A sensor system for measuring moisture content in combustibles

Abstract

The present invention relates to a sensor system for measuring the moisture content of combustible materials, comprising: a sensing detection unit for detecting the reflection frequency of an object whose moisture content is to be measured and the reflection frequency of the soil in the environment in which the object whose moisture content is to be measured is located; a main control unit for controlling the sensing detection unit and transmitting detection data detected by the sensing detection unit; and a calculation unit for receiving the detection data, performing temperature compensation on the detection data, and performing environmental compensation processing on the temperature-compensated detection data in combination with meteorological factors to obtain the final moisture content of the object. The present invention is capable of long-term, unattended monitoring of the moisture content of combustible materials.

Description

Technical Field

[0001] The present invention relates to the technical field of forest fire early warning, in particular to a combustible moisture content measurement sensor system. Background Art

[0002] Forests are a crucial ecosystem worldwide. They not only provide immense social value by improving human living environments and providing resources and habitats, but also offer immense natural value in protecting species, preventing wind and sand storms, modifying climate, and conserving water resources. Known as the "lungs of the Earth," forest fires can cause significant damage and loss to both ecosystems and humans. Therefore, preventing, detecting, and suppressing forest fires remains a crucial and arduous task.

[0003] Forest fire warning and monitoring are two key components of forest fire prevention. Pre-disaster prediction and on-site monitoring are crucial for protecting forest resources. Traditional forest fire prevention methods primarily rely on building watchtowers for high-altitude observation, ground patrols, and video surveillance. However, the vast majority of forests are located in remote areas with extremely poor living conditions, complex terrain, and steep mountains, making comprehensive observation and patrol difficult. These methods also require significant manual effort, resulting in high underreporting rates and significant errors. In contrast, aerial patrols by aircraft, while capable of more comprehensive inspections and eliminating blind spots, are expensive, labor-intensive, and subject to underreporting. With the rapid development of advanced technologies such as computers and wireless sensors, the research and design of real-time forest fire monitoring systems based on wireless multi-sensor networks has become a hot topic. Unmanned wireless sensor systems can be freely deployed within the monitored forest area, enabling convenient and rapid real-time monitoring of various fire factors over a wide area. These systems provide information such as temperature and humidity, wind speed and direction, and moisture content for fire prediction models, and are therefore becoming a mainstream approach in forest fire research.

[0004] As for forest and grassland fire prediction and forecasting technologies, existing models are generally based on the well-known fire triangle model, which uses climate, topography, and fuel as three key indicators for assessing forest fire risk. While climate and topography information is relatively easy to obtain, obtaining accurate information on the fuel content of the vegetation canopy becomes crucial for assessing wildfire risk. Changes in fuel moisture content, also known as fuel humidity, significantly impact fire occurrence and spread by affecting combustion characteristics such as ignition point. Furthermore, the cooling effect of moisture reduces the inherent heat content of the fuel, requiring more external heat to maintain combustion. Therefore, fuel moisture content can be used to predict fire risk levels, estimate the extent of fire occurrence and spread, and predict the energy released by a fire and the area burned. Accurate, real-time acquisition and monitoring of fuel moisture content data not only provides a reference for fire suppression but also allows for the determination of forest fire management strategies, such as planned burning, based on fuel moisture content. However, the existing detection system for the moisture content of forest understory fuels still has defects such as insufficient accuracy, low temporal and spatial resolution, and difficulty in equipment deployment, making it impossible to obtain accurate and real-time fuel moisture content data.

[0005] Existing fuel moisture detection schemes can be categorized into two main types. One is an indirect detection method based on inversion of high-resolution remote sensing imagery. This method exploits the sensitivity of vegetation canopy moisture content in the near-infrared and shortwave infrared bands. After training a statistical learning model, it can rapidly and automatically generate spatiotemporally continuous fuel moisture estimates over a wide range of areas. While this approach has some general applicability, it also has significant drawbacks. First, its temporal resolution depends on the temporal resolution of remote sensing image acquisition. For example, MODIS only acquires remote sensing images of a specific area every one to two days. Consequently, the resulting moisture content estimates have a significant lag, making them inadequate for emergency situations. Second, accurate inversion relies heavily on measured hyperspectral data, which is limited and expensive, making it difficult to use across large areas and multiple temporally. Furthermore, this method struggles to accurately invert the moisture content of understory fuels in dense forests. These shortcomings significantly limit the application of inversion models in real-world fire risk prediction and early warning, as well as fire spread. Another direct detection method is based on the electrical resistance method. The basic idea of ​​this method is to monitor the changes in the electrical resistance of a small piece of wood exposed to air, and then derive its moisture content after temperature correction, thereby reflecting the fire risk of fallen branches in the forest. This sensor, called a hygrometer or hygrometer stick, can be easily deployed in forest areas. Compared with remote sensing inversion methods, its measurement results are more representative of the moisture content of thin branches in the forest environment. However, this method also has some unavoidable drawbacks. First, the thin wooden sticks used in this type of method must be exposed to the air for a long time. Therefore, in order to avoid errors in the measured values ​​due to corrosion of the sticks, the sticks need to be replaced regularly (often at intervals of less than one year), making long-term unattended operation difficult to achieve, which greatly increases operating costs and limits the deployment area to a certain extent. At the same time, the combustible materials accumulated under the forest are often a mixture of leaves and dead branches. It is difficult to simulate such a complex combustible composition by only measuring thin wooden sticks in the same environment, which causes errors and affects the quality of predictions. Finally, the time lag of the wooden sticks themselves to environmental changes is also greater than that of the mixed combustible materials of fallen leaves and dead branches under the forest, which reduces the reference value of their measured values. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a combustible moisture content measurement sensor system, which can accurately monitor the moisture content of combustibles without being monitored for a long time.

[0007] The technical solution adopted by the present invention to solve the technical problem is to provide a combustible moisture content measurement sensor system, comprising:

[0008] A sensing detection unit, configured to detect a reflection frequency of an object whose moisture content is to be measured and a reflection frequency of soil in an environment where the object whose moisture content is to be measured is located;

[0009] A main control unit, configured to control the sensing detection unit and transmit detection data detected by the sensing detection unit;

[0010] The calculation unit is used to receive the detection data, perform temperature compensation on the detection data, and perform environmental compensation processing on the temperature-compensated detection data in combination with meteorological factors to obtain the final moisture content of the object.

[0011] The sensing detection unit includes an insertion rod, on which a first detector and a second detector are distributed. The insertion rod is inserted into the soil in the environment where the object whose moisture content is to be measured is located, so that the first detector is located in the environmental soil layer and the second detector is located in the object whose moisture content is to be measured; a signal generator is also provided on the insertion rod, and the signal generator is used to send electromagnetic wave signals of different frequencies to the object whose moisture content is to be measured and the soil in the environment where the object whose moisture content is to be measured is located. The first detector is used to receive the electromagnetic wave signal passing through the environmental soil layer and extract the reflection frequency of the soil in the environment where the object whose moisture content is to be measured is located therefrom; the second detector is used to receive the electromagnetic wave signal passing through the object whose moisture content is to be measured and extract the reflection frequency of the object whose moisture content is to be measured therefrom.

[0012] The first detector and the second detector have the same structure, and both include an anode ring and a cathode ring, and the anode rings and cathode rings are alternately and vertically distributed on the insertion rod.

[0013] An isolation cavity is further provided between the first detector and the second detector. The diameter of the isolation cavity is larger than that of the first detector and the second detector, and the isolation cavity is made of plastic material.

[0014] The calculation unit includes a temperature compensation module and an environmental compensation module; the temperature compensation module performs temperature compensation on the detection data based on a sensor temperature characteristic curve model; the environmental compensation module inputs the temperature-compensated detection data and current meteorological factors into the environmental compensation model to obtain the final moisture content of the object.

[0015] The sensor temperature characteristic curve model is established through experiments, wherein during the heating process, the sensor temperature characteristic curve model is y=-8.8846x+b, and during the cooling process, the sensor temperature characteristic curve model is y=-8.8791x+b, where y represents the data after temperature compensation, x represents the detection data detected by the sensing detection unit, and b is a constant.

[0016] The environmental compensation model includes a feature extractor and a linear classifier; the feature extractor includes a long-term feature extraction network and a short-term feature extraction network, the input of the long-term feature extraction network is an ordered sequence constructed according to daily values, which is used to extract meteorological factor data features in units of days; the input of the short-term feature extraction network is an ordered sequence constructed according to hourly values, which is used to extract meteorological factor data features in units of hours; the long-term and short-term feature extraction networks are long short-term memory neural networks with two independent parameters; the linear classifier is a fully connected network, the input is the spliced ​​meteorological factor data features in units of days and the meteorological factor data features in units of hours, and the output is a category.

[0017] The environmental compensation model is trained in the following way:

[0018] Data cleaning and preprocessing: remove data with combustible moisture content exceeding the threshold, and normalize the removed data to a mean of zero and a variance of one;

[0019] Sample composition: The data is constructed into an ordered sequence of length n according to the daily and hourly values, and arranged in chronological order. The sample labels use the absolute value error between the sensor output and the actual moisture content, and are graded.

[0020] Model training: Use the gradient descent method to train the parameters of the environmental compensation model, input the data-enhanced samples into the environmental compensation model batch by batch to obtain sample classification, input the classification results into the loss function to obtain the loss value for error backpropagation and parameter update; the data enhancement refers to adding a random fluctuation to the input sample; the loss function includes cross entropy loss, mean loss and variance loss.

[0021] The meteorological factors include air temperature, relative humidity, rainfall, wind speed, illumination and air pressure.

[0022] Beneficial effects

[0023] Due to the adoption of the above-mentioned technical solution, the present invention offers the following advantages and positive effects compared to existing technologies: It is suitable for long-term monitoring of the moisture content of forest floor combustibles. The sensor detection unit adopts a contactless design, which reduces maintenance and facilitates unattended deployment. The calculation unit utilizes a neural network algorithm to model environmental parameters, effectively improving measurement accuracy in complex environments. The entire sensor circuit is simple, with low overall cost, making it highly feasible for large-scale deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is an overall block diagram of an embodiment of the present invention;

[0025] Figure 2 is a schematic diagram of a sensing detection unit in an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram of the detector structure in an embodiment of the present invention;

[0027] Figure 4 It is a structural block diagram of the environmental compensation model in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0029] The embodiment of the present invention relates to a combustible moisture content measurement sensor system, such as Figure 1 As shown, it includes: a sensing detection unit, which is used to detect the reflection frequency of the object whose moisture content is to be measured and the reflection frequency of the soil in the environment where the object whose moisture content is to be measured is located; a main control unit, which is used to control the sensing detection unit and send the detection data detected by the sensing detection unit; and a calculation unit, which is used to receive the detection data, perform temperature compensation on the detection data, and perform environmental compensation processing on the temperature-compensated detection data in combination with meteorological factors to obtain the final moisture content of the object.

[0030] This embodiment detects and measures the moisture content of surface litter. The practical application of the sensor detection unit is limited by the layout conditions and the on-site environment and must be operable on-site. Therefore, this application needs to consider the following aspects of the design: first, the surface litter of forests and grasslands has a certain thickness, and the detector must be able to measure the moisture content of fallen objects in an area of ​​a certain thickness; at the same time, the sensor detection unit should not affect the evaporation of water under sunlight in the natural environment, and the infiltration of rainwater after rainfall, and it is necessary to avoid or reduce the impact of the soil on the detector.

[0031] Therefore, Figure 2As shown, the sensing detection unit in this embodiment includes an insertion rod 1, on which a first detector 2 and a second detector 3 are distributed. The insertion rod 1 is inserted into the soil of the environment where the object whose moisture content is to be measured is located, so that the first detector 2 is located in the environmental soil layer 5 and the second detector 3 is located in the object whose moisture content is to be measured 6. The insertion rod 1 is also provided with a signal generator 4, which is used to send electromagnetic wave signals of different frequencies to the object whose moisture content is to be measured and the soil in the environment where the object whose moisture content is to be measured is located. The first detector 2 is used to receive the electromagnetic wave signal passing through the environmental soil layer and extract the reflection frequency of the soil in the environment where the object whose moisture content is to be measured is located. The second detector 3 is used to receive the electromagnetic wave signal passing through the object whose moisture content is to be measured and extract the reflection frequency of the object whose moisture content is to be measured. There is also an isolation cavity 7 between the first detector 2 and the second detector 3.

[0032] In this embodiment, the first detector and the second detector are placed in the environmental soil layer and the fallen combustible layer respectively. Changes in moisture content will cause the detectors to sense different frequencies, so the moisture content of the combustible can be obtained by detecting the frequency.

[0033] When measuring the moisture content of surface litter, the detector is easily affected by the soil below. Therefore, the detection method of the detector has an important impact on the measurement accuracy of the moisture content of surface litter. In this embodiment, the detector adopts a vertical layered ring design, such as Figure 3 As shown, two sets of circular rings are used, with cathode rings B and anode rings A placed alternately and vertically. This design is easy to implement with an insert rod and enables layered detection. The electric field distribution pattern also reduces soil interference to a certain extent. Furthermore, the presence of an isolation cavity between the two detectors can further reduce the impact of soil on the measured values. The diameter of the isolation cavity needs to be larger than that of the detector. Simulations show a diameter of 250 cm and a thickness of 60 cm. A plastic material with a relative dielectric constant of 3 is selected to ensure that the electric field distribution of metal detectors is not affected.

[0034] It can be seen that by arranging the fallen objects whose moisture content is to be measured around the sensing detection unit, the moisture in the fallen objects affects the electric field around the sensing detection unit, thereby affecting the capacitance value of the detector, and further affecting the circuit resonant frequency of the sensing detection unit. The moisture content value of the fallen objects with measured moisture content is obtained through the change in frequency. In this embodiment, a 100MHz active crystal oscillator is selected as the signal generator to achieve stable measurement accuracy. Since the high frequency of the crystal oscillator is not conducive to the frequency detection of the microprocessor, this embodiment can divide the circuit output frequency. At the same time, considering the attenuation of the transmission line, the output signal of the detector can be amplified.

[0035] The main control unit is responsible for operating and managing the system, sending detection commands to the sensor detection unit and reading the frequency of its current output signal. It then transmits this information and detection data to the computing unit via the data transmission module. In this embodiment, the computing unit can be a cloud-based monitoring platform, and the data transmission module can be a 4G module. The main control unit in this embodiment can also receive commands and parameters from the cloud-based monitoring platform to execute corresponding tasks.

[0036] The calculation unit is mainly responsible for processing the signal frequency reported by the main control unit and finally obtaining the moisture content value of the combustible material. It is divided into two stages. The first stage is to perform temperature calibration on the reported signal frequency value and use the mapping relationship obtained from the experiment (i.e., the temperature compensation model) to obtain the moisture content of the combustible material under an ideal interference-free environment; the second stage is to compensate for the environmental factors by integrating multiple meteorological factors to obtain the final moisture content of the combustible material that can better reflect the actual situation.

[0037] The calculation unit of this embodiment includes a temperature compensation module and an environmental compensation module; the temperature compensation module performs temperature compensation on the detection data based on the sensor temperature characteristic curve model; the environmental compensation module inputs the temperature-compensated detection data and the current meteorological factors into the environmental compensation model to obtain the final moisture content of the object.

[0038] Affected by the physical properties of the sensing detection unit, the impedance, capacitive reactance and inductive reactance of the sensing detection unit will change to a certain extent at different temperatures, thereby affecting the detection accuracy of the sensing detection unit. This embodiment establishes a sensor temperature characteristic curve model through an experimental method to perform temperature compensation on the measured value of the sensing detection unit. Due to the hysteresis effect caused by the physical properties of the material device, the temperature response curve of the heating process and the temperature response curve of the cooling process do not overlap. Therefore, the temperature response characteristics of the heating and cooling processes of the sensing detection unit are tested, and it can be found that it has good linearity. Therefore, a linear function y=kx+b is used for fitting, and the slope k of the heating process is obtained experimentally. 升 =-8.8846, the slope k of the cooling process 降 =-8.8791, and the intercept b depends on the slight difference of each sensor detection unit and can be calibrated before leaving the factory.

[0039] Although the sensing detection unit itself has achieved high measurement accuracy for fallen objects and soil moisture content in a steady-state environment after linear temperature correction, in actual more complex natural environments, the influence of meteorological factors such as light, rainfall, and wind speed on the moisture content of fallen objects and the sensing detection unit will bring certain challenges to the measurement accuracy. Therefore, this embodiment also establishes an environmental compensation model based on a neural network to ensure that the reported data is stable and accurate.

[0040] The input of the environmental compensation model is shown in Table 1, which includes multiple meteorological factors. The output is the absolute error between the output of the sensor detection unit and the true moisture content. Among them, the true value label in the model training is the moisture content of the litter measured by the drying method. Since meteorological factors such as temperature and humidity are periodic on a daily and annual basis, additional time parameters are added for modeling. In Table 1, the daily value data refers to the daily maximum / minimum / average value of this factor in the period before the sampling time point, and the hourly value refers to the maximum / minimum / average value per hour. The starting time for daily value calculation is 0:00 Beijing time.

[0041] Table 1 Meteorological factors used in model training

[0042]

[0043] The training of the environmental compensation model goes through the following steps:

[0044] (1) Data cleaning: Considering that when the moisture content of combustibles is higher than 45%, they are almost incombustible, so this part of the data is first eliminated; for the measured value samples, a single multi-point sampling method can be used to take the average value to reduce the sampling error.

[0045] (2) Data preprocessing: normalize to a mean of 0 and a variance of 1.

[0046] (3) Sample composition: For the data part, an ordered sequence of length n is constructed for the daily value and the hourly value, where n is an adjustable hyperparameter; the sequences are arranged in chronological order; the sample labels are obtained by using the absolute value error between the sensor output and the true moisture content, and are classified according to Table 2, making it a classification problem.

[0047] Table 2 Model classifier output value mapping table

[0048]

[0049]

[0050] (4) Model training: Use the gradient descent method to train the environmental compensation model; input the data-enhanced samples into the environmental compensation model in batches to extract sample features, and the sample features enter the linear classifier to obtain sample classification. The classification results are input into the loss function to obtain the loss value for error backpropagation and parameter update.

[0051] (5) Data enhancement: Adding a small random fluctuation to the input sample value can alleviate overfitting and improve the generalization ability of the model.

[0052] (6) Loss function: It consists of three parts: cross entropy loss, mean loss, and variance loss. Cross entropy loss is a common loss for classification tasks, guiding the model to output the correct classification; mean loss is a regular term in the loss function, guiding the model's output (weighted average) to be close to the true value label, thereby improving model performance. The specific formula is as follows:

[0053]

[0054]

[0055] L=λ·L mean +(1-λ)·L crossEntropy

[0056] in, is the weighted average of the classification, j is the category estimate, p i is the soft cross entropy category coefficient, is the normalized model output, indicating the probability of classification; L is the loss function, N is the number of samples, and λ is an adjustable hyperparameter.

[0057] The environmental compensation model in this embodiment is as follows Figure 4 As shown in the figure, the feature extractor consists of two long-short-term memory recurrent neural networks, one for extracting daily and one for extracting hourly meteorological factor data features. This is because daily and hourly environmental factors have different impacts on moisture content, and extracting them separately yields better results. The linear classifier is a fully connected network with trainable parameters, whose input dimension is the number of features and output dimension is the number of categories. It ultimately outputs the absolute error of the prediction through a weighted average.

[0058] The present invention will be further described below by taking a combustible moisture content measurement sensor system deployed somewhere in Jiading, Shanghai as an example.

[0059] In order to verify the accuracy and stability of the measurement of this system, the embodiment analyzed the output data of this sensing system in March 2022 and compared it with the moisture content measured by the drying method at the same time point.

[0060] The hyperparameters of the environmental compensation model are as follows: a) The sample batch size is 8, the initial learning rate is set to 0.01, and the Adagrad optimization algorithm is used for 100 epochs, with the learning rate decayed to 0.001 at epoch 80. b) The two LSTM networks use the same network structure, with a hidden layer feature length of 128, and features extracted by stacking three LSTM modules. The linear classifier has 128 neurons. c) The length of the data sequence is set to 7 (to account for meteorological factors of the previous 7 days / hours). d) A small random fluctuation is added to the samples to mitigate overfitting, with a maximum random amplitude of 5%. e) The loss function coefficient λ is adjusted during training using linear annealing, starting with λ = 0.1 and ending with λ = 0.8, with an adjustment interval of every 30 iterations.

[0061] The results are shown in Table 3 below. Among the 30 samples, after compensation using the environmental compensation model, only two samples, or 6.67%, had an error greater than 2.5. The average absolute error for all samples was 0.918, and all errors were less than 5, with a maximum error of 4.43. This demonstrates that the combustible moisture content sensing system of this embodiment has excellent environmental adaptability and measurement accuracy.

[0062] Table 3 Example sampling data

[0063]

[0064]

[0065] It's easy to see that this invention is suitable for long-term monitoring of the moisture content of forest floor combustibles. The sensor detection unit utilizes a contactless design, reducing maintenance and facilitating unattended deployment. The calculation unit utilizes a neural network algorithm to model environmental parameters, effectively improving measurement accuracy in complex environments. The entire sensor circuit is simple, resulting in low overall cost and strong feasibility for large-scale deployment.

Claims

1. A combustible moisture content measurement sensor system, characterized in that: include: A sensing detection unit, configured to detect a reflection frequency of an object whose moisture content is to be measured and a reflection frequency of soil in an environment where the object whose moisture content is to be measured is located; A main control unit, configured to control the sensing detection unit and transmit detection data detected by the sensing detection unit; a calculation unit, configured to receive the detection data, perform temperature compensation on the detection data, and perform environmental compensation processing on the temperature-compensated detection data in combination with meteorological factors to obtain a final moisture content of the object; The calculation unit includes a temperature compensation module and an environmental compensation module; the temperature compensation module performs temperature compensation on the detection data based on the sensor temperature characteristic curve model; the environmental compensation module inputs the temperature-compensated detection data and the current meteorological factors into the environmental compensation model to obtain the final moisture content of the object; the sensor temperature characteristic curve model is established experimentally, wherein, during the heating process, the sensor temperature characteristic curve model is y=-8.8846x+b, and during the cooling process, the sensor temperature characteristic curve model is y=-8.8791x+b, y represents the temperature-compensated data, x represents the detection data detected by the sensing detection unit, and b is a constant.

2. The combustible moisture content measurement sensor system according to claim 1, characterized in that: The sensing detection unit includes an insertion rod, on which a first detector and a second detector are distributed. The insertion rod is inserted into the soil in the environment where the object whose moisture content is to be measured is located, so that the first detector is located in the environmental soil layer and the second detector is located in the object whose moisture content is to be measured; a signal generator is also provided on the insertion rod, and the signal generator is used to send electromagnetic wave signals of different frequencies to the object whose moisture content is to be measured and the soil in the environment where the object whose moisture content is to be measured is located. The first detector is used to receive the electromagnetic wave signal passing through the environmental soil layer and extract the reflection frequency of the soil in the environment where the object whose moisture content is to be measured is located therefrom; the second detector is used to receive the electromagnetic wave signal passing through the object whose moisture content is to be measured and extract the reflection frequency of the object whose moisture content is to be measured therefrom.

3. The combustible moisture content measurement sensor system according to claim 2, characterized in that: The first detector and the second detector have the same structure, and both include an anode ring and a cathode ring, and the anode rings and cathode rings are alternately and vertically distributed on the insertion rod.

4. The combustible moisture content measurement sensor system according to claim 2, characterized in that: An isolation cavity is further provided between the first detector and the second detector. The diameter of the isolation cavity is larger than that of the first detector and the second detector, and the isolation cavity is made of plastic material.

5. The combustible moisture content measurement sensor system according to claim 1, characterized in that: The environmental compensation model includes a feature extractor and a linear classifier; the feature extractor includes a long-time feature extraction network and a short-time feature extraction network, the input of the long-time feature extraction network is an ordered sequence constructed according to daily values, which is used to extract meteorological factor data features in units of days; the input of the short-time feature extraction network is an ordered sequence constructed according to hourly values, which is used to extract meteorological factor data features in units of hours; the long-time and short-time feature extraction networks are long-short-time memory recursive neural networks with two independent parameters; the linear classifier is a fully connected network, the input is the spliced ​​meteorological factor data features in units of days and the meteorological factor data features in units of hours, and the output is a category.

6. The combustible moisture content measurement sensor system according to claim 1, characterized in that: The environmental compensation model is trained in the following way: Data cleaning and preprocessing: remove data with combustible moisture content exceeding the threshold, and normalize the removed data to a mean of zero and a variance of one; Sample composition: construct an ordered sequence of length n according to the daily value and hourly value respectively, and arrange them in chronological order; the sample label uses the absolute value error between the sensor output and the actual moisture content, and performs hierarchical processing; model training: use the gradient descent method to train the parameters of the environmental compensation model, input the data-enhanced samples into the environmental compensation model in batches to obtain sample classification, input the classification results into the loss function to obtain the loss value, perform error backpropagation and update the parameters; the data enhancement refers to adding a random fluctuation to the input sample; the loss function includes cross entropy loss, mean loss and variance loss.

7. The combustible moisture content measurement sensor system according to claim 1, characterized in that: The meteorological factors include air temperature, relative humidity, rainfall, wind speed, illumination and air pressure.

Images