Coal quality online detection method and system

Abstract

The invention relates to the technical field of coal detection, in particular to a coal quality online detection method and system. The method comprises the following steps: acquiring values in all dimensions at all moments during historical coal detection; determining the noise possibility of each dimension at each moment; determining a correction value of each moment in each dimension; determining feature importance in each dimension at each moment; determining the adjustment of each dimension of each door of the LSTM model at each moment and then outputting the door; and based on the adjusted output, training a coal quality detection model constructed by an LSTM model to realize online detection of coal quality. According to the invention, based on statistical characteristics of a correlation mean value, a mean value, a standard deviation and the like of a target moment and a target dimension, the noise possibility in data of each dimension at each moment is dynamically discriminated, and the noise is corrected; and the noise interference problem caused by the influence of complex environments such as high dust, high humidity, vibration and electromagnetic interference on sensor signals is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal detection, and in particular to an online coal quality detection method and system. Background Art

[0002] In online coal quality monitoring, the physical and chemical properties of coal samples (such as ash, sulfur, and calorific value) directly impact subsequent combustion efficiency, pollution emissions, and resource assessment. Therefore, high-precision, real-time online monitoring methods are urgently needed. Currently, multimodal sensors (such as infrared spectrometers, gamma / neutron detectors, and temperature sensors) can be used to non-destructively sense coal conditions and generate time series data for modeling and prediction. To model the long-term dependencies and complex dynamic changes in this data, the Long Short-Term Memory network (LSTM), an improved recurrent neural network (RNN) model, effectively alleviates the vanishing gradient problem in traditional neural networks due to its unique gating mechanism.

[0003] However, coal transmission sites are often located in environments with high dust, high humidity, vibration, and electromagnetic interference. These factors seriously interfere with the signal quality of various sensors. The information entropy of sensor outputs at certain moments increases after being disturbed, and the data fluctuates violently, making it difficult for traditional LSTM models to accurately distinguish between noise and effective features, thereby mistakenly identifying interference signals as effective features. At the same time, there is a complex nonlinear coupling relationship between coal quality parameters and multi-source sensor data, and the discrimination capabilities of various dimensions under different working conditions vary significantly. When processing such multimodal data, the LSTM model defaults to equal-weighted input and lacks the ability to dynamically identify "key dimensions", which easily causes the model to fall into problems of overfitting or weak generalization performance, resulting in reduced accuracy in subsequent online coal quality detection. Summary of the Invention

[0004] In order to solve the problem that the traditional LSTM model has poor model robustness due to interference from the on-site environment, and is unable to dynamically identify key feature dimensions, resulting in low detection accuracy, thereby reducing the accuracy of subsequent online coal quality detection, the present invention provides a method and system for online coal quality detection.

[0005] In a first aspect, the present invention provides a method for online detection of coal quality, which adopts the following technical solution: A method for online detection of coal quality, comprising: obtaining the value of each dimension at each moment during historical coal detection; recording any moment as a target moment, recording any dimension as a target dimension, and determining the noise possibility of the target moment in the target dimension according to the mean of the correlation between the value in the target dimension of the time period to which the target moment belongs and the values ​​in the remaining dimensions, the value in the target dimension of the target moment, the mean of the target dimension of the time period to which the target moment belongs, and the standard deviation of the value in the target dimension of the time period to which the target moment belongs; based on the noise possibility, correcting the value in the target dimension at the target moment to obtain the corrected value in the target dimension at the target moment; and correcting the value in the target dimension according to the value in the target dimension at the target moment. The feature importance of the target moment in the target dimension is determined based on the maximum value of the information entropy and variance of the corrected value of the time period in the target dimension, and the information entropy and variance of the corrected value of the time period to which the target moment belongs in each dimension; the adjusted output of each gate of the LSTM model in the target dimension at the target moment is determined based on the projection of the original output of each gate of the LSTM model in the target dimension at the target moment, the feature importance, the corrected value of the target moment in the target dimension, and the mean value of the corrected value of the time period to which the target moment belongs in the target dimension; based on the adjusted output, the coal quality detection model constructed by the LSTM model is trained to realize online detection of coal quality.

[0006] The beneficial effects are: by dynamically judging the possibility of noise in the data of each dimension at each moment based on statistical characteristics such as the correlation mean, average and standard deviation between the target moment and the target dimension, and correcting the noise, the noise interference problem caused by complex environments such as high dust, high humidity, vibration and electromagnetic interference of the sensor signal is significantly improved; the information entropy and variance of the corrected value are used to measure the importance of features, and dynamic identification and adaptive weight adjustment of key features in different dimensions and different time periods are realized, which solves the limitation of the traditional LSTM model that defaults to equal-weighted input of multimodal data, and enhances the expression ability of complex nonlinear coupling relationships of coal quality parameters; by combining feature importance with the adjustment output of each gate of the LSTM model, the gating mechanism of the model is dynamically adjusted, so that the model focuses more on effective features and key dimensions, enhances the robustness and generalization ability of the model, reduces the risk of overfitting, and effectively improves the accuracy of subsequent online coal quality detection.

[0007] Furthermore, the dimensions include infrared spectrum dimension, temperature dimension, pressure dimension, vibration dimension and humidity dimension.

[0008] Furthermore, the values ​​in each dimension at each moment during the historical coal detection are values ​​that have undergone maximum-minimum normalization processing.

[0009] Furthermore, the noise probability satisfies: Where, For the moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the correlations between the values ​​in and the values ​​in the remaining dimensions, For the moment in dimension The value in For the The time period to which the moment belongs is in the dimension The mean of the values ​​in , For the The time period to which the moment belongs is in the dimension The standard deviation of the values ​​in , is a hyperparameter, is the max-max normalization function.

[0010] The beneficial effects are: by introducing the correlation mean as a weight and combining the standardized deviation of the target moment and dimension, it is possible to quantify the degree to which each data point deviates from its time period trend, while reducing the noise judgment probability of data points with high correlation with the remaining dimensions, thereby more accurately distinguishing valid signals from noise; the introduction of the standard deviation and ensures that the calculation of noise possibility not only depends on the deviation of the data value, but also takes into account the fluctuation range and noise tolerance of the data, thereby improving the adaptability and robustness of the model to different working conditions and environmental fluctuations.

[0011] Furthermore, the correlation adopts the Pearson correlation coefficient.

[0012] Furthermore, the correction value satisfies: Where, For the moment in dimension The correction value in For the A moment in dimension The value in For the A moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the values ​​in .

[0013] The beneficial effects are: by dynamically adjusting the contribution ratio of the original observation value to the mean of the period to which it belongs through the noise possibility weight, the abnormal data that may be interfered by noise can be smoothed and corrected to the stable period mean in a targeted manner, thereby reducing the impact of noise on the model input; for data points with low noise possibility, the corrected value is closer to the original observation value, ensuring the retention of effective signals and feature integrity, and avoiding the loss of important information caused by the unified processing of all data by traditional simple filtering.

[0014] Furthermore, the feature importance satisfies: Where, For the A moment in dimension The feature importance of For the The time period to which the moment belongs is in the dimension The information entropy of the corrected value in For the The time period to which the moment belongs is in the dimension The variance of the correction value in , For the The maximum value of the information entropy of the correction value in each dimension of the time period to which the moment belongs, For the The maximum value of the variance of the correction value in each dimension of the time period to which the moment belongs, is the number of dimensions.

[0015] The beneficial effects are: by combining the two indicators of information entropy and variance, the complexity and volatility of the data of each dimension in the time period to which the target moment belongs are comprehensively measured, so that the key feature dimensions with the most discriminative power for coal quality detection can be scientifically and dynamically identified; the maximum value of the information entropy and variance of each dimension is used for normalization to eliminate the influence of different dimensions and data ranges, so that the feature importance is reasonably distributed among all dimensions, ensuring that the model can fairly and effectively utilize multidimensional data; through the normalization of feature importance weights, the model can highlight the role of key dimensions and weaken the dependence on redundant or noisy dimensions, thereby improving the LSTM model's ability to capture important signals in complex nonlinear coupling environments, which helps to improve prediction accuracy and generalization performance.

[0016] Furthermore, the adjusted output satisfies: Where, The first The door in The dimension of a moment The adjusted output, The first The original output of the gate is The dimension of a moment The projection, For the A moment in dimension The feature importance of For the A moment in dimension The correction value in For the The time period to which the moment belongs is in the dimension The mean of the corrected values ​​in , In order to use the adaptive learning rate algorithm to dynamically adjust the learning rate parameters according to the model training process, for function.

[0017] The beneficial effects are: by combining feature importance with correction value deviation and applying weighted adjustment of dynamic learning rate, the original gated output is effectively corrected, making the model more sensitive to effective changes in key dimensions, thereby enhancing the ability of LSTM to capture complex nonlinear time series features; using feature importance weights, the gating mechanism can dynamically amplify or suppress information flow according to the importance of each dimension, which helps to distinguish key features from noise, strengthen the model's selective forgetting and memory functions, and avoid the performance degradation caused by traditional LSTM treating all input dimensions equally; the adjusted output is mapped through the activation function to ensure that the gated output varies between 0 and 1, smoothly and stably controlling the flow of information, helping to prevent gradient explosion or disappearance, and improving the stability of model training.

[0018] Furthermore, the training LSTM model to construct a coal quality detection model includes: using the coal quality parameters corresponding to the values ​​in each dimension at each moment during the pre-acquired historical coal detection as labels for the LSTM model to construct the coal quality detection model, and forming a vector of the adjusted outputs of each dimension of each gate of the LSTM model at the target moment as the final output of each gate of the LSTM model at the target moment, and training the coal quality detection model of the LSTM model, wherein the coal quality parameters include ash content, sulfur content, calorific value, volatile matter and moisture.

[0019] In a second aspect, the present invention provides an online coal quality detection system, which adopts the following technical solution: A coal quality online detection system includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned coal quality online detection method is implemented.

[0020] By adopting the above technical solution, the above-mentioned coal quality online detection method is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0021] The present invention has the following technical effects: By comprehensively evaluating the correlation, mean, and standard deviation of data in each dimension at each moment, it is possible to accurately identify and correct noise data generated in environments such as high dust, high humidity, vibration, and electromagnetic interference, significantly improving the effectiveness and stability of sensor signals; based on the "feature importance" measurement of information entropy and variance, the model can automatically adjust the input weights of each dimension under different working conditions, highlighting the discrimination ability of key dimensions, and avoiding the feature redundancy or omission problems caused by traditional LSTM equal-weighted input; pre-correction of noise and combining it with dynamic feature importance allow the model to focus more on real and effective quality features during training, thereby reducing the overfitting tendency caused by random interference and improving the generalization performance of the model; the LSTM coal quality detection model constructed by the present invention can more accurately capture the nonlinear coupling relationship between coal quality parameters and multi-source sensor data, realize real-time, high-precision online quality assessment, and meet the needs of industrial sites for fast and reliable detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a method for online detection of coal quality in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0024] The embodiment of the present invention discloses a method for online detection of coal quality, referring to Figure 1 , including steps S01 to S06: S01: Obtain the values ​​of each dimension at each moment during historical coal detection.

[0025] It should be noted that near-infrared spectrometers, temperature sensors, pressure sensors, vibration sensors, and humidity sensors are deployed along the coal transportation belt of the coal mine. For example, data is collected at intervals of 1 second, and time synchronization is ensured. The data is normalized and each eigenvalue is mapped to interval, and obtain the median value of each dimension at each moment in the historical coal testing; and obtain the coal quality parameters (such as ash content, sulfur content, calorific value, etc.) corresponding to the values ​​in each dimension at each moment in the historical coal testing through laboratory analysis.

[0026] Specifically, the dimensions include infrared spectrum dimension, temperature dimension, pressure dimension, vibration dimension and humidity dimension.

[0027] Specifically, the values ​​in each dimension at each moment during the historical coal detection are values ​​that have undergone maximum-minimum normalization processing.

[0028] S02: Determine the noise probability in each dimension at each moment.

[0029] It's important to note that there's a close physical connection between sensor data at the coal testing site. Changes in coal quality often cause coordinated changes in data across multiple dimensions, while noise typically manifests as abnormal fluctuations in a single dimension. This is due to factors like dust, water mist, vibration, and electromagnetic interference at the site. Its scope and mode of action are localized and targeted. For example, electromagnetic interference typically only affects sensors with electromagnetic induction characteristics (such as pressure and temperature sensors that output electrical signals) and does not simultaneously interfere with optical sensors like infrared spectrometers. Dust adhering to the lens of a near-infrared spectrometer only causes anomalies in the spectral signal and does not directly affect the coal transportation vibration data collected by the vibration sensor. These interference sources often target specific sensor types or the detection of a specific physical quantity, resulting in single-dimensional data fluctuations. Therefore, this step determines the potential for noise in each dimension at each moment by quantifying the correlation between the data in each dimension.

[0030] Record any moment as the target moment and any dimension as the target dimension. Determine the noise possibility in the target dimension at the target moment based on the mean of the correlation between the value in the target dimension of the time period to which the target moment belongs and the values ​​in the remaining dimensions, the value in the target dimension of the target moment, the mean of the target dimension of the time period to which the target moment belongs, and the standard deviation of the value in the target dimension of the time period to which the target moment belongs.

[0031] Implementers can set the number of moments within the time period to which the target moment belongs based on specific implementation circumstances, for example, 15.

[0032] Specifically, the noise probability satisfies: ; Where, For the A moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the correlations between the values ​​in and the values ​​in the remaining dimensions, For the A moment in dimension The value in For the The time period to which the moment belongs is in the dimension The mean of the values ​​in , For the The time period to which the moment belongs is in the dimension The standard deviation of the values ​​in , is a hyperparameter, is the max-max normalization function.

[0033] Implementers can set hyperparameters according to specific implementation conditions, for example, 0.001. The existence of hyperparameters is to prevent , making the formula meaningless.

[0034] Specifically, the correlation adopts the Pearson correlation coefficient.

[0035] in, It reflects the degree of coordinated change between the dimension data and the rest of the dimension data (the Pearson correlation coefficient is used to measure linear correlation, the higher the mean, the greater the dimension The more consistent the change trend is with the other dimensions; on the contrary, it means that the dimension The weaker the correlation with other dimensions), the larger the value, the stronger the dimension The more independent the data change of is from the other dimensions, the less it conforms to the multi-dimensional collaborative fluctuation characteristics caused by coal quality changes. The smaller the size, the bigger it is; and vice versa. Indicates the A moment in dimension The standardized deviation of the data from the mean of the period (normalized by standard deviation) reflects the A moment in the time period in the dimension The degree of abnormality in the period, the mean and standard deviation of the period describe the normal fluctuation range. The greater the deviation, the The more the data at a moment exceeds the normal fluctuation pattern, the larger the value is, indicating that the The higher the degree of deviation from the historical fluctuation range at a moment, the more likely it is a single-dimensional anomaly caused by local interference (such as dust, electromagnetic interference). The smaller the size, the bigger it is; and vice versa.

[0036] S03: Determine the correction value in each dimension at each moment.

[0037] It's important to note that noise in coal testing data typically manifests as abnormal fluctuations in a single dimension, while true coal quality changes can trigger coordinated changes in multiple dimensions. Therefore, this step weights the raw data based on the noise potential, suppressing outliers caused by local interference while retaining the true signal, making the data more consistent with the physical laws of coal quality variation.

[0038] Based on the noise possibility, the value of the target moment in the target dimension is corrected to obtain a corrected value of the target moment in the target dimension.

[0039] Specifically, the correction value satisfies: ; Where, For the A moment in dimension The correction value in For the A moment in dimension The value in For the A moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the values ​​in .

[0040] in, Indicates the degree of preservation of the original value during correction, reflecting the possibility that the data at that moment belongs to the real signal. When it is small (noise possibility is low), When it approaches 1, the original value accounts for a high proportion, indicating that the data is more likely to be caused by changes in coal quality, and the corrected value is close to the original collection result. Indicates the weight of the period mean in the correction, reflecting the degree of interference of noise on the data. When it is large (noise possibility is high), If it approaches 1, the time period average has a high proportion, indicating that the data is more likely to be caused by local interference (such as dust and electromagnetic interference), and the correction value is close to the normal fluctuation range (mean).

[0041] S04: Determine the importance of features in each dimension at each moment.

[0042] It's important to note that in multi-dimensional coal testing data, different dimensions contribute differently to coal quality parameters (such as ash content and calorific value). Furthermore, during the mining process, coal ash, sulfur content, calorific value, and other parameters naturally vary between different coal seams and mining faces. For example, a coal seam mined in the morning may be rich in kaolin (the source of ash), significantly increasing the infrared spectrum's sensitivity to ash changes during this period. In the afternoon, when the data shifts to a pyrite-rich layer, sulfur-related spectral features become dominant, shifting the infrared spectrum's sensitivity focus from ash to sulfur. Therefore, the sensitivity of each dimension changes dynamically over time as coal quality changes. While the LSTM model has the ability to model long-term dependencies, its standard structure assumes equal weighting for all input dimensions, which can easily introduce high noise or low-contribution signals, leading to problems such as mislearning and overfitting. Therefore, it's necessary to dynamically assess the importance of each feature before data enters the LSTM model, thereby guiding the model's focus on the feature dimensions most sensitive to the current coal quality state. Therefore, this step quantifies the information richness and fluctuation significance of data in each dimension, dynamically evaluates the importance of features in different dimensions at each moment, increases the attention of subsequent models to key features, and enhances the adaptability and robustness of the detection system to environmental disturbances.

[0043] The feature importance of the target moment in the target dimension is determined based on the information entropy and variance of the corrected value of the time period to which the target moment belongs in the target dimension, and the maximum value of the information entropy and variance of the corrected value of the time period to which the target moment belongs in each dimension.

[0044] Specifically, the feature importance satisfies: ; Where, For the A moment in dimension The feature importance of For the The time period to which the moment belongs is in the dimension The information entropy of the corrected value in For the The time period to which the moment belongs is in the dimension The variance of the correction value in , For the The maximum value of the information entropy of the correction value in each dimension of the time period to which the moment belongs, For the The maximum value of the variance of the correction value in each dimension of the time period to which the moment belongs, is the number of dimensions.

[0045] in, Indicates passing After normalization, The time period to which the moment belongs is in the dimension The information uncertainty reflects the fluctuation diversity of the value of this dimension within the time window. The larger the value, the more uniform the distribution of the data in this dimension, the higher the state diversity, and the more high entropy information features related to coal quality changes may be included. This means that the distinguishing ability of this dimension is strong and it should be given a higher importance weight. The smaller the size, the bigger it is; and vice versa. Indicates passing After normalization, The time period to which the moment belongs is in the dimension The data discreteness reflects the fluctuation intensity of the dimension value within the time window. The larger the value, the more dramatic changes have occurred in the dimension during the period, and it may have captured dynamic signals related to coal quality changes, such as particle size, moisture, ash content fluctuations, etc. Therefore, its weight should also be increased. The larger the value, the greater the value; vice versa. Finally, by dividing Make sure the sum of feature importances for all dimensions is 1.

[0046] S05: Determine the adjusted output of each gate of the LSTM model in each dimension at each moment.

[0047] It should be noted that in the traditional LSTM structure, the states of the forget gate, input gate, and output gate are calculated from the input features and hidden state. By default, all input dimensions are treated equally, making it difficult to distinguish the differences in the contributions of different sensor features to coal quality changes at a given moment. This can cause high-noise dimensions to mislead the gating state, thereby affecting the update and output of memory units. Therefore, to enhance the LSTM's focus on highly discriminative dimensions, this step introduces a feature-importance-based gating adjustment term into the LSTM's gating mechanism. This dynamically amplifies the influence of key dimensional features on the gating state and suppresses the interference of less discriminative features.

[0048] The adjusted output of each gate of the LSTM model in the target dimension at the target moment is determined based on the projection of the original output of each gate of the LSTM model in the target dimension at the target moment (the standard forget gate output is mapped to the feature dimension space through linear transformation), the feature importance, the correction value in the target dimension at the target moment, and the average of the correction values ​​in the target dimension for the time period to which the target moment belongs.

[0049] Specifically, the adjusted output satisfies: ; Where, The first The door in The dimension of a moment The adjusted output, The first The original output of the gate is The dimension of a moment The projection, For the A moment in dimension The feature importance of For the A moment in dimension The correction value in For the The time period to which the moment belongs is in the dimension The mean of the corrected values ​​in , In order to use the adaptive learning rate algorithm to dynamically adjust the learning rate parameters according to the model training process, for function.

[0050] Implementers can set an adaptive learning rate algorithm, such as the Adam algorithm, based on specific implementation circumstances.

[0051] in, Represents LSTM The original output of the gates (forget gate / input gate / output gate) is in dimension The linear projection value on reflects the basic influence of the dimension feature on the gating state in the traditional LSTM structure. This value is generated by the input feature and hidden state of LSTM through matrix operation. Its size directly determines the regulation strength of the gating signal on the memory unit. Reflects the dimension In the The larger the value is, the more significant the coal quality change information is in the dimension during the current period, and its contribution to the gate state should be increased. The smaller the size, the bigger it is; and vice versa. Indicates the A moment in dimension The deviation degree between the correction value and the mean value of the period reflects whether there is a sudden change or abnormal fluctuation in coal quality at the current moment. The larger the value, the more likely there is a key event such as a sudden change in coal quality. Therefore, the response to the change should be strengthened. The smaller the size, the bigger it is; and vice versa. It is used to control the adjustment amplitude of feature importance and deviation on the gate, which can be adjusted dynamically according to the training process to avoid oscillation or over-correction of the model.

[0052] S06: Based on the adjusted output, training a coal quality detection model constructed by the LSTM model to achieve online detection of coal quality.

[0053] Specifically, the training LSTM model to construct a coal quality detection model includes: The coal quality parameters corresponding to the values ​​in each dimension at each moment during the historical coal detection obtained in advance are used as labels for the LSTM model to construct a coal quality detection model. The adjusted outputs of each gate of the LSTM model in each dimension at the target moment are used as vectors as the final outputs of each gate of the LSTM model at the target moment. The coal quality detection model of the LSTM model is trained. The coal quality parameters include ash content, sulfur content, calorific value, volatile matter and moisture.

[0054] After the model training is completed, the trained coal quality detection model is used to perform online coal quality detection on the coal to be tested. That is, the multimodal sensor data of the coal to be tested is input into the trained coal quality detection model, and the trained coal quality detection model is used to make real-time predictions on the coal quality parameters of the coal to be tested to complete the online coal quality detection. The LSTM model is trained and optimized according to actual conditions to improve the model's adaptability to new coal quality conditions and environmental changes, and to continuously ensure the accuracy of online coal quality detection (for example, when it is detected that the ash content prediction error exceeds 1.5% for three consecutive times due to the mining of a new type of coal, or the online detection cycle reaches 10 times, etc., new multimodal data and corresponding coal quality parameters are obtained to retrain the model).

[0055] An embodiment of the present invention further discloses an online coal quality detection system, comprising a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, an online coal quality detection method according to the present invention is implemented.

[0056] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0057] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for online detection of coal quality, characterized in that: include: Obtain the values ​​of each dimension at each moment during historical coal testing; Record any moment as the target moment and any dimension as the target dimension. Determine the noise possibility of the target moment in the target dimension based on the mean of the correlation between the value in the target dimension of the time period to which the target moment belongs and the values ​​in the remaining dimensions, the value in the target dimension of the target moment, the mean of the value in the target dimension of the time period to which the target moment belongs, and the standard deviation of the value in the target dimension of the time period to which the target moment belongs. Based on the noise possibility, correcting the value of the target moment in the target dimension to obtain a corrected value of the target moment in the target dimension; Determine the feature importance of the target moment in the target dimension based on the information entropy and variance of the corrected value of the time period to which the target moment belongs in the target dimension, and the maximum value of the information entropy and variance of the corrected value of the time period to which the target moment belongs in each dimension; determine the adjusted output of each gate of the LSTM model in the target dimension at the target moment based on the projection of the original output of each gate of the LSTM model in the target dimension at the target moment, the feature importance, the corrected value of the target moment in the target dimension, and the mean of the corrected value of the time period to which the target moment belongs in the target dimension; Based on the adjusted output, a coal quality detection model constructed by the LSTM model is trained to achieve online detection of coal quality.

2. A method for online detection of coal quality according to claim 1, characterized in that: The dimensions include infrared spectrum dimension, temperature dimension, pressure dimension, vibration dimension and humidity dimension.

3. A method for online detection of coal quality according to claim 1, characterized in that: The values ​​in each dimension at each moment during the historical coal detection are values ​​that have undergone maximum-minimum normalization processing.

4. A method for online detection of coal quality according to claim 1, characterized in that: The noise probability satisfies: ； Where, For the A moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the correlations between the values ​​in and the values ​​in the remaining dimensions, For the A moment in dimension The value in For the The time period to which the moment belongs is in the dimension The mean of the values ​​in , For the The time period to which the moment belongs is in the dimension The standard deviation of the values ​​in , is a hyperparameter, is the max-max normalization function.

5. A method for online detection of coal quality according to claim 1 or 3, characterized in that: The correlation was calculated using the Pearson correlation coefficient.

6. A method for online detection of coal quality according to claim 1, characterized in that: The correction value satisfies: ； Where, For the A moment in dimension The correction value in For the A moment in dimension The value in For the A moment in dimension The possibility of noise, For the The time period to which the moment belongs is in the dimension The mean of the values ​​in .

7. A method for online detection of coal quality according to claim 1, characterized in that: The feature importance satisfies: ； Where, For the A moment in dimension The feature importance of For the The time period to which the moment belongs is in the dimension The information entropy of the corrected value in For the The time period to which the moment belongs is in the dimension The variance of the correction value in , For the The maximum value of the information entropy of the correction value in each dimension of the time period to which the moment belongs, For the The maximum value of the variance of the correction value in each dimension of the time period to which the moment belongs, is the number of dimensions.

8. The method for online detection of coal quality according to claim 1, characterized in that: The adjusted output satisfies: ; Where, The first The door in The dimension of a moment The adjusted output, The first The original output of the gate is The dimension of a moment The projection, For the A moment in dimension The feature importance of For the A moment in dimension The correction value in For the The time period to which the moment belongs is in the dimension The mean of the corrected values ​​in , In order to use the adaptive learning rate algorithm to dynamically adjust the learning rate parameters according to the model training process, for function.

9. The method for online detection of coal quality according to claim 1, characterized in that: The training of the LSTM model to construct a coal quality detection model includes: The coal quality parameters corresponding to the values ​​in each dimension at each moment during the historical coal detection obtained in advance are used as labels for the LSTM model to construct a coal quality detection model. The adjusted outputs of each gate of the LSTM model in each dimension at the target moment are used as vectors as the final outputs of each gate of the LSTM model at the target moment. The coal quality detection model of the LSTM model is trained. The coal quality parameters include ash content, sulfur content, calorific value, volatile matter and moisture.

10. A coal quality online detection system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an online coal quality detection method according to any one of claims 1 to 9 is implemented.

Images