Method for detecting moisture content of combustible materials in forest and grassland

Abstract

The invention relates to a forest steppe combustible moisture content detection method, and relates to the field of combustible moisture content detection.The forest steppe combustible moisture content detection method comprises the steps that a combustible sample to be detected is weighed and then put into a container, and the container is put into a high-low temperature test box; setting the temperature and humidity of a high-low temperature test box and the wind speed of an airflow generator; acquiring atmospheric humidity, atmospheric temperature and wind speed for recording weather elements; adding quantitative water into the container, and recording the volume of the added water and the hygrometer value of the combustible; dividing the collected data set into a training set, a verification set and a test set; calculating input parameters by using a neural network, and adjusting the input parameters to receive atmospheric temperature, humidity, wind speed and combustible humidity parameters as input; a training set is used for training a model, hyper-parameters of the model are adjusted through a verification set, an optimization algorithm is utilized, network weight is updated, an error between a predicted value and the actual moisture content is minimized, and the problem that the moisture content of the combustible is not accurately measured is solved.

Description

Technical Field

[0001] This invention relates to the field of moisture content detection of combustibles, and specifically to a method for detecting the moisture content of combustibles in forests and grasslands. Background Technology

[0002] With the increasing impact of global climate change and human activities on the natural environment, the moisture content of combustibles has become a crucial factor affecting forest fires, water resource management, and the ecological environment. Accurate monitoring of combustible moisture content is of great significance for improving forest fire prevention, ensuring food security, and maintaining ecological balance. Traditional methods for measuring combustible moisture content, such as capacitive sensor methods, infrared sensor methods, and drying methods, while meeting measurement needs to some extent, suffer from limitations such as cumbersome operation, high cost, and poor real-time performance. Meanwhile, weather factors, such as air temperature, air humidity, and wind speed, are closely related to combustible moisture content. Therefore, researching a method for monitoring combustible moisture content that combines weather element information and combustible moisture data has significant theoretical and practical implications.

[0003] Existing methods for measuring the moisture content of combustibles have some shortcomings: 1. They do not consider the influence of actual ambient temperature and humidity on the moisture content of combustibles. [1-2] 2. Although actual ambient temperature and humidity are considered, data on the combustible material itself is not incorporated, resulting in low measurement accuracy. [3] 3. While some methods offer high accuracy in measuring the moisture content of combustibles, they cannot provide real-time, continuous monitoring of the moisture content. [4] 4. Some methods, while capable of continuously measuring the moisture content of combustibles, do not consider outdoor environmental factors and are therefore not widely applicable. [5-6] .

[0004] [1] Tong Guanjun. A device and method for measuring the moisture content of combustibles [P]. Shanghai: CN106841563A, 2017.03.03.

[0005] [2] Gao Zhongliang, Shu Lifu, Li Zhi, Long Tengteng, Wang Qiuhua, Zhou Ruliang, Diao Jianpeng, Cheng Cheng, Liu Dewen, Wang Hongrun, Cui Fei, Wei Jianheng, Yang Zhuwei, Ma Zenan, Wang Hechenyang, et al. A method and system for determining the moisture content of surface combustibles in Pine yunnanensis [P]. Yunnan Province: CN112213230A, 2021.01.12.

[0006] [3] He Binbin, Fan Chunquan, Quan Xingwen, et al. A method for estimating the 10-h dead combustible content of forest surface [P]. Sichuan Province: CN112462028A, 2020.12.15.

[0007] [4] Jin Sen, Yu Hongzhou, et al. Automatic lifting weighing measurement method and sampling weighing device for surface combustible moisture content [P]. Heilongjiang Province: CN103487350A, 2013.10.22.

[0008] [5] Guo Zaijun, Ye Qiaolin, Wang Haijiao, Gao Demin, Li Yuntao, Niu Haifeng, Wang Runsheng, et al. A device for detecting the moisture content of forest combustibles [P]. Hebei Province: CN113252499A, 2021.05.19.

[0009] [6] Hu Haiqing, Guo Yan, Li Donghui, Man Ziyuan, Yu Hongzhou, Hu Tongxin, Sun Long, et al. A device for measuring the moisture content of forest combustibles in the field [P]. Heilongjiang Province: CN114199716A, 2021.12.11. Summary of the Invention

[0010] A method for detecting the moisture content of forest and grassland combustibles utilizes weather element sensors and combustible humidity sensors to collect data on atmospheric temperature, atmospheric humidity, wind speed, and combustible humidity. These factors exhibit a non-linear relationship with the measured moisture content, thus solving the problem of inaccurate measurement. The method employs a neural network for non-linear fitting to derive the moisture content, demonstrating strong environmental adaptability and enabling more accurate measurement of moisture content in field combustibles.

[0011] A method for detecting the moisture content of forest and grassland combustibles includes the following steps:

[0012] S1. Weigh the combustible sample to be tested, put it into a container, and place it in a high and low temperature test chamber.

[0013] S2. Set the temperature of the high and low temperature test chamber from -40℃ to +60℃, and conduct a test every 10℃.

[0014] Humidity, from 20%RH to +90°C, tested every 10%RH;

[0015] The wind speed of the airflow generator was tested every 3 m / s, from 0 m / s to 21 m / s.

[0016] Acquire and record atmospheric humidity, atmospheric temperature, and wind speed every 30 minutes;

[0017] S3. Add a fixed amount of water to the container and record the volume of water added and the hygrometer reading of the combustible material.

[0018] S4. Divide the collected dataset into training set, validation set, and test set; the training set is used to build the model; the validation set is used for model parameter tuning and optimization; and the test set is used to evaluate the final performance of the model; usually, the ratio of the dataset to the test set and validation set is 8:1:1.

[0019] S5. Use a neural network to calculate the input parameters and adjust them to receive atmospheric temperature, humidity, wind speed, and combustible material humidity parameters as input.

[0020] S6. Train the model using the training set and adjust the model's hyperparameters using the validation set to avoid overfitting and underfitting; use optimization algorithms to update network weights and minimize the error between the predicted value and the actual moisture content.

[0021] S7. In practical applications, input the untrained atmospheric temperature, humidity, wind speed, and combustible humidity parameters into the trained deep learning model to obtain the predicted moisture content of combustibles.

[0022] Furthermore, the neural network includes:

[0023] The input layer has a size of (b_s, 4), where b_s represents the number of samples trained simultaneously in each training round, and 4 represents that each sample contains 4 parameters, including atmospheric temperature, humidity, wind speed, and combustible material humidity.

[0024] In a linear layer, an input array of size (b_s, 1, 4) is multiplied by the layer's weights to obtain an output of size (b_s, 1, 512), where 512 is a value set during model construction, as are 1024 and 128. The output sequence is then transformed into three vectors: query (Q), key (K), and value (V). The calculation of Q, K, and V allows the model to capture the dependencies between any two elements in the input sequence, regardless of their distance within the sequence. The key vector represents the vector representation of all elements in the sequence and is used for comparison with the query vector. Similarly, the value vector also represents the vector representation of all elements in the sequence. Finally, a weighted sum is generated to produce the output for the next fully connected layer.

[0025] The attention layer generates a weighted context vector by calculating the importance weights of different parts of the input data. This vector emphasizes the parts of the input data that are most relevant to the current task. The attention layer is applied after the fully connected layer, and the context is then fed into another fully connected layer. Through this structure, the model can effectively capture key information in the input data and make more accurate predictions or classifications in subsequent fully connected layers.

[0026] The fully connected layer performs activation function correction and dropout operations in every two adjacent layers. The activation function can add non-linear characteristics to the model, enabling the model to better fit various complex curves. At the same time, the dropout method sets the parameters in the input array to 0 according to a certain proportion, which means that the parameters in this part will no longer participate in the subsequent calculations. This can greatly improve the model's anti-interference ability.

[0027] Beneficial Effects: Based on weather elements such as atmospheric temperature, humidity, wind speed, and combustible material moisture, this model accurately measures the moisture content of combustible materials using a neural network model. The model combines a fully connected layer and an attention layer. The fully connected layer captures all preceding features and effectively fuses them to form a global feature representation, integrating feature information and making the final decision. The attention layer dynamically adjusts its focus on the input data, enhancing the model's ability to capture key information and enabling it to adaptively select important features, reducing the need for human intervention. This structure allows the model to capture both global and local features. Through its unique feature fusion and dynamic weighting mechanism, it effectively combines atmospheric temperature, humidity, and other elements, accurately inferring the moisture content of litter, demonstrating powerful representation capabilities and predictive accuracy. This invention is characterized by its comprehensive approach. The method integrates multiple meteorological factors, such as atmospheric temperature and humidity, and wind speed, as well as their correlation with litter moisture content. This allows for a comprehensive and accurate reflection of the moisture status of the forest ecosystem. Furthermore, it boasts a high degree of intelligence, overcoming the limitations of traditional methods by utilizing advanced sensors and data processing technologies. Previous studies often focused only on the impact of a single meteorological factor on litter moisture content, neglecting the interactions between multiple factors, leading to inaccurate and incomplete predictions. This method effectively overcomes this limitation by comprehensively considering multiple meteorological factors. This invention has a wider range of applications and provides more accurate outdoor measurements of combustible material moisture content. Attached Figure Description

[0028] Figure 1 This is a diagram of the overall system structure.

[0029] Figure 2 It is a neural network model;

[0030] Figure 3 This is a flowchart of the model optimization process;

[0031] Figure 4 This is Table 1 - Table of Humidity Changes in Fallen Items;

[0032] Figure 5 Table 2 shows the changes in atmospheric temperature.

[0033] Figure 6 Table 3 shows the changes in atmospheric humidity.

[0034] Figure 7 Table 4 shows the wind speed changes;

[0035] Figure 8 Table 5 shows the parameters of the BP neural network training model.

[0036] Figure 9 This is Table 6 - Results without environmental factors;

[0037] Figure 10 Table 7 shows the verification results with added environmental factors;

[0038] Figure 11 This is a visualization of model training. Detailed Implementation

[0039] This invention discloses a method for detecting the moisture content of forest and grassland combustibles, such as... Figure 1-11 The specific implementation steps are shown below:

[0040] Step 1: Weigh the combustible sample to be tested, put it into a container, and place it in a high and low temperature test chamber.

[0041] Step 2: Set the temperature of the high and low temperature test chamber. Set the temperature of the high and low temperature test chamber from -40℃ to +60℃, and conduct a test every 10℃.

[0042] Humidity, from 20%RH to +90°C, tested every 10%RH;

[0043] The wind speed of the airflow generator was tested every 3 m / s, from 0 m / s to 21 m / s.

[0044] Acquire and record atmospheric humidity, atmospheric temperature, and wind speed every 30 minutes.

[0045] Step 3: Add a measured amount of water to the container and record the volume of water added and the hygrometer reading of the combustible material.

[0046] Step 4: Divide the collected dataset into training, validation, and test sets. Tables 1 to 4 show randomly selected data samples. The training set is used to build the model, the validation set is used for model parameter tuning and optimization, and the test set is used to evaluate the final performance of the model. Typically, the ratio of the dataset to the test and validation sets is 8:1:1.

[0047] Step 5: See Figure 2The neural network model diagram shows an input layer with a size of (b_s, 4), where b_s represents the number of training samples per training epoch, and 4 indicates that each sample contains four parameters: atmospheric temperature, humidity, wind speed, and combustible material humidity. In the linear layer, the input array of size (b_s, 1, 4) is multiplied by the weights of that layer to obtain an output of size (b_s, 1, 512), where 512 is the value. The same applies to other parameters such as 1024 and 128. This number comes from the settings used when building the model. The output sequence is then converted into three vectors: query (Q), key (K), and value (V). The calculation of Q, K, and V allows the model to capture the dependency between any two elements in the input sequence, regardless of their distance within the sequence. The key vector, representing the vector representation of all elements in the sequence, is used for comparison with the query vector. Similarly, the value vector represents the vector representation of all elements in the sequence. These vectors are then weighted and summed to generate the output for the next fully connected layer. The attention layer generates a weighted context vector by calculating the importance weights of different parts of the input data. This vector emphasizes the parts of the input data most relevant to the current task. After the fully connected layer, the attention layer applies weights based on the importance of different parts. This context is then fed into another fully connected layer. Through this structure, the model can effectively capture key information from the input data and perform more accurate predictions or classifications in subsequent fully connected layers. This design not only improves model performance but also enables the model to better understand and process complex data patterns. Afterwards, activation function correction and dropout operations are performed between every two adjacent fully connected layers. Activation functions can add non-linearity to the model, allowing it to better fit various complex curves. Meanwhile, the dropout method we use sets a certain proportion of the parameters in the input array to 0, meaning that these parameters will not participate in subsequent calculations. This can significantly improve the model's robustness to interference. In our model, the dropout parameter is set to 0.05. After multiple linear layer operations, the model finally obtains a unique output, which is the predicted water content value.

[0048] Step 6: Train the model using the training set (training parameters are shown in Table 5), and adjust the model's hyperparameters using the validation set to avoid overfitting and underfitting. Use optimization algorithms, such as stochastic gradient descent (SGD), to update the network weights and minimize the error between the predicted and actual water content. Figure 3 process.

[0049] Step 7: In practical applications, input the untrained atmospheric temperature, humidity, wind speed, and combustible humidity parameters into the trained deep learning model to obtain the predicted moisture content of the combustible.

[0050] Based on the verification results without environmental factors in Table 6, the average error can be calculated as follows: By verifying the results using Table 7 and adding environmental factors, and calculating the average error, we can obtain the following: Under the same conditions without considering environmental factors, the average error in detecting the moisture content of combustibles is [missing value]. The average error in detecting the moisture content of combustibles using a model trained based on atmospheric temperature, humidity, wind speed, and combustible material humidity is: In comparison, the accuracy of the combustible material moisture content detection method based on BP neural network is significantly improved.

[0051] like Figure 11 The graph shows the validation results of the final output model. The top left corner shows the validation results on the training set, the bottom left corner shows the validation results on the test set, the top right corner shows the validation results on the validation set, and the bottom right corner shows the validation results on the entire dataset. In these four sub-graphs, the vertical axis represents the actual water content value, while the horizontal axis corresponds to the water content value output by the model. The "x" symbols in the graphs indicate the relative positional relationship between the model output value and the true value; these points are roughly distributed around the line y = x. Ideally, these "x" symbols should be completely above the line y = x. Ultimately, the overall model fit R reached 0.994.

[0052] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting the moisture content of forest and grassland combustibles, characterized in that, Includes the following steps: S1. Weigh the combustible sample to be tested, put it into a container, and place it in a high and low temperature test chamber. S2. Set the temperature of the high and low temperature test chamber from -40℃ to +60℃, and conduct a test every 10℃. Humidity, from 20%RH to +90°C, tested every 10%RH; The wind speed of the airflow generator was tested every 3 m / s, from 0 m / s to 21 m / s. Acquire and record atmospheric humidity, atmospheric temperature, and wind speed every 30 minutes; S3. Add a fixed amount of water to the container and record the volume of water added and the hygrometer reading of the combustible material. S4. Divide the collected dataset into training set, validation set, and test set; the training set is used to build the model; the validation set is used for model parameter tuning and optimization; and the test set is used to evaluate the final performance of the model; usually, the ratio of the dataset to the test set and validation set is 8:1:

1. S5. Use a neural network to calculate the input parameters and adjust them to receive atmospheric temperature, humidity, wind speed, and combustible material humidity parameters as input. S6. Train the model using the training set and adjust the model's hyperparameters using the validation set to avoid overfitting and underfitting; use optimization algorithms to update network weights and minimize the error between the predicted value and the actual moisture content. S7. In practical applications, input the untrained atmospheric temperature, humidity, wind speed, and combustible humidity parameters into the trained deep learning model to obtain the predicted moisture content of combustibles.

2. The method for detecting the moisture content of forest and grassland combustibles according to claim 1, characterized in that, The neural network includes: The input layer has a size of (b_s, 4), where b_s represents the number of samples trained simultaneously in each training round, and 4 represents that each sample contains 4 parameters, including atmospheric temperature, humidity, wind speed, and combustible material humidity. In a linear layer, an input array of size (b_s, 1, 4) is multiplied by the layer's weights to obtain an output of size (b_s, 1, 512), where 512 is a value set during model construction, as are 1024 and 128. The output sequence is then transformed into three vectors: query (Q), key (K), and value (V). The calculation of Q, K, and V allows the model to capture the dependencies between any two elements in the input sequence, regardless of their distance within the sequence. The key vector represents the vector representation of all elements in the sequence and is used for comparison with the query vector. Similarly, the value vector also represents the vector representation of all elements in the sequence. Finally, a weighted sum is generated to produce the output for the next fully connected layer. The attention layer generates a weighted context vector by calculating the importance weights of different parts of the input data. This vector emphasizes the parts of the input data that are most relevant to the current task. The attention layer is applied after the fully connected layer, and the context is then fed into another fully connected layer. Through this structure, the model can effectively capture key information in the input data and make more accurate predictions or classifications in subsequent fully connected layers. The fully connected layer performs activation function correction and dropout operations in every two adjacent layers. The activation function can add non-linear characteristics to the model, enabling the model to better fit various complex curves. At the same time, the dropout method sets the parameters in the input array to 0 according to a certain proportion, which means that the parameters in this part will no longer participate in the subsequent calculations. This can greatly improve the model's anti-interference ability.

Images