WO2019001824A1 - METHOD AND SYSTEM FOR DATA COLLECTION - Google Patents

Abstract

In a method for collecting data by means of a sensor network (2) having a central point (3) and a plurality of sensor nodes (4, 5), the central point (3) generates data acquisition orders (7) for the data collection and transmits at least one of the data acquisition orders (7 ) to at least one of the sensor nodes (4, 5). The sensor nodes (4, 5) respectively capture order-specific data (8) and transmit them to the central location (3). It is provided that the central point (3) the data acquisition orders (7) for optimizing at least one utility value (x ₁ , ..., x _n ) of the order-specific data (8) using an algorithm (f [x ₁ , .. ..... x _n ]) for machine learning.

Description

 METHOD AND SYSTEM FOR DATA COLLECTION

The invention relates to a method for collecting data by means of a sensor network having a central point and a plurality of sensor nodes according to the type defined in more detail in the preamble of claim 1.

The invention also relates to a data collection system comprising a sensor network having a central location and a plurality of sensor nodes, according to the preamble of claim 10.

The use of a sensor network for acquiring environmental data of the associated sensor nodes is well known in the art. In particular, it is known from practice to use vehicles of a vehicle fleet for data collection of information relevant to road traffic and for monitoring the vehicle fleet itself. Reference is made, for example, to DE 10 2013 223 217 A1, which relates to a method for operating a server, whereby individual messages from several vehicles are provided to the server, wherein the individual messages comprise environment data, each describing an identical vehicle environment of the plurality of vehicles, wherein the Server divides individual messages into one or more classes and forms aggregated environment data based on the environment data of the individual messages. The aggregated environment data describe a vehicle environment, whereby the server determines a respective quality of the environment data of the individual messages per class by means of a comparison operation against the aggregated environment data.

Usually, for such methods, the data to be collected, for example concerning traffic signs, are already known in the planning phase of a vehicle project. Thus, the vehicles of the vehicle fleet are already preconfigured for the collection of data on specific data. However, the configuration of all vehicles of a vehicle fleet is in not very efficient in practice. For example, it may be necessary to capture only traffic signs relating to a particular region. The data of the vehicles, which in turn evaluate traffic signs from their environment that are not located in the desired region, are then unnecessarily transmitted to the central location. Among other things energy and data volume are wasted.

It has also been shown that a predefined and defined data extraction is not future-oriented and sufficiently flexible. Thus, a configurable, function-specific data removal from the vehicles of the vehicle fleet is desirable. Such a method is known from the generic DE 195 13 640 A1. DE 195 13 640 A1 relates to a method for reducing a data volume of vehicle data to be transmitted from vehicles of a vehicle fleet, in particular a sampling vehicle fleet, which contain information about the operating state and / or the surroundings of the vehicles, and position data which contain information about the position of the vehicle Contain vehicles in a predetermined coordinate system each at a given time, after detecting the resulting vehicle and position data in the vehicle for the wireless transmission of vehicle and position data to a central point. The vehicles of the vehicle fleet are specified which of the resulting vehicle and position data are to be transmitted to the central location under which conditions. The defaults can be influenced wirelessly by a predefined device outside the vehicles.

A disadvantage of the known method is that the data acquisition jobs are usually created statically and with high manual effort and thereby bring a high cost and resource consumption. Frequently, the measured data have no economic value or have this only for a short time. For a further technical background, reference is also made to DE 101 33 945 A1, which relates to a method and a device for exchanging and jointly processing object data between sensors and a processing unit. In this case, position information and / or speed information and / or further object attributes of sensor objects and fusion objects are transmitted and processed.

Finally, DE 10 2015 210 881 A1 relates to a method for determining the position and / or orientation of a vehicle, wherein at least two sensors are provided whose data are fused. It is provided that a quality measure is dynamically generated for the data of each sensor by means of artificial intelligence, and that the sensor data and the quality measures are fused to determine position and / or orientation.

It is an object of the present invention to provide an improved method for data collection, in which the data acquisition jobs are determined in particular goal-oriented and the amount of data to be transmitted is preferably low. It is also an object of the present invention to provide an improved system for data collection in order to carry out an advantageous data collection by means of a sensor network. According to the invention, the object of the method is achieved by the features mentioned in claim 1.

For the system, the object is achieved by the features mentioned in claim 10.

In the method according to the invention for collecting data by means of a sensor network having a central point and a plurality of sensor nodes, it is provided that the central point generates data acquisition jobs for the data collection and at least one of the data acquisition order transmitted to at least one of the sensor nodes. In each case, the sensor node acquires order-specific data and transmits it to the central location.

The central location may be any location independent of the sensor nodes; in particular, but not exclusively, one or more external servers or server devices, service vehicles or other other vehicles, through to diagnostic devices and building automation systems. However, the invention is particularly suitable for a method for collecting data by means of an external server device and a plurality of sensor nodes.

The central location may also be referred to as a "backend" and the sensor nodes as a "frontend" regarding data collection.

The communication between the plurality of sensor nodes and the central location can be provided wirelessly, in particular based on a wireless standard, or wired.

According to the invention, the central point prepares the data acquisition jobs to optimize at least one useful value of the job-specific data using a machine learning algorithm. In particular, a data acquisition task according to the invention can contain a description and / or coding for the data to be derived from or to be detected by the sensor node.

The basic idea of the invention is thus to create the orders for data acquisition as goal-oriented as possible or efficiently and application-specific using machine learning and to transmit them to the sensor nodes. The invention thus describes a method of using a machine learning algorithm compared to the known one State of the art, in particular taking into account the economic benefits in the collection, processing and distribution of data collected by a sensor network, may be advantageous.

Automated control of the central location may be provided to provide optimal data selection by creating the data acquisition jobs that are sent from the sensor nodes to the central office for economical use.

There is the advantage that machine learning reliably enables the determination of the optimum, in particular the yield of the data collection, by controlling the data acquisition orders. Preferably, only the data are recorded and transmitted by the sensor nodes, for which a temporally and spatially dependent economic benefits can be expected.

Thus, a particularly advantageous cost-benefit ratio is to be expected in the data collection by a sensor network, which can not be achieved with the known methods of the prior art.

In one development of the invention, provision can be made for at least one of the sensor nodes to be designed as a vehicle, mobile terminal, building automation module, traffic monitoring module and / or hardware module of an intelligent personal assistant.

The invention can be implemented particularly advantageously if the sensor nodes are designed as vehicles or if the sensor node is designed as a vehicle fleet, vehicles of all types being suitable. Such a sensor node can thus be, for example, a land vehicle, watercraft, aircraft and / or spacecraft. However, the invention is particularly suitable for use with motor vehicles. There may also be provided several types of sensor nodes within the sensor network, for. B. some sensor nodes that are designed as vehicles and some sensor nodes, which are designed as mobile devices, such as mobile phones or smartphones. In one development of the invention, provision can be made for the data acquisition orders to relate to movement information, status information and / or environment information relating to the sensor nodes.

For example, the data acquisition orders may relate to position information, for example a relative position of the respective sensor node to a reference and / or a global position of the respective sensor node, for example GPS data.

Furthermore, the speed and / or acceleration of a sensor node (that is to say, for example, a vehicle) can be detected.

A state information may be, for example, an operating state of a sensor node, for example of a vehicle. With reference to a vehicle, it may in particular be a driving situation, a quality or maintenance condition and / or the age of the vehicle.

The invention can be used particularly advantageously if the data acquisition jobs concern environmental information relating to the sensor nodes. This may be, for example, image information, sound information and / or video information and / or already evaluated or classified data, such as traffic signs, construction sites and / or parking facilities. For example, it may thus be provided that a data acquisition order relates to the detection of traffic signs and their global position, wherein the data acquisition order specifically to vehicles of a particular region, eg. In Ingolstadt. Furthermore, data concerning the state of public facilities or public infrastructure, such as roads, in particular data on potholes, other defects, etc., may be collected. Also information regarding parking facilities can be collected. In one development of the invention, it can be provided that at least one of the utility values relates to an economic yield, a data quality, a data quantity and / or a data actuality.

Thus, data collection jobs can be optimized using machine learning for the economic return on the re-use of the collected data. In addition to the economic aspect, other useful values can also be optimized. For example, the data up-to-dateness, together with the expected economic return, can be included in the evaluation of a data collection order, in order to ensure customer satisfaction and customer loyalty, in addition to maximizing returns on the sale of the data.

In a preferred development it can be provided that the central point offers the data acquired by the sensor nodes on a trading platform. For example, then an optimization of the data acquisition orders can be made with regard to the (expected) demand on the trading platform and in this way the economic benefit of the data can be optimized.

Of course, instead of a trading platform, the data collected can also be directly offered to third parties; for example, the collected data can be offered to card service providers, trade fair providers, town planning offices, etc. In one development of the invention, it can also be provided that the acquisition of the order-specific data by the sensor nodes is linked to order-specific conditions, wherein the order-specific conditions for optimizing the at least one value of the order-specific data using the algorithm for machine learning created by the central body become.

As a result, the data load can be further reduced since the sensor nodes only acquire data and / or transmit them to the central location, if this makes sense in terms of the utility value. In a development, it may be provided that the order-specific conditions relate to a time specification and / or an attribute of the sensor nodes, in particular a position of the respective sensor node, a state of the respective sensor node and / or a sensor node type. It may thus be provided that a sensor node only records and / or transmits job-specific data if it is located at a predetermined location. For example, the "perimeter Munich" (location relationship) and / or an arbitrary period (for example "the next four weeks") may be provided as an order-specific condition. The sensor node may itself "decide" whether to capture the data or not based on the order-specific conditions of the data collection job. An elaborate coordination with the central office is therefore unnecessary.

In an advantageous development of the invention, it can be provided that the algorithm for machine learning as an artificial neural network, as a Bayesian network, as a regression analysis, as a support vector chine, as an ensemble method, as a cluster analysis and / or as a principal component analysis, or comprises the said methods.

In particular, an artificial neural network can advantageously be used to optimize the at least one useful value of the order-specific data.

In one embodiment of the invention, the machine learning center may provide a training algorithm that is executed based on real-time data prior to the creation of the data collection jobs based on inventory data and / or during the creation of the data collection jobs.

In particular, when using a vehicle as a sensor node can be provided that the data from at least one vehicle bus, in particular from a CAN bus, a LIN bus and / or a FlexRay bus are detected. However, it can also be provided that the data is detected by using on-board sensors.

The invention also relates to a data collection system comprising a sensor network having a central location and a plurality of sensor nodes. The central office is arranged to create data collection jobs for data collection and to communicate at least one of the data acquisition jobs to at least one of the sensor nodes. Furthermore, the sensor node is set up to capture order-specific data and to transmit it to the central office.

For the system according to the invention it is further provided that the central point comprises a control device on which a machine learning algorithm can be executed in order to create the data acquisition orders with regard to the optimization of at least one useful value of the order-specific data. Features and advantages that have already been described for the method according to the invention described above, can of course be used for the inventive system.

The data collection system may be configured using the central location, preferably a server facility, such that only those order-specific data from the plurality of sensor nodes are collected and transmitted variable in time and location (with respect to the sensor nodes) for optimum yield expectation were calculated. Components of the central location may in particular be a receiving unit for the data transmitted by the sensor nodes, a processing unit and a memory unit. The machine learning algorithm can thereby optimize the data selection on the basis of metrics or at least one utility value. The result of the evaluation can finally be forwarded to a device for configuring the data acquisition orders.

The following components may be provided for the system:

 - A device for detecting the data, comprising the at least one sensor node;

- Each device for transmitting the data from the central location to the at least one sensor node and vice versa;

 - a device for calculating the economic benefits based on machine learning;

 - An apparatus for training or training a model for mechanical learning to calculate the economic benefits based on data (by inventory data and / or real-time data).

Below is an embodiment of the invention with reference to the drawing in principle shown in more detail. It shows:

1 shows a system for data acquisition by means of a sensor network, comprising a plurality of sensor nodes and a central point in a schematic representation; and FIG. 2 is a schematic representation of the method for collecting data.

FIG. 1 shows very schematically a system 1 for data collection, comprising a sensor network 2 with a central point 3 and a plurality of sensor nodes 4, 5. The sensor nodes 4, 5 are essentially motor vehicles 4 in the exemplary embodiment However, any sensor nodes can be provided, for example a mobile terminal 5, a building automation module, a traffic monitoring module and / or a hardware module of an intelligent personal assistant. The use of a vehicle fleet is to be understood in the present case only as an example and not restrictive. For clarification, two mobile terminals 5 (in the present case smartphones) are shown in dashed lines in FIG. It can also be provided to use different types of sensor nodes in the sensor network 2 mixed, for example, motor vehicles 4 and smartphones 5. A sensor node, for example, a motor vehicle 4, may include one or more sensors 6, for example, to capture environmental information.

The central point 3 is a server device in the exemplary embodiment, wherein the closer details of the subsequently described FIG. 2 can be seen.

In Fig. 2, the flow is shown schematically within the central point 3 for clarity. It is provided that the central point 3 creates data acquisition orders 7 for the data collection and transmits at least one of the data acquisition orders 7 to at least one of the sensor nodes, in the present case a motor vehicle 4. A data acquisition task 7 can include a data description part 7.1 and a condition part 7.2. In the data description part 7.1, the type of order-specific data 8 to be recorded can thus be described or coded. This may be, for example, movement information, status information and / or environment information relating to the motor vehicle 4. It is envisaged that the central body 3, the data acquisition orders 7 for optimizing the at least one utility value xi, ..., x _n of the order-specific data 8 using an algorithm f [xi, ..., x _n] created for machine learning.

The capture of the order-specific data 8 can be linked to order-specific conditions, which can be stored in the condition part 7.2 of the data acquisition order 7. It can be provided that the order-specific conditions in order to optimize at least one utility value xi, ..., x of the order-specific data 8 _n using the algorithm f [xi, ..., x _n] to the machine learning by the central body 3 it - be poses. An order-specific condition may, for example, be a time specification and / or an attribute of the sensor nodes, in particular a position of the respective sensor node, a state of the respective sensor node and / or a sensor node type.

The data 8, which is captured by the sensor nodes, that is, for example, a motor vehicle 4 or a mobile terminal 5, can be transmitted to the central location 3 and stored in the central location 3 in a data memory 9. The central point 3 preferably provides the data 8 recorded by the sensor nodes, which are stored, for example, in the data memory 9 of the central point 3, on a trading platform 10. Within the trading platform 10, the acquired data 8 can be offered to one or more customers 10.1, 10.2, 10. However, the data 8 can also be offered directly to corresponding customers without the use of a trading platform 10.

Finally, the at least one useful value Xi,... , x _n , which is to be optimized to be an economic yield, a data quality, a data quantity and / or a data actuality. For the evaluation of the economic yield with regard to the costs of data acquisition and data transmission, for example, upon the sale of the acquired data 8, feedback, for example about the yield, can be given to the central location 3 (indicated by dashed arrows 11 in FIG. 2), in which case, for example, the yield in addition to the actual data 8 is stored in the data memory 9 or a further data memory of the central point 3.

On a control device 12 of the central point 3, the algorithm [xi, ... , x _n ] for machine learning, which is preferably an artificial neural network. However, it may also be provided that any other algorithm is f [xi, ..., x _n] machine learning used, such as a Bayesian network, a regression analysis, a support vector machine, an ensemble method, a cluster analysis and / or a principal component analysis. It can be provided that the machine learning central point 3 uses a training algorithm 13 which, prior to the creation of the data acquisition jobs 7, is based on inventory data and / or during the data acquisition tasks Creation of the data acquisition jobs 7 is performed on the basis of real-time data.

Claims

P a n t a n s p r e c h e Method for collecting data by means of a sensor network (2) having a central point (3) and a plurality of sensor nodes (4, 5), wherein the central point (3) generates data acquisition jobs (7) for the data collection and at least one of the data acquisition jobs (7) transmitted to at least one of the sensor nodes (4, 5), and wherein the sensor nodes (4, 5) each capture order-specific data (8) and transmit them to the central location (3), characterized in that the central point (3) the data acquisition orders (7) for optimizing at least one utility value (xi, ..., x _n ) of the order-specific data (8) using an algorithm (f [xi, ..., x _n ]) machine learning. Method according to claim 1, characterized in that at least one of the sensor nodes as a vehicle (4), mobile terminal (5), building automation module, traffic monitoring module and / or hardware module of a smart personal assistant is formed. Method according to claim 1 or 2, characterized in that the data acquisition orders (7) relate to movement information, status information and / or environment information relating to the sensor nodes (4, 5). Method according to one of claims 1 to 3, characterized in that at least one of the utility values (xi, ..., x _n) an economic yield, a data quality, a data quantity and / or a data update. Method according to one of claims 1 to 4, characterized in that the central point (3) offers the data acquired by the sensor nodes (4, 5) on a trading platform (10). Method according to one of claims 1 to 5, characterized in that the acquisition of the order-specific data (8) by the sensor nodes (4, 5) is linked to order-specific conditions, wherein the order-specific conditions for optimizing the at least one utility value (xi, ..., x _n ) of the order-specific data (8) using the algorithm (f [xi, ..., x _n j) are created for machine learning by the central body (3). Method according to claim 6, characterized in that the order-specific conditions relate to a time specification and / or an attribute of the sensor nodes (4, 5), in particular a position of the respective sensor node (4, 5), a state of the respective sensor node (4, 5) and / or a sensor node type. Method according to one of claims 1 to 7, characterized in that the algorithm (f [xi, ..., x _n ]) is designed for machine learning as an artificial neural network, as a Bayesian network, as a regression analysis, as a support vector machine, as an ensemble method, as a cluster analysis and / or as a principal component analysis or such Includes method. 9. The method according to any one of claims 1 to 8,  characterized in that  the central machine learning point (3) provides a training algorithm (13), which is executed on the basis of inventory data and / or during the creation of the data acquisition orders (7) on the basis of real-time data prior to the creation of the data acquisition orders (7). A data collection system (1) comprising a sensor network (2) having a central location (3) and a plurality of sensor nodes (4, 5), the central location (3) being arranged for collecting data (7) and at least one of the data acquisition orders (7) to be transmitted to at least one of the sensor nodes (4, 5), and wherein the sensor nodes (4, 5) are arranged to acquire order-specific data (8) and to the central location (3) to convey  characterized in that the central point (3) comprises a control device (12) on which an algorithm (f [xi, ..., x _n ]) for machine learning is executable, at least in order to optimize the data acquisition orders (7) a utility value (xi, ..., x _n ) of the order-specific data (8) to create.