Current determination at a consumer in a motor vehicle

Abstract

A method for determining an individual electrical current through a consumer (120) on board a motor vehicle (105) comprises steps of determining a total current flowing through the consumer (120) and at least one other consumer (120); providing a machine learning technique (160) trained to determine individual currents through the consumers (120) based on the total current; and determining the individual current through the consumer (120) using the machine learning technique (160) based on the total current.

Description

[0001] The present invention relates to a consumer on board a motor vehicle. In particular, the invention relates to the determination of an electric current by such a consumer.

[0002] A motor vehicle comprises numerous electrical components powered by one or more energy sources. To control the vehicle, for example, with regard to energy management, attempts are made to determine and monitor as many currents as possible flowing through each component over time. These measurements can be collected on board the vehicle or transmitted to a central location where they can be correlated with measurements from other vehicles.

[0003] This technique requires significant metrological effort and the transmission of a large number of measured values. One of the problems underlying the present invention is to provide an improved technique for determining electrical currents from consumers on board a motor vehicle. The invention solves this problem by means of the subject matter of the independent claims. Dependent claims describe preferred embodiments.

[0004] According to a first aspect of the present invention, a first method for determining an individual electrical current through a consumer on board a motor vehicle comprises steps of determining a total current flowing through the consumer and at least one other consumer; providing a machine learning technique trained to determine individual currents through the consumers based on the total current; and determining the individual current through the consumer using the machine learning technique based on the total current.

[0005] It has been recognized that the electrical consumers on board a motor vehicle often form groups that are functionally coupled or interdependent. Electrical currents through such consumers can be correlated, but this correlation is often not obvious or difficult to detect using conventional methods. It is therefore proposed to determine the correlation using a machine learning technique, allowing an individual current to be determined based on the total current.

[0006] In practice, this allows for the creation of a virtual sensor capable of determining the current flowing through a device without physically measuring it. Based on the device's electrical current consumption, it's then possible to determine its condition and whether it's functioning normally.

[0007] In a further development of the invention, in the event of an anomaly in the current flow through the consumer, information can be collected on board the vehicle and transmitted to an external body for analysis. The volume of transmitted data can be significantly reduced compared to the prior art, typically by approximately 90% or more. Processing and transmission costs can be saved. Storage and processing costs can be reduced for a central body that collects and potentially analyzes information from a large number of vehicles.

[0008] According to a further aspect of the invention, a second method for training a machine learning technique to determine an electric current through a consumer on board a motor vehicle is proposed. The second method comprises steps of detecting individual currents through multiple consumers on board the motor vehicle; detecting a total current through the consumers; and training the machine learning technique to determine the individual currents based on the total current.

[0009] The second method can be performed on board a motor vehicle that includes real sensors for determining the individual currents. This motor vehicle could, for example, be a development or pre-production model. The trained technology can be transferred to another motor vehicle to perform the first method described herein. The other motor vehicle does not need to have sensors for determining the individual currents through the consumers.

[0010] It is preferred that the individual currents and the total current at corresponding devices on board various motor vehicles are recorded. The motor vehicles can be of the same type or construction, or have identical or sufficiently similar systems in which the electrical consumers are housed. It has been shown that the determination of the individual currents is relatively robust, so that good results can already be achieved when the machine learning technique is trained on data from a small number of motor vehicles. In one example, only two motor vehicles were required to collect sufficient data that can be applied on board virtually any number of corresponding motor vehicles or devices.

[0011] Machine learning techniques primarily work with temporal sequences of measured values. Individual streams and the total stream can be sampled at a predetermined sampling rate, for example, approximately once per second, ten times per second, or one hundred times per second. This allows for the generation of corresponding sequences of measured values, enabling the training of the machine learning system and the determination of an individual stream based on the total stream.

[0012] The technique of machine learning preferably employs a recurrent neural network. A recurrent (feedback) neural network has neurons in multiple layers, and unlike a feedforward network, it allows connections between neurons in one layer and neurons in the same or a preceding layer. This enables improved evaluation of the temporal sequence of measurements. Such a network can, for example, be an Elman network, a Jordan network, a Hopfield network, or a fully interconnected neural network.

[0013] It is particularly advantageous that the machine learning technique includes a long short-term memory (LSTM). This overcomes the problem of vanishing gradients, which can be problematic in some recurrent neural networks. The LSTM can encompass a short-term memory ranging from a few to several thousand time steps. The LSTM allows gradients to be propagated with little or no attenuation during training, enabling the training of neurons in a multilayer network, even those relatively far from the starting layer, via backpropagation.

[0014] Furthermore, it is proposed that the machine learning technique incorporates a transformer. A transformer comprises a deep learning architecture that integrates an attention mechanism. Classic use cases include translation, text generation, or summarizing longer texts. The transformer can be more efficient than long short-term memory and implement soft weighting for processed words to operate within a current contextual window.

[0015] According to yet another aspect of the present invention, a device for determining an individual electrical current through a consumer on board a motor vehicle comprises a sensor for determining a total current flowing through the consumer and at least one other consumer; and a machine learning technique trained to determine the individual currents on the basis of the total current.

[0016] The device may include a processing unit configured to implement a machine learning technique described herein.

[0017] The processing device is preferably configured to partially or completely execute a method described herein. For this purpose, the processing device may be electronic and may, for example, comprise an integrated circuit, a programmable logic device, or a programmable microcomputer. The method may be implemented as a configuration or as a computer program product with program code means for the processing device. The configuration or the computer program product may be stored on a computer-readable data carrier. Features or advantages of the method may be transferred to the device or vice versa.

[0018] The consumers can form a group connected to a power distribution unit. The total current can be exchanged between the power distribution unit and an electrical power source. Preferably, the power distribution unit is connected to only a single power source on board the vehicle. The power source can, for example, comprise a generator, a battery, or a DC / DC converter powered by electrical energy from a high-voltage battery, which is essentially configured to store electrical energy for propelling the vehicle by means of an electric drive motor. A plurality of such groups can exist on board the vehicle, each of whose individually connected consumers can be monitored using the technology described herein.

[0019] Normally, electric current flows from the source to the power distribution unit and is counted as positive. However, in certain operating states of the vehicle or device, a negative current can also occur, flowing from the power distribution unit back to the power source. Similarly, the current through a load can also be temporarily negative, causing the load to feed electrical energy back to the power distribution unit. This electrical energy is usually absorbed by another load, so it may not be detectable in the total current.

[0020] Consumers can be comprised of a component or system on board the motor vehicle. The group can form a functionally logical unit whose elements are related to a common purpose.

[0021] For example, the consumers might include a headlight, an air conditioning system, a chassis control unit, or a central control unit of the vehicle. It has been shown that consumers serving a common purpose on board the vehicle can have such coordinated electrical current draws that their determination using the machine learning technique described herein is possible with good results.

[0022] According to yet another aspect of the present invention, a motor vehicle comprises a device as described herein. The motor vehicle can, in particular, be a passenger car or a motorcycle. Optionally, the motor vehicle can also be another land vehicle, for example, a truck or a bus.

[0023] The invention will now be described in more detail with reference to the attached drawings, in which: Fig. 1 a system; Fig. 2 a flowchart of a first procedure; Fig. 3 a flowchart of a second procedure; and Fig. 4 A diagram of a single current and a predicted single current is illustrated.

[0024] Fig. Figure 1 shows a system 100 with a motor vehicle 105 and an external location 110. On board the motor vehicle 105 is a device 115 comprising several electrical consumers 120 and a power distributor 125. The electrical consumers 120 are preferably functionally related, with the device 115 dedicated to a predetermined purpose on board the motor vehicle 105. For example, the device 115 can include a headlight that incorporates different electrical consumers 120 for implementing parking lights, daytime running lights, low beams, high beams, fog lights, cornering lights, and / or turn signals. During normal operation of the motor vehicle 105, the consumers 120 are operated simultaneously only in predetermined combinations.

[0025] The power distributor 125 typically distributes electrical current from the power source 130 to the electrical loads 120. The power distributor 125 can be passive, for example, implemented as a busbar, or active, for example, in the form of an electrical control unit. Regarding the technology described herein, the power distributor 125 can, in a first variant, include sensors for determining the individual currents flowing through the loads 120. In a second variant, these sensors can be omitted.

[0026] A device 140 on board the motor vehicle 105 comprises a processing unit 150, a sensor 145 for determining a total current exchanged between the power source 130 and the consumers 120, and a processing unit 165. Optionally, a wireless interface 155 is also provided.

[0027] It is proposed that the processing device 150 is configured to implement a machine learning technique 160. In a particularly preferred embodiment, the technique 160 comprises a combination of a transformer and a long short-term memory (LSTM). The technique 160 is trained to determine, based on the time course of a total current flowing between the power source 130 and the device 115, corresponding time courses of individual currents through the consumers 120 of the device 115. It generally holds that at any given time, the sum of the individual currents equals the total current.

[0028] In one embodiment, an anomaly can be detected in such a defined individual current. Information about the device 115, in particular the defined total current, the defined individual current, possibly another individual current or a driving condition of the motor vehicle 105, can be transmitted to the external location 110 via the wireless interface 155.

[0029] The external device 110 can be configured to collect and evaluate such information regarding a large number of motor vehicles 105. For example, the average energy consumption of a consumer 120 can be determined by averaging across one or more motor vehicles 105. In another embodiment, an evaluation of a specific individual current on board the motor vehicle 105 can be performed. For example, the maximum current consumption of a single consumer 120 can be determined. The consumer 120 can then be protected by an electrical fuse to ensure that no current higher than the determined maximum current flows through the consumer 120. The fuse can be designed, in particular, as a cartridge fuse or as an electronic fuse.

[0030] The machine learning technique 160 can, for example, be trained by the external entity 110. For this purpose, the external entity 110 can communicate with one or more motor vehicles 105, each comprising a power distribution unit 125 according to a first variant described herein. Time-dependent profiles of measured values ​​can be received from these motor vehicles 105, each reflecting a total current and several associated individual currents from consumers 120 of a device 115 on board the motor vehicle 105. The external entity 110 preferably comprises an interface 170 for communication with the interface 155 of the motor vehicle 105, a processing unit 165 configured to implement and train the machine learning technique 160, and an optional data storage device 170 for storing information received from a motor vehicle 105.

[0031] The external entity 110 can train the machine learning technique 160 to find a combination of individual currents flowing through the assigned consumers 120 of the facility 115, resulting in a total current. Once the machine learning technique 160 is sufficiently trained, it can be transmitted to a motor vehicle 105. This motor vehicle 105 can include a power distribution unit 125 according to a second variant described herein. Overall, the technique 160 can be trained using data from a few motor vehicles 105 and then deployed on board a large number of motor vehicles 105.

[0032] Fig. Figure 2 shows a flowchart of a first method 200 for determining an individual current through a consumer 120 on board a motor vehicle 105. For this purpose, the motor vehicle 105 includes a current distributor 125, which only requires one sensor 145 to detect a total current through several consumers 120 of a group 115. Sensors for determining individual currents are not required or not provided.

[0033] In step 205, the total current is determined using sensor 145. In step 210, individual currents through consumers 120, corresponding to the measured total current, can be determined using the trained machine learning technique 160. The total current and the individual currents are available as time series or temporal sequences. Using the approach described herein, only a few measurements of the total current are required to determine the individual currents with sufficient certainty, probability, or accuracy using machine learning technique 160.

[0034] Fig. Figure 3 shows a flowchart of a second procedure 300 for training a machine learning technique 160, which can then be applied in the first procedure 200. The technique 160 can, in particular, comprise an artificial neural network (ANN). The procedure 300 is executed with a motor vehicle 105, which is equipped with a power distributor 125 of the first variant described herein, i.e., it includes current sensors for all connected consumers 120. In step 305, a total current is recorded that flows between a device 115 with several electrical consumers 120 and a power source 130. The total current can be determined, in particular, by means of the sensor 145. In addition, in step 310, individual currents flowing through the consumers 120 connected to the device 115 are recorded. The sum of the individual currents essentially corresponds to the total current.

[0035] In step 315, the technology 160 can be trained to reconstruct the individual currents based on the total current. The total currents and the individual currents are each provided as time sequences of measured values. The technology 160 operates on these data streams and is trained to determine corresponding profiles of the individual currents through the included consumers 120, based on a measured profile of the total current.

[0036] In an optional step 320, the technology 160 is fully trained and forms an artificial intelligence (AI) that can be transferred to a motor vehicle 105, which includes a device 115. Typically, no sensors for determining individual currents through the consumers 120 of the device 115 are required at the power distribution unit 125.

[0037] Fig.Figure 4 shows a diagram 400 with a first curve 405, which corresponds to an actual individual current through a consumer 120 of a device 115 on board a motor vehicle 105, and a second curve 410, which corresponds to a determination or prediction based on the total current using a technique described herein. In the horizontal direction, a time in seconds is shown, and in the vertical direction, a relative current flow with respect to a predetermined maximum is shown.

[0038] It can be seen that the specified curve 410 corresponds with good accuracy to the actual curve 405. In an experiment based on representation 400, an accuracy of approximately 89% was achieved. The settling-in period until sufficiently accurate individual currents can be determined based on a total current may only require a few tens to a few hundred data points. In practice, this settling-in process can be successfully completed in less than approximately one second.

[0039] In an exemplary setup 115 with 9 electrical consumers 120, 10 measured values ​​are generated per measuring point according to the state of the art. For a comprehensive analysis of the electrical processes on board the motor vehicle 105, all 10 measured values ​​are currently transmitted to the external location 110 at a selected measuring frequency.

[0040] Using a technique described herein, it may be sufficient to measure only the total current and transmit a corresponding measurement value at the selected measurement frequency to the external location 110. There, based on the trained machine learning technique 160, it can be determined which individual currents have flowed through the assigned consumers 120. In this way, the data volume to be transmitted can be reduced by approximately 90%.

[0041] The technique described herein can be used with devices 115 with theoretically any number or few electrical loads 120. Generally, the savings in data volume are proportional to the number of loads 120 included in the device 115. However, if too many electrical loads 120 are considered, the accuracy or reliability of the described technique may decrease. Good results have so far been achieved with devices 115 that include at least two and at most approximately 15, preferably approximately 4 to 12, electrical loads 120. Reference sign 100 System 105 motor vehicles 110 external position 115 facility 120 electrical consumers 125 power distribution units 130 Power source 140 Device 145 Sensor 150 processing units 155 wireless interface 160 Machine Learning Techniques 165 Processing unit 170 data storage devices 175 interface 200 first procedure 205 Total flow Determine 210 individual streams 300 second procedure 305 Record total flow 310 individual streams are recorded 315 Kl train 320 AI transferred 400 diagram 405 first run 410 second run

Claims

Method (200) for determining an individual electrical current through a consumer (120) on board a motor vehicle (105), the method comprising the following steps: - Determining (305) a total current flowing through the consumer (120) and at least one other consumer (120); - Providing a machine learning technique (160) trained to determine the individual currents based on the total current; and - Determining the individual current through the consumer (120) using the machine learning technique (160) based on the total current. Method (200) for training a machine learning technique (160) to determine an electric current through a consumer (120) on board a motor vehicle (105), wherein the method comprises the following steps: - capturing individual currents through several consumers (120) on board the motor vehicle (105); - capturing a total current through the consumers (120); and - training the machine learning technique (160) to determine the individual currents based on the total current. Method (200) according to claim 2, wherein the individual currents and the total current are detected at corresponding devices on board different motor vehicles (105). Method (200, 300) according to one of claims 1 to 3, wherein the machine learning technique (160) operates with respect to temporal sequences of measured values. Method (200, 300) according to claim 4, wherein the machine learning technique (160) comprises a recurrent neural network. Method (200, 300) according to claim 5, wherein the machine learning technique (160) comprises a Long Short Term Memory. Method (200, 300) according to one of the preceding claims, wherein the machine learning technique (160) comprises a transformer. Device (140) for determining an individual electrical current through a consumer (120) on board a motor vehicle (105), the device comprising the following elements: - a sensor for determining a total current flowing through the consumer (120) and at least one other consumer (120); and - a machine learning technique (160) trained to determine the individual currents on the basis of the total current. Device (140) according to claim 8, wherein the consumers (120) form a group which is connected to a power distributor (125); and the total current is exchanged between the power distributor (125) and an electrical power source. Device (140) according to claim 9, wherein the consumers (120) are comprised of an assembly or system on board the motor vehicle (105). Device (140) according to one of claims 8 to 10, wherein the consumers (120) are comprised of a headlight, an air conditioning system, a chassis control system or a central control unit of the motor vehicle (105). Motor vehicle (105) comprising a device (140) according to one of claims 8 to 11.

Images