a destination. As will be evident, the data storage device 32 can be part of the controller 34, separate from the controller 34, or part of the controller 34 and part of a separate system. The controller 34 implements the logic model for the pseudo-ANN based on the ANN prediction model trained with a set of values ​​and comprises at least one processor 44 and a computer-readable memory device or media 46. The processor 44 can be any custom or off-the-shelf processor, central processing unit (CPU), graphics processing unit (GPU), auxiliary processor among multiple processors assigned to the controller 34, semiconductor-based microprocessor (in the form of a microchip or chipset), any combination thereof, or, more generally, any device for executing instructions. The computer-readable memory device or media 46 can include volatile and non-volatile memory, such as read-only memory (ROM), random-access memory (RAM), and keep-alive memory (KAM).KAM is a persistent or non-volatile memory that can be used to store various operating variables while the processor 44 is switched off. The computer-readable memory device(s) 46 can be implemented using any number of known memory devices such as PROMs (programmable read-only memory), EPROMs (electrically erasable PROMs), EEPROMs (electrically erasable PROMs), flash memory, or other electrical, magnetic, optical, or combined memory devices capable of storing data, some of which represent executable instructions used by the controller 34 in controlling the vehicle 10. The instructions can comprise one or more separate programs, each containing an ordered list of executable instructions for implementing logical functions. When executed by the processor 44, the instructions receive and process signals from the sensor system 28, perform logic, calculations, methods, and / or algorithms for the automatic control of the vehicle 10 components, and generate control signals that are transmitted to the actuator system 30 to automatically control the vehicle 10 components based on the logic, calculations, methods, and / or algorithms. Although in Fig.1 where only one control unit 34 is shown, embodiments of the vehicle 10 may include any number of control units 34 which communicate via any suitable communication medium or combination of communication media and which cooperate to process the sensor signals, perform logic, calculations, methods and / or algorithms and generate control signals to automatically control features of the vehicle 10. The communication system 36 is configured to wirelessly transmit information to and from other units 48, such as, but not limited to, other vehicles (“V2V” communication), infrastructure (“V2I” communication), remote transport systems, and / or user devices (described in more detail with reference to Fig. 2). In one exemplary embodiment, the communication system 36 is a wireless communication system configured to communicate via a wireless local area network (WLAN) using IEEE 802.11 standards or using cellular data communication. However, additional or alternative communication methods, such as a dedicated short-range communication channel (DSRC channel), are also considered within the scope of this description.DSRC channels refer to one- or two-way short- to medium-range wireless communication channels specifically designed for automotive applications, as well as a corresponding set of protocols and standards. In various embodiments, the communication system 36 is used for communication between the control unit 34, including data relating to a projected future path of the vehicle 10, including projected future steering instructions. In various embodiments, the communication system 36 can also enable communication between the steering control system 84 and / or other systems and / or devices. In certain embodiments, the communication system 36 is further configured for communication between the sensor system 28, the input device 27, the actuator system 30, one or more controllers (e.g., the controller 34), and / or other systems and / or devices. For example, the communication system 36 can comprise any combination of a Controller Area Network (CAN) bus and / or direct wiring between the sensor system 28, the actuator system 30, one or more controllers 34, and / or one or more other systems and / or devices. In various embodiments, the communication system 36 can include one or more transceivers for communication with one or more devices and / or systems of the vehicle 10, passenger devices (e.g., the user device 54 of Fig. 2), and / or one or more remote information sources (e.g., GPS data, traffic information, weather information, etc.). Referring to Fig. 2, Fig. 2 shows an exemplary diagram of an APC controller that uses the pseudo-ANN root resolver for the torque prediction model according to one embodiment. The torque prediction model is implemented by the pseudo-ANN system, avoiding complex ANN inversion techniques. The APC controller is paired with an ANN, which can be implemented using the neural network to predict torque characteristics based on measured vehicle operating inputs according to various embodiments. The ANN solves problems with torque prediction accuracy and calibration overhead that can occur during vehicle operation. In one exemplary embodiment, a current calibration process can be based on an ever-increasing number (~52) of calibration lookup tables. The process can be made capable of tolerating higher dimensionality than a conventional process and also handling increases in engine power by performing the required torque calibrations more quickly. Figure 2 shows an exemplary torque prediction system 200 that includes a pseudo-torque prediction model 210 (i.e., an implemented pseudo-ANN structure) embedded in a control unit with a set of measured vehicle operating inputs 205. These inputs represent the vehicle operating inputs of the current vehicle operation with respect to RPM, ICAM, ECAM, APC, and SPARK. Also included is an output 207 of the torque prediction model, which generates a predicted torque based on the measured vehicle operating inputs 205 of the torque prediction model. Furthermore, in an exemplary embodiment, the APC controller 215 operates for torque control using fixed variables of RPM, ICAM, and ECAM.In an exemplary embodiment, the APC controller 215 operates when a driver requests a new or desired torque to modify the values ​​of one or both APC and SPARK vehicle operating inputs to achieve an output that closely approximates the requested torque value. For example, the driver may request the desired torque, and this torque request is received as a vehicle operating input for the pseudo-ANN root solver processor 220; additionally, the set of vehicle operating inputs may include the current RPM, ICAM, ECAM, and SPARK received from the vehicle operating system.The pseudo-ANN root solver 220 implements the ANN torque prediction model 210 based on the received set of measured vehicle operating inputs, representing the measured operating conditions of the vehicle's current RPM, ICAM, and ECAM, for reception by a pseudo-ANN model to perform root resolution instead of traditional inversion methods. The pseudo-ANN is based on a prediction model with values ​​that have already been trained offline; therefore, the ANN can be considered a trained supervised or unsupervised model. The pseudo-ANN model outputs a desired APC 240 to achieve the requested torque 225. The pseudo-ANN root solver 220 can also be reapplied with a current SPARK input 230. In one exemplary embodiment, the trained ANN torque prediction model is a statistically based machine learning method called a Deep Artificial Neural Network (ANN) for predicting braking torque in the embedded software of the control unit. This method can replace complex, traditional, lookup table-based methods to increase accuracy and improve calibration effort. The ANN increases the accuracy of torque prediction and reduces the calibration effort in current software and calibration processes, which are caused by a constantly increasing number (~52) of calibration lookup tables. In an exemplary embodiment, the trained ANN torque prediction model is implemented with an ANN prediction model or algorithm configured with various elements, including a data split of approximately 80% training and 20% testing; inputs / outputs: 5 inputs, 1 output; vehicle operating inputs: RPM, ICAM, ECAM, APC, Spark; output: braking torque; number of neurons per hidden layer: 12-18; hidden layers: 2+ ("Deep"); and an activation function of a tangential sigmoid. In one exemplary embodiment, the ANN torque prediction model is configured through the steps enabled for deep ANN and for calibration using dyno data collected using a design of experiments (DoE) covering a full range of the input space in a pseudo-space-filling configuration. The data collection process can be performed, for example, by a calibrator running a MATLAB® algorithm to train the deep ANN and importing the DoE dyno data on demand. The user only inputs data, algorithm details, options, and structure, which are automatically configured by the algorithm. The training algorithm is executed and outputs a final ANN model with optimally chosen weights and distortion parameters, resulting in a model with improved torque prediction.The calibrator considers a set of results that are automatically plotted for output by the algorithm to ensure that an appropriate set of requirements is met. The calibrator incorporates the weights and distortions selected by the algorithm into the software provided by the ECU for calibration. The configured ANN torque prediction model can accept a set of vehicle measurements as vehicle operating inputs, which may have a greater degree of freedom (i.e., compared to tabular methods). These greater degrees of freedom include RPM, ICAM, ECAM, APC, SPARK, and others.Furthermore, the degrees of freedom are not limited to the use of a conventional set of vehicle measurements including RPM, ICAM, and ECAM, nor are the freedoms limited to the extended set of vehicle measurements including RPM, ICAM, ECAM, APC, and SPARK, but can be extended to include other variables as desired to ensure modeling with reasonable numerical (i.e., higher degrees of accuracy) accuracies and other variable considerations. In various exemplary embodiments, the prediction logic implemented through offline training is either logic derived from supervised or unsupervised learning processes and can be activated using other neural networks, including trained convolutional neural networks (CNNs) and / or recurrent neural networks (RNNs) in which the root-solution methodology can be applied and used in vehicle operation. Furthermore, alternative embodiments can be considered that include a neural network consisting of multiple layers (i.e., three layers) of a convolutional neural network (CNN), also with dense layers (i.e., two dense layers), trained offline, enabling the control of operations in accordance with the system shown in Fig. 1, in accordance with various embodiments. The neural network is used to inform the APC / SPARK controller about torque characteristics and is configured as a pre-trained neural network. Therefore, in certain embodiments, the torque prediction system process is configured in only one operating mode. In various embodiments, for example, the neural network is trained in a training mode before being deployed or made available in the vehicle (or other vehicles). Once the neural network is trained, it can be implemented in a vehicle (e.g., vehicle 10 in Fig. 1) in an operating mode where the vehicle is operated autonomously, semi-autonomously, or manually. In various alternative exemplary embodiments, the neural network can also be implemented in a vehicle in both training and operating modes and trained during an initial operating period in conjunction with time-delay operations or a similar methodology for torque control predictions. Alternatively, a vehicle can operate exclusively in operating mode with neural networks that have already been trained via a training mode of the same vehicle and / or other vehicles in different embodiments. Figure 3 shows an exemplary flowchart for the APC control of the torque prediction system according to one embodiment. In Figure 3, the APC control using the ANN is initiated at Problem 305. It is expected that if an ANN were used for the prediction model, there would be no practical way to control this type of ANN model using mathematical inversion. This is because true inversion of an ANN is a highly complex method and computation, and depending on the ANN structure, it may not be possible without recently published numerical estimation methods. The process described in flowchart 300 of Figure 3 completely avoids the need for mathematical inversion of an ANN, thus enabling the use of ANNs for the prediction model and simultaneously providing a method of controlling the model that might otherwise not have been possible.The process reduces the dimensionality of the mathematical expression, the complexity of the ANN expression, the need for a true inversion, and the computational complexity, while still meeting the requirements for an adequate level of torque control. In Fig. 3, the ANN in Problem 310 receives a set of "n" vehicle operating inputs, comprising a plugin of the actual RPM, a plugin of the actual ICAM, a plugin of the actual ECAM, APC, and the plugin of the actual SPARK. In Problem 310, fixed current values ​​are substituted into the ANN for 5 of the 5 independent variables. In Problem 320, the expression is simplified to the function of a single variable, APC. This will be a composition of activation functions ∅ due to several hidden layers in the ANN structure. The expression is then set equal to the torque desired by the driver, TDes. TDes = A + B * Ø (Ø(...APC))+...+C*Ø (Ø(...APC)) where T, A and B are constants and ∅ is the chosen NN activation function. In problem 330, a nonlinear root resolution algorithm is used to solve this expression for the only remaining variable APC (i.e., to find the roots of this expression). That is, 0 = -TDes + A + B * Ø (Ø(...APC)) + ... + C * Ø(Ø(...APC)) where APC is the desired APC = Root 1, Root 2, etc.... and T, A and B are constants and Ø is the chosen NN activation function. In another exemplary embodiment, the same or a similar three-stage process described above can be applied in a different cycle for the spark by leaving the spark as the remaining variable and resolving the spark for a desired torque value requested by the driver. Figure 4 shows an exemplary flowchart illustrating the steps for use in the control procedure for torque control according to one embodiment. In Figure 4, the procedure includes the following tasks. Task 410: An ANN is embedded in the control as a torque prediction model. Then, in Tasks 420 and 425, the three-stage control process is applied in successive multiples (i.e., twice) to determine the roots in the first cycle for the desired APC value in Task 420 and in the second cycle in Task 425 to determine the desired SPARK value. The logic, discovered through historical or expert knowledge or by empirical testing of various logic schemes from a list of roots in Task 430, that enables APC or SPARK control for the desired torque requested by the driver, is implemented.The selected root is implemented in Task 435 as a control variable in a software control module to obtain the torque requested by the driver. The converted desired APC root (or Spark root) in Task 440 is implemented via a set of constants (i.e., the constants T, A, and B) and the selected NN activation function ∅ used in the control module. In various exemplary embodiments, the features sent to the predictive model as vehicle operating inputs can include other vehicle operating measurements, such as engine speed, air mass per cylinder event, vent valve duty cycle, wastegate position (optional), oxygen sensor output, refueling command, camshaft adjuster position, boost pressure, charge air temperature, ignition timing, boost pressure ratio (optional), and throttle position. The presented set of input features should not be considered exhaustive. For example, the input set of "n" vehicle operating inputs can be augmented, modified, or reduced depending on the vehicle operating inputs required for the torque control to function. An output from the neural network is a continuous output to instruct a vehicle control unit to perform an action that affects the torque value.In various exemplary embodiments, the output from the neural network is configured as a continuous output to instruct a vehicle control system to perform a control action to regulate the value of a predicted torque. In various embodiments, the ANN is stored in onboard memory in the vehicle, such as the computer-readable storage device or medium 46 of Fig. 1. In various embodiments, the described methods, systems and vehicles provide a canonical representation of the output of a hybrid recurrent neural network, together with the use of a deep neural network for regression over this canonical representation and for predicting vehicle actions (e.g. torque actions) for a vehicle using multiple neural networks, as described above. As briefly mentioned above, the various modules and systems described above can be implemented as one or more machine learning models that undergo supervised, unsupervised, semi-supervised, or reinforcement learning. Such models can be trained to perform classification (e.g., binary or multiclass classification), regression, clustering, dimensionality reduction, and / or similar tasks. Examples of such models include artificial neural networks (ANNs) (e.g., recurrent neural networks (RNNs) and convolutional neural networks (CNNs)), decision tree models (e.g., classification and regression trees (CARTs)), ensemble learning models (e.g., boosting, bootstrapped aggregation, gradient boosting machines, and random forests), Bayesian network models (e.g., naive Bayes), principal component analysis (PCA), support vector machines (SVMs), and clustering models (e.g.,K-nearest-neighbor, K-means, expectation-maximization, hierarchical clustering, etc.) and models of linear discriminant analysis.

Claims

Method for implementing a torque control using a neural network (NN), comprising: in response to a torque request (225), executing, by a processor (220), a torque prediction model (210) of the NN to receive a set of measured vehicle operating inputs (205) associated with the torque prediction; replacing a set of multiple independent variables of the torque prediction model (210) implemented by the NN with a simplified mathematical expression in a pseudo-NN containing a reduced set of variables including at least one independent variable to replace a previous set of multiple independent variables of the NN;Processing the set of measured vehicle operating inputs by the pseudo-NN based on the torque prediction model (210) using at least one independent variable and one or more constants that replace a set of substituted independent variables in a simplified mathematical expression of the pseudo-NN; and solving at least one root of the simplified mathematical expression of the pseudo-NN by obtaining a root value without having to rely on an inversion operation of a mathematical expression consisting of a whole set of independent variables to obtain a requested torque value. Method according to claim 1, further comprising: solving at least one root of the simplified mathematical expression of the pseudo-NN by the processor (220) to obtain a root value in order to predict a change in an air-per-cylinder (APC) control (215) to obtain the requested torque value. Method according to claim 2, further comprising: solving at least one root of the simplified mathematical expression of the pseudo-NN by the processor (220) to obtain a root value in order to predict a difference in an ignition control in order to change the engine timings in order to obtain the requested torque value. The method according to claim 3, further comprising: solving at least one root of the simplified mathematical expression of the pseudo-NN without having to resort to an inversion operation, despite a high degree of nonlinearity of the NN. The method of claim 4, further comprising: applying the torque prediction model (210) comprising a composition of an activation function with multiple hidden layers by the processor (220), wherein a root resolution algorithm is used to solve the simplified mathematical expression for a single remaining variable. Method according to claim 1, wherein the set of measured vehicle operating inputs (205) comprises intake camshaft adjuster (ICAM), exhaust camshaft adjuster (ECAM), revolutions per minute (RPM), air per cylinder (APC) and ignition timing of a vehicle (10) in operation. Method according to claim 6, wherein the pseudo-NN comprises a pseudo-artificial NN (ANN), wherein the pseudo-ANN is a trained prediction model. System (100), comprising: a set of vehicle operating inputs (205) received by a processor (220) relating to one or more measurements of vehicle operations used to predict a requested torque of a vehicle (10); a vehicle control system instructed by a torque prediction model (210) implemented by the processor (220) using an artificial neural network (ANN) to solve a root of the prediction model used to obtain the torque request (225); in response to the torque request (225), the processor (220) is configured to execute the torque prediction model (210) of the ANN to obtain a set of measurements as vehicle operating inputs (205) for torque prediction;wherein the processor (220) is configured to replace a set of several independent variables of the torque prediction model (210) implemented by the ANN in order to formulate a functional mathematical expression in a pseudo-ANN containing a reduced set of variables comprising only one independent variable of a reduced set comprising several independent variables used by the ANN; the processor (220) is configured to receive the set of measurements of a currently operating vehicle (10) as vehicle operating inputs (205) for the pseudo-ANN in order to apply the torque prediction model (210) with a single independent variable and one or more constants that have been replaced for a full set of independent variables;the processor (220) is configured to process the single independent variable in a pseudo-ANN containing a simplified mathematical expression; and the processor (220) is configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN to obtain a root value without relying on an inversion operation of a mathematical expression consisting of an entire set of independent variables to predict a requested torque value. System (100) according to claim 8, further comprising: the processor (220) is configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN to obtain a root value to predict a change in an air-per-cylinder (APC) control (215) to obtain the requested torque value. System (100) according to claim 9, further comprising: the processor (220) being configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN to obtain a root value in order to predict a difference in an ignition control to change the engine timing to obtain the requested torque value; the processor (220) being configured to solve at least one root of the simplified mathematical expression of the pseudo-ANN without relying on an inversion operation, despite a high degree of nonlinearity of the ANN; and the processor (220) being configured to apply the torque prediction model (210) comprising a composition of an activation function with multiple hidden layers, using a root resolution algorithm to solve an expression for a single remaining variable;wherein the set of measured vehicle operating inputs (205) includes intake camshaft adjuster (ICAM), exhaust camshaft adjuster (ECAM), revolutions per minute (RPM), air per cylinder (APC) and ignition timing of a vehicle in operation (10).

Images