A dry fog dust reduction system for online monitoring of dust concentration

Abstract

The present invention relates to the field of data processing technology, and specifically to a dust concentration online monitoring dry fog dust reduction system; a first feature classification hidden layer is constructed according to the difference in concentration change characteristic values ​​between a dust concentration sequence and a dust concentration historical sequence; a first classification condition is constructed according to the data distribution characteristics in the dust concentration sequence; a second classification model and a third classification model are constructed according to the data difference characteristics and data discrete characteristics at the same time in the dust concentration sequence and the dust concentration historical sequence. A second feature classification hidden layer and a second feature classification additional hidden layer are constructed according to the first classification condition, the second classification model and the third classification model. The present invention constructs an LSTM concentration prediction improved network model according to different hidden layers to predict the dust concentration sequence, obtain the predicted concentration value at the future moment and adjust the dry fog dust reduction system; the data prediction accuracy and dust reduction effect are improved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a dry fog dust reduction system for online monitoring of dust concentration. Background Art

[0002] The dry fog dust reduction system is a device that reduces the dust concentration in the air by spraying water mist or chemicals. It effectively reduces the dust concentration in the air and improves the air quality of the environment. It is more energy-saving and environmentally friendly than traditional dust removal methods. The higher the dust concentration in the environment, the greater the spraying volume in the dust reduction system. The existing dust reduction system adjusts the spraying volume based on the dust concentration collected in real time. The disadvantage is that the spraying volume can only be increased when the dust concentration is monitored to increase, resulting in untimely dust reduction. Therefore, in order to improve the dust reduction effect, it is necessary to improve the dust reduction system and adjust the spraying volume in advance by predicting the dust concentration at the next moment.

[0003] Dust concentration can be predicted through the existing LSTM long short-term memory prediction network, but the change of dust concentration is easily affected by environmental factors and human factors, and the data fluctuation is relatively complex. The existing LSTM prediction network cannot effectively capture the complex change patterns and laws of dust concentration data only through the input layer features; resulting in low accuracy of dust concentration prediction results and poor dust reduction effect of dry fog dust reduction system. Summary of the invention

[0004] In order to solve the technical problem that the use of the existing LSTM prediction network leads to inaccurate prediction of dust concentration, which affects the dust reduction effect of the dry fog dust reduction system, the purpose of the present invention is to provide a dry fog dust reduction system for online monitoring of dust concentration. The technical solution adopted is as follows:

[0005] A data acquisition module, used to acquire a dust concentration sequence for monitoring dust concentration in a current period, and at least two dust concentration history sequences;

[0006] A hidden layer construction module is used to obtain concentration change characteristic values ​​according to data difference characteristics of adjacent data points in a dust concentration sequence and a dust concentration history sequence; construct a first feature classification hidden layer according to difference characteristics and data distribution characteristics of concentration change characteristic values ​​at the same time between the dust concentration sequence and the dust concentration history sequence; construct a first classification condition according to the data distribution characteristics in the dust concentration sequence; construct a second classification model according to data difference characteristics at the same time in the dust concentration sequence and the dust concentration history sequence; construct a third classification model according to discrete characteristics of data at the same time in the dust concentration sequence and the dust concentration history sequence; construct a second feature classification hidden layer according to the first classification condition, the second classification model and the third classification model; and construct a second feature classification additional hidden layer according to the second classification model and the third classification model;

[0007] A prediction network construction module, used for constructing an LSTM concentration prediction improved network model according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer;

[0008] The concentration prediction module is used to predict the dust concentration sequence according to the LSTM concentration prediction improved network model to obtain the predicted concentration value at the future moment; and adjust the dry fog dust reduction system according to the concentration prediction value.

[0009] Furthermore, the step of obtaining the concentration change characteristic value according to the data difference characteristics of adjacent data points in the dust concentration sequence and the dust concentration history sequence includes:

[0010] For any data point in the dust concentration sequence and the dust concentration history sequence, the absolute value of the difference between the arbitrary data point and other data points at adjacent moments is calculated to obtain a concentration change characteristic value of the arbitrary data point.

[0011] Furthermore, the step of constructing a first feature classification hidden layer according to the difference characteristics of the concentration change characteristic values ​​at the same time between the dust concentration sequence and the dust concentration history sequence and the data distribution characteristics comprises:

[0012] For any collection moment, the premise that the data point values ​​of all historical dust concentration history sequences at the said arbitrary collection moment are greater than the preset concentration threshold is taken as a precondition; on the premise that the precondition is met, the average value of the absolute value of the difference between the dust concentration sequence and the concentration change characteristic value of each dust concentration history sequence at the said arbitrary collection moment is calculated to obtain the change difference characteristic value of the dust concentration sequence at the said arbitrary collection moment; the change difference characteristic value is negatively correlated mapped to obtain the first judgment model; the first feature classification hidden layer is constructed according to the said precondition and the first judgment model.

[0013] Furthermore, the step of constructing a first classification condition according to the data distribution characteristics in the dust concentration sequence includes:

[0014] The first classification condition is that the data point value of the dust concentration sequence at any collection time is greater than the preset concentration threshold.

[0015] Furthermore, the step of constructing a second classification model according to the data difference characteristics at the same time in the dust concentration sequence and the dust concentration history sequence includes:

[0016] Calculate the average value of the data point values ​​of the dust concentration historical sequence at the arbitrary collection time to obtain the historical concentration average value at the arbitrary collection time; calculate the absolute value of the difference between the data point values ​​of the dust concentration sequence at the arbitrary collection time and the historical concentration average value and normalize them to obtain the second classification model.

[0017] Furthermore, the step of constructing a third classification model according to the discrete features of data at the same time in the dust concentration sequence and the dust concentration history sequence includes:

[0018] The absolute value of the difference between any two data points of the dust concentration sequence and the dust concentration history sequence at any collection time is calculated to obtain a concentration difference characteristic value; the variance of the concentration difference characteristic value is calculated and normalized to obtain the third classification model.

[0019] Furthermore, the step of constructing an improved LSTM concentration prediction network model according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer includes:

[0020] According to the data features of the dust concentration historical sequence, the LSTM network is trained to obtain the LSTM concentration prediction network model; the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer are added to the LSTM concentration prediction network model and optimized and learned to obtain the LSTM concentration prediction improved network model; the input of the LSTM concentration prediction improved network model is the dust concentration sequence of the current time period, and the output is the predicted concentration value at the future moment.

[0021] Furthermore, the step of predicting the dust concentration sequence according to the LSTM concentration prediction improved network model to obtain the predicted concentration value at the future moment includes:

[0022] The dust concentration sequence is input into the LSTM concentration prediction improved network model; for the target data point at the target collection time in the dust concentration sequence, the target data point is placed in the first feature classification hidden layer, and when the target data point satisfies the first judgment model and exceeds the preset first threshold, the target data point is regarded as a normal change data point; when the target data point is not a normal change data point; the target data point is placed in the second feature classification hidden layer, and when the target data point satisfies the first classification condition, and the second classification model and the third classification model both exceed the preset second threshold, the target data point is regarded as an abnormal mutation data point;

[0023] When the target data point is an abnormal mutation data point, other data points in a preset neighborhood of the target data point are placed in a second feature classification additional hidden layer; the number of other data points that satisfy both the second classification model and the third classification model and exceed a preset second threshold is counted as a neighborhood quantity value; the ratio of the neighborhood quantity value to the number of other data points in a preset neighborhood is calculated to obtain a neighborhood proportion; when the neighborhood proportion exceeds a preset proportion threshold, the target data point is regarded as an environmental abnormality data point, otherwise the target data point is regarded as a human interference data point;

[0024] In the LSTM concentration prediction improvement network model, the output value is adjusted and improved according to the type of the target data point to obtain the predicted concentration value at the future moment.

[0025] The present invention has the following beneficial effects:

[0026] In the present invention, obtaining the concentration change characteristic value can characterize the dust concentration change characteristics at adjacent moments, so as to construct the first feature classification hidden layer according to the difference characteristics of the dust concentration change characteristics of different periods, and can preliminarily determine the change inducement of dust concentration according to the change characteristics of dust concentration in the prediction process; adjust the prediction result according to the change inducement, and improve the prediction accuracy. Obtaining the second classification model and the third classification model can construct the second feature classification hidden layer according to the data characteristics of the same data points of the same and different periods, and can further judge the change inducement of dust concentration according to the difference of dust concentration in different periods in the prediction process, so that the prediction model can adjust the prediction result according to the change inducement, and improve the prediction accuracy. Obtaining the second feature classification additional hidden layer can further distinguish the change inducement on the basis of the second feature classification hidden layer, so as to make the prediction result more accurate. The obtained LSTM concentration prediction improved network model can judge the change inducement of dust concentration according to the data characteristics of the dust concentration sequence, and make different degrees of dust concentration prediction adjustment according to different change inducements; compared with the existing LSTM prediction model, it can improve the prediction accuracy. Finally, the dry fog dust reduction system is adjusted according to the concentration prediction value to improve the dust reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 A block diagram of a dry fog dust reduction system for online monitoring of dust concentration provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, characteristics and effects of a dust concentration online monitoring dry fog dust reduction system proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0030] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0031] The specific scheme of the dry fog dust reduction system for online monitoring of dust concentration provided by the present invention is described in detail below with reference to the accompanying drawings.

[0032] See also Figure 1 , which shows a block diagram of a dust concentration online monitoring dry fog dust reduction system provided by an embodiment of the present invention, the system includes the following modules:

[0033] The data acquisition module S1 is used to acquire a dust concentration sequence for monitoring the dust concentration in the current period and at least two dust concentration history sequences.

[0034] In an embodiment of the present invention, the implementation scenario is to predict dust concentration and adjust the dry fog dust reduction system according to the prediction results. LSTM long short-term memory neural network is an existing commonly used prediction method. The LSTM prediction network is trained and tested according to the historical data sequence of the prediction object to obtain a special prediction model for the prediction object; therefore, the dust concentration sequence for monitoring the dust concentration in the current period and at least two dust concentration historical sequences are first obtained; the dust concentration sequence of the current period is updated in real time with the change of the latest collection time, and the data point with the farthest collection time is continuously discarded to achieve the same length of the dust concentration sequence and the dust concentration historical sequence. Because the dust concentration will show periodic changes, a prediction model can be constructed according to the historical dust concentration changes during prediction; in the embodiment of the present invention, the dust concentration historical sequence has a cycle length of 24 hours, and the collection time of the data points between different dust concentration historical sequences corresponds one to one; the implementer can determine the collection frequency and the length of the dust concentration historical sequence by himself.

[0035] Because the change of dust concentration is affected by many factors, such as environmental changes and the influence of human activities, the data fluctuation characteristics are relatively complex, resulting in the inability of the existing LSTM prediction model to effectively capture the complex change patterns and laws of dust concentration only through the input layer characteristics, affecting the accuracy of the prediction results, resulting in poor dust reduction effects. Therefore, it is necessary to improve the LSTM prediction model, analyze the change characteristics of the data before prediction, and determine the change inducement of the data, so that the prediction model adjusts the prediction result value according to the change inducement to improve the prediction accuracy. In the present invention, multiple LSTM hidden layers are stacked on the prediction model in combination with the change characteristics of dust concentration to increase the depth of the model, so that the prediction model can better capture the complex patterns in the time series data, improve the understanding and characterization capabilities of the prediction model, and increase the accuracy of the prediction results of dust concentration.

[0036] Hidden layer construction module S2 is used to obtain concentration change characteristic values ​​according to the data difference characteristics of adjacent data points in the dust concentration sequence and the dust concentration historical sequence; construct a first feature classification hidden layer according to the difference characteristics and data distribution characteristics of the concentration change characteristic values ​​at the same time between the dust concentration sequence and the dust concentration historical sequence; construct a first classification condition according to the data distribution characteristics in the dust concentration sequence; construct a second classification model according to the data difference characteristics at the same time in the dust concentration sequence and the dust concentration historical sequence; construct a third classification model according to the discrete characteristics of the data at the same time in the dust concentration sequence and the dust concentration historical sequence; construct a second feature classification hidden layer according to the first classification condition, the second classification model and the third classification model; and construct a second feature classification additional hidden layer according to the second classification model and the third classification model.

[0037] There are several main factors that lead to changes in dust concentration: changes in the quality of the air environment itself, human activities, meteorological conditions or seasonal changes, etc. Therefore, in order to better build a prediction model, it is necessary to analyze the causes of changes in dust concentration at a certain collection time, so that the prediction model can effectively enhance or weaken the values ​​of the concentration value change nodes caused by different causes, so as to achieve the purpose of optimizing the prediction results.

[0038] First, the concentration change characteristic value is obtained according to the data difference characteristics of adjacent data points in the dust concentration sequence and the dust concentration historical sequence. Preferably, in the embodiment of the present invention, obtaining the concentration change characteristic value includes: for any data point in the dust concentration sequence and the dust concentration historical sequence, calculating the absolute value of the difference between any data point and other data points at adjacent moments, and obtaining the concentration change characteristic value of the arbitrary data point; in the embodiment of the present invention, the other data points at adjacent moments are the data points at the last acquisition moment of the arbitrary data point. The larger the concentration change characteristic value, the more obvious the dust concentration change characteristic of the arbitrary data point is; the specific cause of the dust concentration change can be analyzed according to the difference in the concentration change characteristic values ​​corresponding to the data points at the same acquisition moment in the dust concentration sequence and the dust concentration historical sequence, so the first feature classification hidden layer is constructed according to the difference characteristics of the concentration change characteristic values ​​at the same moment between the dust concentration sequence and the dust concentration historical sequence and the data distribution characteristics.

[0039] Preferably, in one embodiment of the present invention, the step of constructing a first feature classification hidden layer includes: for any collection moment, taking the data point values ​​of all historical dust concentration historical sequences at any collection moment as a precondition greater than a preset concentration threshold, and the implementer can set the preset concentration threshold according to the implementation scenario; because the greater the dust concentration, the more intervention is required for the dust reduction system, so when this precondition is met, it means that the dust concentration at the collection moment is relatively abnormal, and then it is necessary to judge the specific cause of the dust concentration change at the collection moment in the dust concentration sequence, analyze the collection moment where abnormal dust concentration may occur, and improve the prediction accuracy. On the premise of meeting the preconditions, calculate the average of the absolute values ​​of the difference between the dust concentration sequence and the concentration change characteristic values ​​of each dust concentration historical sequence at any collection time, and obtain the change difference characteristic value of the dust concentration sequence at any collection time; negatively correlate the change difference characteristic value to obtain the first judgment model; the smaller the change difference characteristic value, the greater the result of the first judgment model, which means that the cause of the dust concentration change in the current period is more likely to be the same as the cause of the dust concentration change at the same collection time in history, which may be caused by meteorological conditions or seasonal periodic changes. It is a normal dust concentration change, and the prediction can be made according to the change law of historical dust concentration. Then, the first feature classification hidden layer is constructed according to the preconditions and the first judgment model. Adding the first feature classification hidden layer in the prediction model can further analyze the cause of dust concentration change, and adjust the prediction result accordingly according to the cause.

[0040] Furthermore, if the data point at a certain collection time in the dust concentration sequence of the current time period does not meet the cause identified by the first judgment model, it is necessary to add a second feature classification hidden layer in the prediction model to identify the cause of the dust concentration change at the data point. First, the first classification condition is constructed according to the data distribution characteristics in the dust concentration sequence, specifically including: taking the value of the data point in the dust concentration sequence at any collection time greater than the preset concentration threshold as the first classification condition. If the value of any data point in the dust concentration sequence in the current time period is greater than the preset concentration threshold, it means that the dust concentration of the arbitrary data point is too high, and the dust reduction system needs more intervention. It is necessary to consider the cause of the dust concentration change at the arbitrary data point when predicting.

[0041] After obtaining the first classification condition, a second classification model is constructed based on the data difference characteristics of the dust concentration sequence and the dust concentration historical sequence at the same time. Preferably, in one embodiment of the present invention, obtaining the second classification model includes: calculating the average value of the data point values ​​of the dust concentration historical sequence at any collection time to obtain the historical concentration average value at the arbitrary collection time; reflecting the average level of dust concentration at the arbitrary collection time in history through the historical concentration average value. Calculate the absolute value of the difference between the data point value of the dust concentration sequence at any collection time and the historical concentration average value and normalize them to obtain the second classification model. The larger the result of the second classification model, the more likely it is that the dust concentration at any collection time in the current period is an abnormal change in dust concentration caused by human activities or spontaneous random changes in the air environment, which is greatly different from the dust concentration at any collection time in history, and cannot be predicted based solely on the historical dust concentration change law.

[0042] In order to improve the accuracy of the cause analysis, a third classification model is constructed based on the discrete features of the data at the same time in the dust concentration sequence and the dust concentration history sequence; preferably, in one embodiment of the present invention, the step of constructing the third classification model includes: calculating the absolute value of the difference between any two data points in the dust concentration sequence and the dust concentration history sequence at any collection time to obtain the concentration difference characteristic value; calculating the variance of the concentration difference characteristic value and normalizing it to obtain the third classification model. The larger the variance of the concentration difference characteristic value, the more random the numerical distribution of the dust concentration at any collection time in the history, and the more likely it is that abnormal changes in dust concentration caused by human activities or the air environment itself will occur at any collection time.

[0043] Furthermore, the second feature classification hidden layer can be constructed according to the first classification condition, the second classification model and the third classification model. Since the second feature classification hidden layer can only analyze the abnormal changes in dust concentration caused by human activities or the air environment itself, the deviation of the two on the prediction results is still large. In order to improve the prediction accuracy, the two need to be further analyzed; because the duration of the abnormal changes in dust concentration caused by human activities and the air environment itself is different, the abnormal changes in dust concentration caused by human activities are shorter. Therefore, the second feature classification additional hidden layer can be constructed according to the second classification model and the third classification model, and the data points at the adjacent moments of any collection moment are analyzed to determine the persistence of the abnormal changes in dust concentration.

[0044] The prediction network construction module S3 is used to construct an improved LSTM concentration prediction network model according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer.

[0045] After obtaining the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer, an LSTM concentration prediction improved network model can be constructed according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer; the LSTM concentration prediction network model is obtained by training the LSTM network according to the data features of the dust concentration historical sequence; it should be noted that the construction of the LSTM concentration prediction network model belongs to the prior art, and the specific steps are not repeated. The first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer are added to the LSTM concentration prediction network model and optimized and learned, different dust concentration historical sequences are put into different hidden layers to learn and judge the causes of dust concentration changes, and the output value of the prediction model is continuously adjusted according to the inducement to improve the accuracy of the prediction model, and finally the LSTM concentration prediction improved network model is obtained; the input of the LSTM concentration prediction improved network model is the dust concentration sequence of the current period, and the output is the predicted concentration value at the future moment.

[0046] The concentration prediction module S4 is used to predict the dust concentration sequence according to the LSTM concentration prediction improved network model to obtain the predicted concentration value at the future moment; and adjust the dry fog dust reduction system according to the concentration prediction value.

[0047] After obtaining an improved LSTM concentration prediction network model with higher prediction accuracy, the dust concentration sequence can be predicted according to the improved LSTM concentration prediction network model to obtain the predicted concentration value at the future moment; specifically including: inputting the dust concentration sequence into the improved LSTM concentration prediction network model; for the target data point at the target collection moment in the dust concentration sequence, the target collection moment is any moment in the dust concentration sequence; placing the target data point in the first feature classification hidden layer, when the target data point satisfies the first judgment model and exceeds the preset first threshold, the preset first threshold is 0.7 in the embodiment of the present invention, which means that the dust concentration change of the target data point is similar to the cause of the dust concentration change at the corresponding collection moment in history, and the target data point is used as a normal change data point; if the historical data point corresponding to the target collection moment does not meet the precondition, it does not participate in the analysis of the first feature classification hidden layer, which means that the dust concentration change corresponding to the target collection moment in the dust concentration history sequence is relatively normal, and the inducement analysis of the first feature classification hidden layer cannot be performed on the target data point.

[0048] Furthermore, when the target data point is not a normal change data point; the target data point is placed in the second feature classification hidden layer. When the target data point meets the first classification condition, and the second classification model and the third classification model both exceed the preset second threshold, the second threshold is preset to 0.8 in the embodiment of the present invention, and the target data point is used as an abnormal mutation data point. At this time, the cause of the abnormal mutation data point is the abnormal change in dust concentration caused by human activities or the air environment itself. However, there are certain differences in the corresponding prediction results of the two, so it is necessary to continue the analysis through the second feature classification additional hidden layer.

[0049] When the target data point is an abnormal mutation data point, other data points in the preset neighborhood of the target data point are placed in the second feature classification additional hidden layer; in the embodiment of the present invention, other data points in the preset neighborhood are data points corresponding to the first 20 acquisition moments of the target data point. The number of other data points that meet the second classification model and the third classification model and exceed the preset second threshold is counted as the neighborhood quantity value; the ratio of the neighborhood quantity value to the number of other data points in the preset neighborhood is calculated to obtain the neighborhood proportion; when the neighborhood quantity value is larger, the neighborhood proportion is larger, which means that the time duration of abnormal changes in dust concentration is longer, and it is more likely to be caused by changes in the air environment itself; on the contrary, the smaller the neighborhood proportion is, the more likely it is that the abnormal change in dust concentration is caused by human activities, and the shorter the duration. When the neighborhood proportion exceeds the preset proportion threshold, the target data point is regarded as an environmental abnormal data point, otherwise the target data point is regarded as a human interference data point; in the embodiment of the present invention, the preset proportion threshold is 0.5. If the target data point does not meet the conditions for entering the first feature classification hidden layer and the second feature classification hidden layer, the target data point is treated as an ordinary data point. The dust concentration of the ordinary data point is low, and there is no need for the cause analysis of the dust concentration change.

[0050] In the LSTM concentration prediction improvement network model, the output value is improved according to the data point type and inducement of the target data point to obtain the predicted concentration value at the future moment. So far, the LSTM concentration prediction improvement network model is used to determine the inducement of the dust concentration change and the corresponding data point type of the data point in the dust concentration sequence, and the predicted value is adjusted to improve the prediction accuracy. Furthermore, the dry fog dust suppression system is adjusted according to the concentration prediction value, and the implementer can determine the spraying amount corresponding to different concentration prediction values.

[0051] In summary, the embodiment of the present invention provides a dry fog dust reduction system for online monitoring of dust concentration; a first feature classification hidden layer is constructed according to the difference in concentration change characteristic values ​​between the dust concentration sequence and the dust concentration historical sequence; a first classification condition is constructed according to the data distribution characteristics in the dust concentration sequence; a second classification model and a third classification model are constructed according to the data difference characteristics and data discrete characteristics at the same time in the dust concentration sequence and the dust concentration historical sequence. A second feature classification hidden layer and a second feature classification additional hidden layer are constructed according to the first classification condition, the second classification model and the third classification model; the present invention constructs an LSTM concentration prediction improved network model according to different hidden layers to predict the dust concentration sequence, obtain the predicted concentration value at the future moment and adjust the dry fog dust reduction system; the data prediction accuracy and dust reduction effect are improved.

[0052] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0053] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A dust concentration online monitoring dry fog dust reduction system, characterized in that: The system includes the following modules: A data acquisition module, used to acquire a dust concentration sequence for monitoring dust concentration in a current period, and at least two dust concentration history sequences; A hidden layer construction module is used to obtain concentration change characteristic values ​​according to data difference characteristics of adjacent data points in a dust concentration sequence and a dust concentration history sequence; construct a first feature classification hidden layer according to difference characteristics and data distribution characteristics of concentration change characteristic values ​​at the same time between the dust concentration sequence and the dust concentration history sequence; construct a first classification condition according to the data distribution characteristics in the dust concentration sequence; construct a second classification model according to data difference characteristics at the same time in the dust concentration sequence and the dust concentration history sequence; construct a third classification model according to discrete characteristics of data at the same time in the dust concentration sequence and the dust concentration history sequence; construct a second feature classification hidden layer according to the first classification condition, the second classification model and the third classification model; and construct a second feature classification additional hidden layer according to the second classification model and the third classification model; A prediction network construction module, used for constructing an LSTM concentration prediction improved network model according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer; The concentration prediction module is used to predict the dust concentration sequence according to the LSTM concentration prediction improved network model to obtain the predicted concentration value at the future moment; and adjust the dry fog dust reduction system according to the concentration prediction value; The step of constructing a first feature classification hidden layer according to the difference characteristics of the concentration change characteristic values ​​at the same time between the dust concentration sequence and the dust concentration history sequence and the data distribution characteristics comprises: For any collection moment, the data point values ​​of all historical dust concentration history sequences at the said any collection moment are greater than the preset concentration threshold as a precondition; on the premise of satisfying the precondition, the average value of the absolute value of the difference between the dust concentration sequence and the said concentration change characteristic value of each dust concentration history sequence at the said any collection moment is calculated to obtain the change difference characteristic value of the dust concentration sequence at the said any collection moment; the said change difference characteristic value is negatively correlated to obtain the first judgment model; the first feature classification hidden layer is constructed according to the said precondition and the said first judgment model; The step of constructing a first classification condition according to the data distribution characteristics in the dust concentration sequence comprises: Taking the value of the data point of the dust concentration sequence at any collection time as greater than the preset concentration threshold as the first classification condition; The step of constructing a second classification model according to the data difference characteristics of the dust concentration sequence and the dust concentration history sequence at the same time comprises: Calculate the average value of the data point values ​​of the dust concentration historical sequence at the arbitrary collection time to obtain the historical concentration average value at the arbitrary collection time; calculate the absolute value of the difference between the data point values ​​of the dust concentration sequence at the arbitrary collection time and the historical concentration average value and normalize them to obtain the second classification model; The step of constructing a third classification model according to the discrete features of data at the same time in the dust concentration sequence and the dust concentration history sequence comprises: The absolute value of the difference between any two data points of the dust concentration sequence and the dust concentration history sequence at any collection time is calculated to obtain a concentration difference characteristic value; the variance of the concentration difference characteristic value is calculated and normalized to obtain the third classification model.

2. The dust concentration online monitoring dry fog dust reduction system according to claim 1 is characterized in that: The step of obtaining the concentration change characteristic value according to the data difference characteristics of adjacent data points in the dust concentration sequence and the dust concentration history sequence comprises: For any data point in the dust concentration sequence and the dust concentration history sequence, the absolute value of the difference between the arbitrary data point and other data points at adjacent moments is calculated to obtain a concentration change characteristic value of the arbitrary data point.

3. The dust concentration online monitoring dry fog dust reduction system according to claim 1 is characterized in that: The step of constructing an improved LSTM concentration prediction network model according to the dust concentration historical sequence, the first feature classification hidden layer, the second feature classification hidden layer, and the second feature classification additional hidden layer comprises: According to the data features of the dust concentration historical sequence, the LSTM network is trained to obtain the LSTM concentration prediction network model; the first feature classification hidden layer, the second feature classification hidden layer and the second feature classification additional hidden layer are added to the LSTM concentration prediction network model and optimized and learned to obtain the LSTM concentration prediction improved network model; the input of the LSTM concentration prediction improved network model is the dust concentration sequence of the current time period, and the output is the predicted concentration value at the future moment.

4. The dust concentration online monitoring dry fog dust reduction system according to claim 3 is characterized in that: The step of predicting the dust concentration sequence according to the LSTM concentration prediction improved network model to obtain the predicted concentration value at the future moment includes: The dust concentration sequence is input into the LSTM concentration prediction improved network model; for the target data point at the target collection time in the dust concentration sequence, the target data point is placed in the first feature classification hidden layer, and when the target data point satisfies the first judgment model and exceeds the preset first threshold, the target data point is regarded as a normal change data point; when the target data point is not a normal change data point; the target data point is placed in the second feature classification hidden layer, and when the target data point satisfies the first classification condition, and the second classification model and the third classification model both exceed the preset second threshold, the target data point is regarded as an abnormal mutation data point; When the target data point is an abnormal mutation data point, other data points in a preset neighborhood of the target data point are placed in a second feature classification additional hidden layer; the number of other data points that satisfy both the second classification model and the third classification model and exceed a preset second threshold is counted as a neighborhood quantity value; the ratio of the neighborhood quantity value to the number of other data points in a preset neighborhood is calculated to obtain a neighborhood proportion; when the neighborhood proportion exceeds a preset proportion threshold, the target data point is regarded as an environmental abnormality data point, otherwise the target data point is regarded as a human interference data point; In the LSTM concentration prediction improvement network model, the output value is adjusted and improved according to the type of the target data point to obtain the predicted concentration value at the future moment.

Images