Method and device for the detection of forest fires

Abstract

The invention relates to a method for implementing ML data in a system for forest fire early detection, comprising the steps of: detecting tn-measurement data from measurements of tn (true negative)-parameters of the forest under tn-conditions; generating tn-ML-data from the tn-measurement data; and implementing tn-ML-data for the detection of forest fires in a forest fire early detection system. The invention also relates to a forest fire early detection system comprising a LoRaWAN network having a terminal, wherein the terminal has a sensor device, a first control device, an evaluation device for evaluating measurement signals provided by the sensor device, and a device for supplying power, as well as a network server, wherein the first control device is designed and provided for accessing a memory, containing data from the adaptation and application of a machine learning model, and wherein the machine learning model comprises tn-ML-data.

Description

[0001] VE RFA HRENUND VO RR ICH TU NGZURD ET E KTI ON VO N WA LDB RÄ NDEN

[0002] The invention relates to a method for implementing ML data (machine learning data) in a system for early forest fire detection, comprising the method steps of acquiring tn measurement data from measurements of tn (true negative) parameters of the forest under tn conditions, creating tn-ML data from the tn measurement data, and implementing tn-ML data for detecting forest fires in a forest fire early detection system.The invention further relates to a forest fire early detection system with a LoRaWAN network comprising a terminal device, wherein the terminal device comprises a sensor device, a first control device, an evaluation device for evaluating measurement signals supplied by the sensor device and a device for supplying energy, as well as a network server, wherein the first control device is suitable and intended to access a memory which comprises data from the adaptation and application of a machine learning model, and wherein the machine learning model comprises tn-ML data.

[0003] State of the art

[0004] The larger a forest fire, the more difficult it is to determine its direction and speed of spread. Weather, wind, soil conditions, and vegetation determine its path and speed of spread, which can change within a short period of time. It is therefore very important to detect a forest fire very early in order to minimize damage and keep the fire manageable, as well as to give the fire department a decisive time advantage.

[0005] During a wildfire, the complex thermal decomposition processes (distillation, pyrolysis, charring, and the oxidation of the resulting gas products during flame combustion) occur simultaneously and often in close proximity to one another. The thermal decomposition of fuels occurs in front of and along the fire line, while pockets of intermittent open flame often persist far behind the flame front.

[0006] Flame combustion generally occurs between 800°C - 1200°C. Smoldering ground fires occur between 300°C - 600°C. Combustible gases, particularly volatile organic compounds (VOCs), are formed more quickly at temperatures above 200°C and reach their peak at 320°C. VOCs are the collective term for organic, carbon-containing substances that evaporate into the gas phase at room temperature or higher temperatures, particularly terpenes. Various organic compounds are also formed, such as methanol, carbon dioxide, carbon monoxide, and molecular hydrogen. Flammable combustion only begins at 425°C to 480°C. Flame temperatures of 700°C to 1300°C are the most common. In this temperature range, carbon dioxide, nitrogen oxides, and volatile sulfur-containing compounds (VSCs), particularly sulfur dioxide, are mainly formed. Smoldering fires spread slowly, approx.3 cm / h, they can generate ground temperatures above 300°C for several hours with peak temperatures of 600°C.

[0007] The following table shows the gases formed during a forest fire, graded by temperature:

[0008] Earth observation data, particularly in the form of aerial and satellite imagery, can potentially help in detecting wildfires. The sharp increase in available earth observation data, particularly aerial and satellite imagery, enables widespread detection of wildfires. While satellite data are useful for detecting and fighting fires, they have one drawback: They usually reach emergency responders only with a delay. Geostationary satellites, due to their great distance, provide only low image resolutions, and non-geostationary satellites must orbit the Earth before they can provide new images.Another option for detecting forest fires is to install a network of gas sensors directly in the forest. These sensors detect gases released during the development of forest fires, thus enabling early detection of forest fires before they can be detected by optical systems from a distance. However, due to the varying vegetation types and soil compositions in forests, different gases and gas concentrations are produced, making accurate detection very difficult. Furthermore, the increasing temperatures during the different phases of forest fire development alone result in different gases and gas concentrations. However, since deploying emergency personnel is very expensive, the detection accuracy needs to be improved.

[0009] It is therefore an object of the present invention to provide a method for implementing ML data in a system for early forest fire detection and / or early forest fire detection, which works reliably, is expandable as required and is cost-effective in installation and maintenance.

[0010] It is also an object of the invention to provide a forest fire early detection system that operates reliably, can be expanded as required and is cost-effective to install and maintain.

[0011] The object is achieved by means of the method for implementing ML data in a forest fire early detection system according to claim 1. Advantageous embodiments of the invention are set forth in the subclaims.

[0012] The inventive method for implementing ML data in a forest fire early detection system comprises three method steps: In the first method step, tn measurement data are collected from measurements of tn parameters of a forest under tn conditions.

[0013] For the purposes of this document, tn parameters (true negative parameters) are parameters recorded from a forest that is not experiencing a forest fire. Such parameters include, for example, temperature, humidity, wind direction and strength, and / or the composition of gases. The tn measurement data recorded with these tn parameters (true negative measurement data) are accordingly measured data from a forest that is also not experiencing a forest fire. The recorded tn measurement data therefore represent measurement data from a forest that is not experiencing a forest fire. By recording tn measurement data, the baseline and long-term deviations of the measured values ​​are compensated. In addition, design-related deviations of the sensors installed in the terminal device are recorded, averaged, and compensated.

[0014] In the second step, tn-ML data is created from the tn measurement data. tn-ML data is data created from the tn measurement data using the algorithm of a machine learning model.

[0015] In the third step, tn-ML data for detecting wildfires is implemented into a wildfire early warning system. The tn-ML data is integrated into a machine learning model to create models for detecting wildfires. The wildfire detection data is made available via APIs and graphical tools.

[0016] The machine learning model is used in this invention to improve the efficiency of the sensor device of the terminal device. The model's algorithm enables improved application-specific detection of the gases to be detected. In particular, the sensitivity of a forest fire early detection system is increased. In addition, the accuracy is increased, i.e., false positives are reduced or eliminated. Forest fire detection based on threshold values ​​is error-prone and generates many false positives. With machine learning models, especially models created based on locally acquired data, the accuracy can be increased enormously, as the model adapts to reality. The ML algorithm enables improved application-specific detection of detected measured values. In addition, the algorithm corrects the detected measured values ​​with regard to air humidity.In addition, the baseline and long-term deviations of the measured values ​​are compensated. By evaluating this data, statements can be made about the current situation during forest fires. ML data in the sense of this invention is data to which a machine learning algorithm has been applied. The ML data contains measurement data on the type and concentrations of different gases as well as their type and concentrations at different temperatures in the different phases of forest fire development. In particular, data on volatile organic compounds (VOCs), volatile sulfur-containing compounds (VSCs) as well as carbon dioxide, carbon monoxide and molecular hydrogen are included, preferably in the ppb range. Furthermore, data on air humidity, ambient temperature and air pressure are optionally included, also recorded at different temperatures in the different phases of forest fire development and created by the machine learning algorithm.

[0017] With the help of the machine learning algorithm, the forest fire early warning system is able to recognize patterns and regularities based on existing data sets and algorithms. The findings can be generalized and used to solve new problems or analyze previously unknown data. Reinforcement learning is preferred for this. Through rewards and punishments, the algorithm learns a strategy for how to act in potentially occurring situations in order to maximize the system's benefit. The machine learning algorithm interacts with the environment and is evaluated by a cost function or reward system. With reinforcement learning, the machine learning algorithm is not shown which action is the right one in which situation, but rather receives positive or negative feedback via the cost function.Based on the cost function, it then assesses which action is the right one at which point in time. Thus, the system learns to maximize the reward function through "reinforcement" through praise or punishment.

[0018] A neural network can also be used. These networks are capable of automatically learning key features relevant to solving a problem from the data. Unlike traditional machine learning methods, in which the features must be laboriously designed and defined by humans, neural networks independently learn the correlations and features from the data.

[0019] In a further development of the invention, a forest fire is detected based on the collected result data. A forest fire early detection system uses sensors to collect various measurement data from an area to be monitored. The result data is acquired using a terminal device. The result data can include the result of comparing the measurement data with the ML data, simply an evaluation of the comparison of the measurement data with the ML data, and / or a simple warning signal. In any case, the result data indicates whether a forest fire was detected by the terminal device's sensor.

[0020] In a further embodiment of the invention, the measurement data are acquired by a terminal device. The terminal device has one or more suitable sensor devices, e.g., for gas analysis, and / or is connected to such sensor devices.

[0021] In a further embodiment of the invention, the result data is determined by a terminal device. The terminal device has an evaluation device for this purpose. The result data is determined by applying ML data to the measurement data acquired by the terminal device.

[0022] In a further embodiment of the invention, the tn-ML data is implemented in the terminal device. The ML data is implemented in the terminal device instead of in a central unit, e.g., a network server. A forest fire early warning system typically has a large number of terminal devices that are widely distributed and have self-sufficient energy supply systems. The implementation of the ML data, i.e., the storage of the ML data on the terminal device and the application of the ML data to measurement data acquired by the terminal device in the terminal device enables the machine learning model to be adapted and applied to the local conditions of the respective terminal device. At the same time, the energy consumption of an individual terminal device is reduced because only a reduced amount of data needs to be transmitted. In an optional embodiment of the invention, the ML algorithm is also installed on the terminal device.Edge computing is used here to evaluate the data locally in the end device. This is important and advantageous because it eliminates the need to transmit the data to a central cloud application, thus reducing the load on the (very narrowband) LoRaWAN network, for example.

[0023] In a further embodiment of the invention, the result data is transmitted to a network server. On the network server, the result data is available to other applications that detect and record a forest fire. By integrating the measurement data into a machine learning model, models for detecting forest fires are also created using the network server. The result data can include the result of comparing the measurement data with the machine learning data, simply an evaluation of the comparison of the measurement data with the machine learning data, and / or a simple warning signal.

[0024] In a further aspect of the invention, only a portion of the result data is transmitted to the network server.

[0025] In a further development of the invention, transmission occurs using a protocol such as LoRa, LoRaWAN, and / or IP. LoRa requires particularly low energy and is based on chirping frequency spread modulation according to US patent US 7791415 B2. Licenses for its use are granted by a founding member of the industry consortium, Semtech. LoRa uses license- and permit-free radio frequencies in the sub-1 GHz range, such as 433 MHz and 868 MHz in Europe or 915 MHz in Australia and North America, thus allowing a range of more than 10 kilometers in rural areas with minimal energy consumption. LoRa technology consists of the physical LoRa protocol and the LoRaWAN protocol, which is defined and managed as the upper network layer by the industry consortium LoRa Alliance.LoRaWAN networks implement a star-shaped architecture using gateway message packets between the end devices and the central network server. The gateways (also called concentrators or base stations) are connected to the network server via the standard Internet Protocol, while the end devices communicate wirelessly with the respective gateway using LoRa (chirp frequency spread spectrum modulation) or FSK (frequency spectrum modulation). The wireless connection is thus a single-hop network, in which the end devices communicate directly with one or more gateways, which then forward the data traffic to the Internet. Conversely, data traffic from the network server to an end device is routed via a single gateway. Data communication generally works in both directions, although data traffic from the end device to the network server is the typical application and the predominant operating mode.By bridging larger distances with very low energy consumption, LoRaWAN is particularly suitable for IoT applications outside of settlements.

[0026] At the physical level, LoRaWAN, like other wireless protocols for IoT applications, uses spread spectrum modulation. It differs from conventional DSSS (Direct Sequence Spread Spectrum Signaling) in that it uses an adaptive technique based on chirp signals. Chirp signals offer a compromise between reception sensitivity and maximum data rate. A chirp signal is a signal whose frequency varies over time. LoRaWAN technology is cost-effective to implement because it does not rely on a precise clock source. LoRa's ranges extend up to 40 kilometers in rural areas. In urban areas, the advantage is good building penetration, as even basements are accessible. Power consumption is very low, at around 10 nA and 100 nA in idle mode. This allows for a battery life of up to 15 years.

[0027] LoRaWAN defines and uses a star-topology network architecture, where all leaf nodes communicate via the most suitable gateway. These gateways handle routing and can redirect communication to an alternative if more than one gateway is within range of a leaf node and the local network is congested.

[0028] Some other IoT protocols (such as ZigBee or Z-Wave) use so-called mesh network architectures to increase the maximum distance of a leaf node from a gateway. The end devices in the mesh network forward messages among themselves until they reach a gateway, which forwards the messages to the internet. Mesh networks program themselves and dynamically adapt to environmental conditions without requiring a master controller or hierarchy. However, to forward messages, the end devices in a mesh network must be either constantly or periodically ready to receive messages and cannot be put into sleep mode for long periods of time. This results in higher energy requirements for the node end devices to forward messages to and from the gateways, resulting in a reduction in battery life.

[0029] LoRaWAN's star network architecture, on the other hand, allows end devices to enter a power-saving sleep mode for extended periods, ensuring that the device's battery is subjected to as little strain as possible and can therefore operate for several years without battery replacement. The gateway acts as a bridge between simple, battery-optimized protocols (LoRa / LoRaWAN), which are better suited for resource-constrained end devices, and the Internet Protocol (IP), which is used to provide IoT services and applications. After the gateway receives the data packets from the end device via LoRa / LoRaWAN, it sends them via the Internet Protocol (IP) to a network server, which in turn has interfaces to IoT platforms and applications.

[0030] In a further embodiment of the invention, the result data is collected on the terminal device. The result data is stored in the terminal device's memory until it is transmitted as a data packet via one or more gateways to the network server within a download / receive window. The terminal device does not need to have a permanently active download / receive window and therefore be permanently active, as is the case with a Class C terminal device, but can also be a Class A or Class B terminal device according to the LoRaWAN specification, for example. The power consumption of a terminal device is thus minimized.

[0031] In a further embodiment of the invention, the collected result data is transmitted to the network server at specified intervals. The end devices are divided into three different bidirectional variants: Class A comprises communication according to the ALOHA access method. With this method, the device sends its generated data packets to the gateway, followed by two download-receive windows that can be used to receive data. A new data transfer can only be initiated by the end device during a new upload. Class B end devices, on the other hand, open download-receive windows at specified times. For this purpose, the end device receives a time-controlled beacon signal from the gateway. This tells a network server when the end device is ready to receive data. Class C end devices have a permanently open download-receive window and are therefore permanently active, but also have increased power consumption.In order to minimize the energy consumption of the terminal devices, usually only class A and B terminal devices are used to carry out the method according to the invention.

[0032] In a further development of the invention, the intervals are defined based on time or data volume. Class B terminals transmit the result data at specified times. Class A terminals can also send the result data to the network server at specified times. However, they can also have the option of transmitting the result data when the result data has a specified data volume. This prevents the data volume from becoming too large for the terminal's memory.

[0033] In a further embodiment of the invention, the terminal device has a communication unit, which is deactivated after the result data has been transmitted. The result data is transmitted from the terminal device to the network server via the communication unit. The communication unit is deactivated after the result data has been transmitted in order to reduce the power consumption of the terminal device.

[0034] In an advantageous embodiment of the invention, an ML algorithm is applied to the result data. The ML algorithm enables improved application-specific detection of the gases to be detected. Furthermore, the algorithm corrects the detected gas concentration in relation to the detected air humidity. In addition, the baseline and long-term deviations of the measured values ​​are compensated. For this purpose, data on different gas compositions and their concentrations are provided to the terminal device, which is compared with the gas compositions and their concentrations determined by the sensor device.

[0035] In a further embodiment of the invention, the first application of the ML algorithm occurs before the software is installed on the end device and / or before the sensor device is mounted within a forest fire monitoring system. Reinforcement learning is preferably used for this purpose. Through reward and punishment, the algorithm learns a strategy for how to act in potentially occurring situations in order to maximize the benefit of the forest fire monitoring system.

[0036] In a further development of the invention, the ML algorithm is applied after the software has been installed on the terminal device and / or after the sensor device has been mounted within a forest fire monitoring system. This approach has the advantage that the adaptation and application of the machine learning model can be adapted to the on-site conditions. In a further development of the invention, the adaptation and application of the machine learning model is carried out via a wireless network. In particular, the adaptation and application of the machine learning model of the terminal device is updated via the control unit, preferably at regular intervals.

[0037] In an advantageous embodiment of the invention, the tn measurement data is obtained from true-negative events, whereby the tn measurement data is obtained from a laboratory test and / or in a natural environment. The recorded tn measurement data is measured from a forest that has not experienced a forest fire. Under laboratory conditions, the tn measurement data is recorded under idealized conditions and is reproducible. By recording tn measurement data, the baseline and long-term deviations of the measured values ​​of a terminal device are compensated. In addition, design-related deviations of the sensors installed in the terminal device are recorded, averaged, and compensated. In a further development of the invention, the tn measurement data is obtained from the natural environment for which the forest fire early detection system is intended.The recorded tn measurement data are therefore specific to the forest in which the system is deployed and therefore only partially transferable to other environments. However, for the environment in which the forest fire early detection system is deployed, the recorded tn measurement data provide a better baseline and correction for long-term deviations in the measured values ​​of a terminal device.

[0038] In a particularly advantageous embodiment of the invention, tp measurement data is additionally acquired, from which tp-ML data is determined. Using exclusively tn measurement data and generating tn-ML data, baseline and long-term deviations are compensated. However, the forest fire early detection system can trigger a false alarm if the acquired measurement data and the resulting result data of a terminal device deviate from these tn-ML data, even if no forest fire is present in the area being monitored.

[0039] Collecting tp (true-positive) measurement data—that is, data from events whose measurement data represent a forest fire—and subsequently generating a tp-ML dataset increases the reliability of the measurement values ​​acquired by a device and the resulting result data, to which both the tn-ML dataset and the tp-ML dataset are applied. This significantly reduces the rate of false alarms.

[0040] In a further development of the invention, the tp measurement data are acquired from true-positive events. For the purposes of this document, true-positive events are events whose measurement data represent a fire, in particular a forest fire.

[0041] In a further aspect of the invention, the true-positive events are simulated in the laboratory. Under laboratory conditions, simulation of the true-positive events is possible safely and reproducibly.

[0042] In a further embodiment of the invention, tp-ML data and tn-ML data are combined to form ML data. The ML data thus comprises tp-ML data and tn-ML data, which are stored in the terminal device and applied to the terminal device's measurement data. This significantly reduces the rate of false alarms.

[0043] In a further development of the invention, the result data is determined from the measurement data and the ML data. Capturing tp measurement data—that is, from events whose measurement data represent a forest fire—and subsequently generating a tp-ML dataset increases the reliability of the measurement values ​​acquired by a terminal device and the resulting result data, to which both the tn-ML dataset and the tp-ML dataset are applied. This significantly reduces the rate of false alarms.

[0044] In a further embodiment of the invention, the tn measurement data are obtained from different regions of the natural environment for which the forest fire early detection system is intended. Depending on the extent of the forest to be monitored, the tn measurement data for different regions of the forest to be monitored also vary. Using different tn measurement data for different regions, a regionally spatially resolved correction of the baseline and long-term deviations can therefore be carried out.

[0045] In a further embodiment of the invention, different tn-ML data are created from the tn measurement data obtained from different regions of the natural environment. Depending on the extent of the forest to be monitored, the tn measurement data for different regions and the tn-ML data generated from the tn measurement data for the forest to be monitored are different.

[0046] In a further embodiment of the invention, different regions of the forest to be monitored are assigned to different tn-ML data. Using different tn-ML data for different regions and the corresponding location information of the tn-ML data, a regionally spatially resolved correction of the baseline as well as long-term deviations of measured values ​​from a terminal device can be carried out. In a further embodiment of the invention, to determine the result data, the measured values ​​recorded from a specific region of the forest to be monitored are compared with the tn-ML data assigned to this region. Using different tn measurement data for different regions, a regionally spatially resolved correction of the baseline as well as long-term deviations can be carried out.

[0047] In a further advantageous embodiment of the invention, measurement data is acquired by the forest fire early detection system, with the resulting data being determined by applying the tn-ML data to the acquired measurement data. Depending on the extent of the forest W to be monitored, tn measurement data from different regions of the natural environment are also acquired. From these tn measurement data acquired in this way, regionally different tn-ML data are then created, with the tn-ML data differing from region to region. The different tn-ML data are therefore assigned to different regions of the forest W to be monitored and applied to the measurement data of the forest fire early detection system with spatial resolution.

[0048] The problem is also solved by means of a forest fire early detection system with a LoRaWAN network. Advantageous embodiments are set forth in the following subclaims.

[0049] The inventive forest fire early detection system with a LoRaWAN network has a terminal device. The terminal device has a sensor device that has one or more sensors, e.g., for gas analysis. The inventive forest fire early detection system also has a first control device, an evaluation device for evaluating the measurement signals supplied by the sensor device, and a power supply device. The power supply device enables autonomous operation of the terminal device, for example by charging a battery via solar cells. The inventive forest fire early detection system also has a network server. The network server has interfaces to other applications with which, for example, the direction and speed of spread of a forest fire can be determined.According to the invention, the first control device is suitable and intended to access a memory containing data from the adaptation and application of a machine learning model. The model's algorithm enables improved application-specific detection of the gases to be detected. Furthermore, the algorithm corrects the detected gas concentration with respect to the detected air humidity. In addition, the baseline and long-term deviations of the measured values ​​are compensated. For this purpose, data on different gas compositions and their concentrations are provided to the sensor system, which is compared with the gas compositions and their concentrations determined by the sensor.

[0050] According to the invention, the machine learning model comprises tn-ML data. tn-ML data is data created from tn measurement data using the algorithm of a machine learning model. The tn measurement data (true negative measurement data) acquired with these tn parameters are measurement data measured from a forest that also has no forest fires. The acquired tn measurement data therefore represents measurement data from a forest that has no forest fires. By acquiring tn measurement data, the baseline and long-term deviations of the measured values ​​are compensated. In addition, design-related deviations of the sensors installed in the end device are recorded, averaged, and compensated.

[0051] In a further development of the invention, the machine learning model comprises tp-ML data. Collecting tp measurement data—that is, for events whose measurement data represent a forest fire—and subsequently generating a tp-ML dataset increases the reliability of the measured values ​​acquired by a terminal device and the resulting result data, to which both the tn-ML dataset and the tp-ML dataset are applied. This significantly reduces the rate of false alarms.

[0052] In one development of the invention, the memory is part of the terminal device. The terminal device has a housing to protect the components from the elements. The memory is also arranged in the housing and connected to the first control device. In a further embodiment of the invention, the network server is coupled to a second control device, which is suitable and intended to execute a machine learning program. The second control device has a system that has a machine learning algorithm. The machine learning algorithm uses training data to improve the machine learning model.

[0053] In a further embodiment of the invention, the second control device has access to the measurement signals acquired by the terminal device. The measurement signals acquired by the terminal device are training data with which a machine learning algorithm of the second control device is trained.

[0054] In a further embodiment of the invention, the second control device is connected to the terminal via two different networks.

[0055] In a further embodiment of the invention, the terminal device has a humidity sensor for detecting the air humidity. Humidity, particularly relative humidity, is an indicator of the risk of forest fires.

[0056] In a further development of the invention, the terminal device has a temperature sensor for detecting the ambient temperature. One indicator of the presence of a forest fire is the air temperature.

[0057] In a further embodiment of the invention, the terminal device has a pressure sensor for detecting the air pressure. By detecting the air pressure, predictions of the wind direction and speed, and thus also of the propagation speed and direction, can be made.

[0058] Exemplary embodiments of the inventive method for implementing ML data in a forest fire early detection system and of the inventive forest fire early detection system are shown in simplified schematic form in the drawings and are explained in more detail in the following description. They show:

[0059] Fig. 1 : Structure of a forest fire early detection system comprising a LoRa radio network with transmission of result data and ML data, ML units fixed

[0060] Fig. 2: Structure of a forest fire early detection system comprising a LoRa radio network with transmission of result data and ML data, mobile ML units

[0061] Fig. 3: Detailed view of a forest fire early detection system comprising a LoRa-

[0062] Wireless network with transmission of result data and ML data, mobile and fixed ML units

[0063] Fig. 4: Mobile forest fire detection unit with an ML unit

[0064] Fig. 5 a: Structure of an ML unit with a forest fire detection sensor

[0065] Fig. 5 b: Structure of an ML unit with two different forest fire detection sensors

[0066] Fig. 5 c: Structure of an exemplary embodiment of an ML unit with three different forest fire detection sensors

[0067] Fig. 1 and Fig. 2 each show an early forest fire detection system 1 according to the invention. The early forest fire detection system 1 has a plurality of terminal devices ED. For detecting a forest fire, a single terminal device ED has a sensor unit that includes sensors for determining air humidity, air pressure, and a temperature sensor. Optionally or additionally, a terminal device ED has sensors for gas analysis and for detecting the prevailing wind direction, with which the composition and concentration of gases as well as their direction of propagation are determined. Furthermore, a terminal device has a communication unit, a control unit, and a memory on which a set of ML data is stored.

[0068] A gateway G has a communication interface to both an end device ED for data exchange and a border gateway BGD. The connection to the border gateway BGD can be established via a meshed multi-hop network (MHF), while the connection to the end device ED is a single-hop connection (FSK). The two communication interfaces of the gateway G use different communication channels, so the sender can be identified via the communication channel used.

[0069] A border gateway (BGD) has a communication interface to a gateway (G) and to the network server (NS). The border gateway (BGD) then sends the data from an end device (ED), which was sent to the border gateway (BGD) via a single-hop and multi-hop connection, directly to the network server (NS) using the Internet Protocol (IP). Communication between the border gateway (BGD) and the network server (NS) can be via wired (WN) or wireless (IP). Each communication interface of the border gateway (BGD) uses its own communication channel, which is different from the other communication interfaces.

[0070] The network server NS is connected to a second control unit that is suitable and intended to execute a machine learning program. In particular, the ML data set of the terminal device ED is updated via the second control unit, preferably at regular intervals.

[0071] The inventive method for implementing ML data in a forest fire early detection system comprises three method steps: In the first method step, tn measurement data are collected from measurements of tn parameters of the forest W under tn conditions.

[0072] The tn parameters are recorded using the ML units ML located in the forest fire early detection system 1 (see Fig. 5). These ML units ML are connected to gateways G via a single-hop connection FSK, similar to an end device ED. A gateway G is connected to the network server NS via border gateways BGD via a multi-hop connection MHF.

[0073] The first control unit C of an ML unit ML collects the measured values ​​of the sensor device of the ML unit ML and stores them in the memory. The first control device C of the ML unit ML creates tn ML data from the acquired tn parameters by applying ML data to the acquired tn measurement data.

[0074] In the third method step, tn-ML data for detecting forest fires are implemented in a forest fire early detection system 1. For this purpose, the memory of a terminal device contains a tn-ML data set that was stored in the memory prior to the installation of the software of the sensor device and / or in particular prior to the installation of the terminal devices ED within a forest fire monitoring system 1.

[0075] The ML unit ML records tn parameters from the forest W to be monitored, which is not experiencing a forest fire. Such parameters include, for example, ambient temperature, air humidity, wind direction and strength and / or the composition of gases, depending on the sensors arranged in the terminal device ED. From these tn parameters, tn measurement data of the forest Win to be monitored are generated in a second process step. The recorded tn measurement data are therefore generated from a forest W that is also not experiencing a forest fire. The tn measurement data are preferably recorded and generated from the natural forest W for which the forest fire early detection system 1 is intended. Depending on the extent of the forest W to be monitored, tn measurement data from different regions of the natural environment are also recorded. From these tn measurement data recorded in this way, regionally different tn ML data are then created, whereby the tn ML data differs from region to region.The different tn-ML data are therefore assigned to different regions of the forest W to be monitored and are compared with the measurement data recorded by the terminal ED.

[0076] To acquire the tn measurement data, mobile, particularly airborne, ML units ML (see Fig. 2, Fig. 4) can be used alternatively or in addition to the fixed ML units ML. A plurality of forest fire detection devices 100 are arranged distributed in and around the forest W. Each forest fire detection device 100 has a forest fire detection station 200 and a mobile forest fire detection unit 300 with an ML unit ML (see Fig. 4).

[0077] Alternatively, or in addition to collecting tn measurement data from natural environments, tn measurement data can also be collected in a defined and standardized laboratory environment. By collecting tn measurement data, baseline and long-term deviations in the measured values ​​are compensated. Furthermore, design-related deviations of the sensors installed in the ED terminal device are recorded, averaged, and compensated.

[0078] Advantageously, as explained, not only are tn parameters and tn measurement data recorded and generated from true-negative events, but tn measurement data are also recorded from true-positive events. In the simplest case, a true-positive event is a forest fire. For this purpose, for example, forest components, such as the fauna found in forests, forest soil components, and / or loose material on the forest floor, are heated and / or burned at various temperatures in a laboratory, and the resulting gases are detected. This can optionally be done specifically for a forest W to be equipped with a forest fire early detection system 1.

[0079] The tp-ML dataset is determined from these experimentally obtained measurement data in the laboratory. Therefore, an ML dataset is generated with data on true-positive events – i.e., events whose measurement data represent a forest fire. To determine the result data, the measured values ​​recorded from a specific region of the forest W to be monitored are compared with the tn-ML data assigned to this region. This enables the terminal device ED, with the help of its control unit, to compare the recorded measurement data with the tn-ML data and tp-ML data. If the recorded result data match the stored tp-ML data, it sends a corresponding message to the network server NS via the communication interface of the terminal device ED. The first ML dataset is loaded onto the terminal device ED before the forest fire early detection system is installed.

[0080] To detect a forest fire, measurement data is collected by the sensor device of the terminal device ED of the forest fire early detection system 1. The measurement data is typically not collected continuously, but at adjustable intervals. This reduces the power consumption of the terminal device ED. From this measurement data, the terminal device ED generates a set of result data using an ML data set stored in the memory of the terminal device ED. The set of result data is sent from a terminal device ED to a gateway G at specified time intervals (time-based or data volume-based). The gateway G forwards this result data to the network server NS, which forwards the result data to the second control unit. On the second control unit, a machine learning algorithm is applied to the result data, thus generating an ML data set.The second control unit sends the ML data set to the network server NS, which sends the ML data set back to the gateway G. The gateway G then forwards the ML data set to the terminal ED. The ML data set is received by the terminal ED and stored in the memory of the terminal ED in such a way that the ML data set sent by the second control unit replaces the ML data set previously stored in the memory of the terminal ED.

[0081] With such machine learning algorithms, fire detection can be reliably achieved even in remote areas. By analyzing this data, conclusions can be drawn about the current situation following forest fires.

[0082] Fig. 3 shows a detailed view of an early forest fire detection system 1 according to the invention. The early forest fire detection system 1 has a LoRaWAN mesh gateway network 10 with a plurality of forest fire detection sensors ED, wherein eight forest fire detection sensors ED each communicate with a gateway G via a single-hop connection FSK. The gateways G are connected to each other and to border gateways BGD. The border gateways BGD are connected to the internet network server NS, either via a wired connection WN or via a wireless connection using the internet protocol IP. A plurality of forest fire detection devices 100 are arranged around the forest W to be monitored.

[0083] To collect tn-ML data, permanently installed ML units ML and a plurality of forest fire detection devices 100 are arranged in the forest W to be monitored. Each forest fire detection device 100 has a forest fire detection station 200 and a mobile forest fire detection unit 300 (see Fig. 4).

[0084] Fig. 4 shows an embodiment of a forest fire detection unit 300, which, like a terminal device, has ED sensors 330, 340 for detecting a forest fire, as well as an ML unit 310. The forest fire detection unit 300 is designed as a flight-capable drone that is autonomous, automatic, and / or remotely controllable. The forest fire detection unit 300 has a drive unit 320 with a plurality of rotors 322 driven by motors 321. The motors 321 are typically electric motors and are supplied with power by a rechargeable energy storage device (battery). The forest fire detection unit 300 can be controlled by pivoting the rotors 322 and varying the speed of the individual motors 321. The forest fire detection unit 300 has the second forest fire detection sensor 330, which in this embodiment is an infrared camera.In this exemplary embodiment, the forest fire detection unit 300 also includes a further forest fire detection sensor 340, which is designed as a pressure sensor. To acquire tn-ML measurement data, the forest fire detection unit 300 includes the ML unit 310, which is detachably connected to the forest fire detection unit 300 via the connection 312 in the receptacle 311.

[0085] The forest fire detection unit 300 according to the invention also has a navigation sensor 350 that detects objects in the vicinity of the forest fire detection unit 300. The navigation sensor 350 has one or a plurality of cameras and / or time-of-flight-based sensors (e.g., radar, ultrasound, lidar) that detect obstacles during the flight of the forest fire detection unit 300. The obstacles are detected, recognized, and analyzed by the control unit arranged in the forest fire detection unit 300 such that the forest fire detection unit 300 automatically avoids the obstacles during its flight.

[0086] Fig. 5 shows three variants of an embodiment of an ML unit ML, which is arranged within a system 1 for early forest fire detection (see Fig. 1, Fig. 3). The ML unit ML, like a terminal device ED, is also a sensor for detecting a forest fire. In order to be able to install and operate the ML unit ML in inhospitable and particularly rural areas far away from any power supply, the ML unit ML is equipped with a self-sufficient power supply E. In the simplest case, the power supply E is a battery, which can also be designed to be rechargeable. However, the use of capacitors, in particular supercapacitors, is also possible. Somewhat more complex and cost-intensive, but a power supply E that offers the ML unit ML a very long service life is the use of solar cells.

[0087] In addition, an ML unit ML has the actual sensor unit, which has one or more sensors S1.1, S1.2, S2.1, S2.2, S3.1, S3.2 for detecting a forest fire. The sensors S1.1, S1.2, S2.1, S2.2, S3.1, S3.2 are connected to the computing unit C. The ML unit ML has a first temperature sensor S1.1 and an identical second temperature sensor S1.2 (Fig. 5 a). Furthermore or additionally, a first S2.1 and identical second humidity sensor S2.2 (Fig. 5 b) as well as a first S3.1 and identical second pressure sensor S3.2 (Fig. 5 c) are arranged in the ML unit. Furthermore, sensors for gas analysis can be arranged in the ML unit. By arranging two (or more) identical sensors S1.1, S1.2, S2.1, S2.2, S3.1, S3.2 for detecting a forest fire in the ML unit, design-related deviations and inaccuracies of the sensors S1.1, S1.2, S2 installed in the terminal device are taken into account when recording measurement data.1 , S2.2, S3.1 , S3.2 recorded, averaged and compensated.

[0088] Using the communication interface K1, ML data packets from the ML unit ML are sent wirelessly as data packets to a gateway G via a single-hop FSK connection via LoRa (chirp frequency spread spectrum modulation) or frequency modulation. The ML unit is connected to one or more gateways G via the output interface A via a single-hop FSK connection. All of the above components are housed in a housing for protection against weather influences.

[0089] B EZ UG S CHARACTERS LIST

[0090] 1 forest fire detection system

[0091] 10 LoRaWAN mesh gateway network

[0092] ED terminal / First forest fire detection sensor

[0093] G Gateway

[0094] NS Internet Network Server

[0095] IP Internet Protocol

[0096] MHF multi-hop radio network

[0097] MDG Mesh Gateways

[0098] BGD Border Gateway

[0099] FSK FSK modulation

[0100] WN Wired connection

[0101] W Forest

[0102] 100 forest fire detection devices

[0103] 200 forest fire detection stations

[0104] 300 forest fire detection unit

[0105] 310 ml unit

[0106] 311 recording

[0107] 312 Detachable connection

[0108] 320 Flight propulsion / propulsion unit

[0109] 321 engine

[0110] 322 Rotor

[0111] 330 Second forest fire detection sensor / IR camera

[0112] 340 forest fire detection sensor

[0113] 350 navigation sensor

[0114] A Exit

[0115] C computing unit

[0116] E Energy supply

[0117] K1 Communication interface ML ML unit

[0118] 51.1, S1.2 Temperature sensor

[0119] 52.1, S2.2 Humidity sensor

[0120] 53.1, S3.2 pressure sensor

[0121] W Forest

Claims

PATENTS Method for implementing ML data in a system (1) for early detection of forest fires, comprising the steps • Determination of tn measurement data from measurements of tn parameters of a forest (W) under tn conditions • Creating tn-ML data from the tn measurement data • Implementation of tn-ML data for detecting forest fires in a forest fire early detection system (1). Method for implementing ML data in a system (1) for forest fire early detection according to claim 1, characterized in that a forest fire is detected based on the determined result data. Method for implementing ML data in a system (1) for forest fire early detection according to claim 1 or 2, characterized in that the measurement data are acquired by a terminal device (ED). Method for implementing ML data in a system (1) for forest fire early detection according to one or more of the preceding claims, characterized in that the result data are determined by a terminal device (ED). Method for implementing ML data in a system (1) for early detection of forest fires according to one or more of the preceding claims, characterized in that the tn-ML data is implemented in the terminal device (ED). Method for implementing ML data in a system (1) for early detection of forest fires according to one or more of the preceding claims, characterized in that the result data is transmitted to a network server (NS). Method for implementing ML data in a system (1) for early detection of forest fires according to claim 6, characterized in that only a portion of the result data is transmitted to the network server (NS). Method for implementing ML data in a system (1) for early detection of forest fires according to claim 6 or 7, characterized in that the transmission takes place using a protocol such as LoRa, LoRaWAN and / or IP.Method for implementing ML data in a system (1) for early forest fire detection according to one or more of claims 6 to 8, characterized in that the result data are collected on the terminal device (ED). Method for implementing ML data in a system (1) for early forest fire detection according to claim 9. characterized in that the collected result data are transmitted to the network server (NS) at specified intervals. Method for implementing ML data in a system (1) for early forest fire detection according to claim 10, characterized in that the intervals are defined based on time or data volume. Method for implementing ML data in a system (1) for early forest fire detection according to one or more of claims 6 to 11, characterized in that the terminal (ED) has a communication unit, wherein the communication unit is deactivated after the transmission of the result data. Method for implementing ML data in a system (1) for early forest fire detection according to one or more of the preceding claims, characterized in that an ML algorithm is applied to the result data.Method for implementing ML data in a system (1) for early detection of forest fires according to one or more of the preceding claims, characterized in that the first application of the ML algorithm takes place before the installation of the software on the terminal (ED) and / or before the installation of the sensor device within a forest fire early detection system (1).

15. Method for implementing ML data in a system (1) for early detection of forest fires according to one or more of the preceding claims, characterized in that an application of the ML algorithm takes place after the installation of the software on the terminal (ED) and / or after the installation of the sensor device within a forest fire early detection system (1).

16. Method for implementing ML data in a system (1) for Early forest fire detection according to claim 15, characterized in that the newly determined ML data are transmitted to the terminal devices (ED) via a wireless network.

17. Method for implementing ML data in a system (1) for early forest fire detection according to one or more of the preceding claims, characterized in that the tn measurement data are obtained from true-negative events, wherein the tn measurement data are obtained from a laboratory test and / or in a natural environment.

18. Method for implementing ML data in a system (1) for Early forest fire detection according to claim 17, characterized in that the tn measurement data are obtained from the natural environment for which the use of the early forest fire detection system (1) is intended.

19. Method for implementing ML data in a system (1) for Early forest fire detection according to one or more of the preceding claims, characterized in that tp measurement data are additionally recorded, from which tp-ML data are determined.

20. Method for implementing ML data in a system (1) for Early forest fire detection according to claim 19, characterized in that the tp measurement data are recorded from true-positive events.

21. Method for implementing ML data in a system (1) for Early forest fire detection according to claim 20, characterized in that the true-positive events are simulated in the laboratory.

22. Method for implementing ML data in a system (1) for Early forest fire detection according to one or more of claims 17 to 21, characterized in that tp-ML data and tn-ML data are combined to form ML data.

23. Method for implementing ML data in a system (1) for Early forest fire detection according to claim 22, characterized in that the result data are determined from the measurement data and the ML data.

24. Method for implementing ML data in a system (1) for Early forest fire detection according to one or more of claims 17 to 23, characterized in that the tn measurement data are obtained from different regions of the natural environment for which the forest fire early warning system (1) is intended to be used.

25. Method for implementing ML data in a system (1) for early detection of forest fires according to claim 24, characterized in that different tn-ML data are created from the tn measurement data obtained from different regions of the natural environment.

26. Method for implementing ML data in a system (1) for early forest fire detection according to one or more of the preceding claims, characterized in that different regions of the forest (W) to be monitored are assigned to different tn-ML data.

27. Method for implementing ML data in a system (1) for early detection of forest fires according to claim 26, characterized in that, in order to determine the result data, the measured values ​​recorded from a specific region of the forest (W) to be monitored are compared with the tn-ML data assigned to this region.

28. Method for implementing ML data in a system (1) for early forest fire detection according to one or more of the preceding claims, characterized in that Measurement data are collected by the forest fire early warning system (1), where result data (RDnn) are determined by applying the tn-ML data to the acquired measurement data.

29. Forest fire early detection system (1) with a LoRaWAN network (10) comprising • a terminal device (ED), wherein the terminal device (ED) comprises a sensor device, a first control device, an evaluation device for evaluating measurement signals supplied by the sensor device and a power supply device (E), • a network server (NS), characterized in that the first control device is suitable and intended to access a memory comprising data from the adaptation and application of a machine learning model, wherein the machine learning model comprises tn-ML data.

30. Forest fire early warning system (1) with a LoRaWAN network (10) according to Claim 29, characterized in that the machine learning model comprises tp-ML data.

31. Forest fire early warning system (1) with a LoRaWAN network (10) according to Claim 29 or 30, characterized in that the memory is part of the terminal (ED).

32. Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 31, characterized in that the network server (NS) is coupled to a second control device that is suitable and intended to execute a machine learning program. Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 32, characterized in that the second control device has access to the measurement signals detected by the terminal device (ED). Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 33, characterized in that the second control device is connected to the terminal device (ED) via two different networks. Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 34, characterized in that the terminal device (ED) has a humidity sensor (S2.1, S2.2) for detecting the air humidity.Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 35, characterized in that the terminal (ED) has a temperature sensor (S1.1, S1.2) for detecting the. ambient temperature. Forest fire early detection system (1) with a LoRaWAN network (10) according to one or more of claims 29 to 36, characterized in that the terminal device (ED) has a pressure sensor (S3.1, S3.2) for detecting the air pressure.

Images