A method for predicting the moisture content of phosphogypsum based on multi-dimensional features

Abstract

The application discloses a kind of phosphogypsum moisture content prediction methods based on multidimensional features, comprising: obtaining the appearance state data and internal state data of phosphogypsum;According to appearance state data, extract appearance feature dataset, including appearance surface feature, appearance humidity anomaly feature and appearance texture anomaly data;According to internal state data, extract internal feature dataset including structure anomaly feature data, distribution anomaly feature data and pore anomaly feature data;Wherein, each type of data in appearance feature dataset and internal feature dataset includes corresponding moisture content;Construct phosphogypsum moisture content prediction calculation model, calculate the optimal phosphogypsum moisture content according to appearance feature dataset and internal feature dataset.It can comprehensively consider the appearance and internal information of phosphogypsum according to the above technical solution, adapt to different complexity of phosphogypsum, realize comprehensive and accurate prediction of the moisture content of phosphogypsum.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of environmental engineering, and in particular, to a phosphogypsum water content prediction method based on multi-dimensional features. BACKGROUND

[0002] With the development of industrialization and urbanization, the amount of phosphogypsum is increasing, and the water content of phosphogypsum is one of the key parameters affecting its treatment, disposal and resource utilization. Accurate prediction of the water content of phosphogypsum is of great significance for optimizing the treatment process, improving resource recovery efficiency and reducing environmental pollution risk. Most traditional methods only focus on a certain characteristic of phosphogypsum, such as predicting the water content only according to the appearance feature or a certain factor in the internal composition. For example, the water content is estimated only by observing the surface humidity of phosphogypsum, while ignoring the influence of internal pore structure, composition distribution and other factors on water storage and migration. The current single-factor consideration method cannot fully reflect the true situation of the water content of phosphogypsum, resulting in low accuracy of the prediction results.

[0003] Therefore, a technical solution is needed that can comprehensively judge the water content of phosphogypsum from multiple dimensions related to the water content of phosphogypsum, and improve the accuracy of the prediction results. SUMMARY

[0004] To achieve the above-mentioned purpose, the present application provides a phosphogypsum water content prediction method based on multi-dimensional features, comprising:

[0005] Obtaining appearance state data and internal state data of phosphogypsum; wherein the appearance state data includes appearance feature, appearance humidity state and appearance texture state; the internal state data includes internal composition structure, internal water distribution condition and internal pore structure condition;

[0006] Extracting appearance feature data set according to the appearance state data, the types of the appearance feature data set include appearance surface feature, appearance humidity abnormal feature and appearance texture abnormal data; wherein each type of data in the appearance feature data set includes the corresponding water content;

[0007] Extracting internal feature data set according to the internal state data, the types of the internal feature data set include structure abnormal feature, distribution abnormal feature and pore abnormal feature; wherein each type of data in the internal feature data set includes the corresponding water content;

[0008] Constructing a phosphogypsum water content prediction calculation model, and calculating the optimal phosphogypsum water content according to the appearance feature data set and the internal feature data set;

[0009] Wherein, the construction of the phosphogypsum water content prediction calculation model comprises:

[0010] Selecting one or more types of feature data from the appearance feature dataset and the internal feature dataset respectively to combine, to form N groups of M-dimensional combination results; wherein N and M are natural numbers;

[0011] Obtaining water content rate data from the N groups of combination results respectively, the water content rate data including prediction type and average water content rate; the prediction type including water content rate prediction result or pending prediction result;

[0012] Obtaining the water content rate prediction result of phosphogypsum according to the N groups of water content rate data.

[0013] Among them, the N groups of M-dimensional combination results refer to selecting one type of feature data or multiple types of feature data from the appearance feature dataset and the internal feature dataset respectively to combine, including:

[0014] Selecting one type from the appearance feature dataset and the internal feature dataset respectively to combine to form one-dimensional combination results, at this time M=1;

[0015] Selecting two types of data from the appearance feature dataset and the internal feature dataset respectively to combine to form two-dimensional combination results, at this time M=2;

[0016] Combining all the appearance feature dataset and the internal feature dataset to form the third combination result, at this time M=3.

[0017] Further, obtaining water content rate data from the N groups of combination results respectively includes:

[0018] Extracting data of X types of abnormal feature points from each group of combination results of the N groups to form a coincident feature group; determining a screening index from the coincident feature group, and determining the average water content rate and the prediction type according to the screening index, wherein the prediction type includes: prediction result or pending prediction result; the screening index includes: water content rate amplitude value, data change relationship feature and water content rate compliance rate.

[0019] Among them, the features of the coincident feature group are determined according to the dimension, including:

[0020] When M=1, the data of X types of abnormal feature points is extracted from the combination results as the coincident feature group, X=2;

[0021] When M=2, the data of X types of abnormal feature points is extracted from the combination results as the coincident feature group, X=3.

[0022] Among them, the water content rate amplitude value represents the change range or fluctuation degree of the water content rate of phosphogypsum under the feature, which is used for the coincident feature group of X=2;

[0023] Determining the average water content rate and the prediction type according to the water content rate amplitude value includes the following steps:

[0024] counting the water content of one type of features in the coincident feature group to generate a first water content amplitude value;

[0025] counting the water content of another type of features to generate a second water content amplitude value;

[0026] matching the first water content amplitude value with the second water content amplitude value to obtain a matching degree: if the amplitude variation curve matching degree is within a preset amplitude value matching degree interval value, defining the prediction type of the average water content of the combined result as a prediction result; otherwise, defining the prediction type of the average water content of the one-dimensional combined result as a pending prediction result.

[0027] The data variation relationship feature refers to the data variation relationship between the distribution proportion of each type of abnormal feature and the abnormal water content data in the coincident feature group, including a direct proportion relationship and an inverse proportion relationship.

[0028] The data variation relationship feature is used for the coincident feature group of X=3.

[0029] Determining the average water content and the prediction type according to the data variation relationship feature includes the following steps:

[0030] Judging the data direct proportion variation relationship between the abnormal feature distribution proportion and the abnormal water content data, if it is a direct proportion relationship, defining the prediction type of the average water content of the combined result as a prediction result, otherwise, defining it as a pending prediction result.

[0031] The water content compliance rate refers to the ratio of judging the water content of each group of data to reach the water content threshold, which is used for the case of X>3.

[0032] Determining the average water content and the prediction type according to the water content compliance rate includes the following steps:

[0033] Setting a water content threshold, which is determined by comprehensively considering the type of phosphogypsum, processing requirements, and environmental standards.

[0034] Checking the water content data of each group in the coincident feature group one by one.

[0035] If the water content of each feature in the coincident feature group exceeds the preset water content threshold, the prediction type of the average water content of the combined result is defined as a prediction result, if the water content in the coincident feature group does not exceed the preset water content threshold, the prediction type of the average water content of the combined result is defined as a pending prediction result.

[0036] Further, obtaining the phosphogypsum water content prediction result according to N groups of water content data includes:

[0037] If there is only one group of data in the N groups of moisture content data whose prediction type is prediction result, the average moisture content corresponding to the prediction type is defined as the phosphogypsum moisture content prediction result.

[0038] If there is more than one group of data with the prediction type "prediction result" in the N groups of moisture content data, calculate the average moisture content of at least two groups corresponding to the prediction type, and define the average moisture content as the phosphogypsum moisture content prediction result.

[0039] If all the prediction types in the N groups of moisture content data are undetermined prediction results, the average value of all groups of moisture content is then taken as the predicted moisture content of phosphogypsum.

[0040] Furthermore, when selecting one class from the appearance feature dataset and one class from the internal feature dataset to combine them into a one-dimensional combination result, the combination method is as follows:

[0041] ,in, For matching degree, For appearance feature data to be concentrated Data at any given time For internal feature data to be concentrated Data at any given time This represents the number of sampling points;

[0042] The largest first match in the appearance feature dataset and the internal feature dataset. The corresponding data are combined to obtain a one-dimensional combination result.

[0043] Furthermore, when selecting two types of data from both the appearance feature dataset and the internal feature dataset to combine them into a two-dimensional combination result, the combination method is as follows:

[0044] ,in, For matching degree, This indicates the proportion of abnormal features in the appearance feature dataset. This indicates the proportion of outlier features in the internal feature dataset. express The mean, express The mean, Indicates sample size;

[0045] Select the matching degree from the appearance feature dataset and the internal feature dataset. The M types of data that meet the requirements are combined to obtain a two-dimensional combination result.

[0046] According to the application, the appearance and internal aspects of the phosphogypsum can be comprehensively considered, multi-dimensional information such as appearance features, humidity conditions, texture conditions, internal component structures, moisture distribution, and pore structures can be fused and analyzed, information in the data can be mined from different angles, different complexity of the phosphogypsum can be adapted, and the water content of the phosphogypsum can be comprehensively and accurately predicted. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 A multi-dimensional feature-based phosphogypsum water content prediction method provided according to an embodiment of the application is shown in a step diagram. DETAILED DESCRIPTION

[0048] The application comprehensively considers the appearance and internal aspects of the phosphogypsum, fuses and analyzes multi-dimensional information such as appearance features, humidity conditions, texture conditions, internal component structures, moisture distribution, and pore structures, the multi-dimensional information fusion can more comprehensively reflect the characteristics of the phosphogypsum and the complex relationship between the characteristics and the water content, and the prediction accuracy can be improved. In the fusion analysis process, a multi-level combination analysis method is used, combination analysis is performed on one type of data, two types of data, and each type of data, respectively, information in the data can be mined from different angles, different complexity of the phosphogypsum can be adapted, and different calculation methods are used to determine the phosphogypsum water content prediction result according to the water content prediction results under various combinations.

[0049] The specific implementation of the application will be described in detail below with reference to the accompanying drawings.

[0050] Figure 1 A step of the phosphogypsum water content prediction method is provided, as shown in the figure, including:

[0051] Step S100: Obtain appearance state data and internal state data of the phosphogypsum.

[0052] The application first extracts features of the appearance characteristics of the phosphogypsum, converts various information contained in the appearance of the phosphogypsum, such as shape, color, texture, and the like, into a numerical form that can be processed by a computer, and obtains appearance state data. After obtaining the appearance state data, a preset appearance feature threshold is set: the threshold can be determined according to a large amount of historical data, experimental results, or experience, and is used to measure the importance or correlation of the appearance features. Each feature data in the appearance extraction feature result is compared with the threshold, and feature data greater than or equal to the preset appearance feature threshold is selected, and these selected data constitute the appearance feature data. In this way, appearance feature data that has little or no relevance to subsequent analysis such as phosphogypsum water content prediction is excluded.

[0053] Therefore, the appearance state data refers to visual characteristics closely related to the water content that can be determined through the surface visual performance of the phosphogypsum, including appearance features, appearance humidity state, and appearance texture state.

[0054] The appearance features include color values, texture roughness values, and shape values, which are quantified data reflecting the surface visual features of the phosphogypsum.

[0055] Color values: For example, the RGB value of the surface of a certain phosphogypsum sample is (200, 180, 160), which can be used to represent the color feature of the surface. Different colors may be associated with the water content, for example, the darker the color, the higher the water content.

[0056] Texture roughness values: The texture roughness of the surface of a certain phosphogypsum is measured by an instrument, for example, 5.2 μm. This value reflects the roughness of the surface, and different roughness may mean different water adhesion and storage capabilities.

[0057] Shape values: The aspect ratio and circularity of a certain phosphogypsum (for example, the aspect ratio is 3:1, and the circularity is 0.6). These values describe the shape of the appearance, and the shape affects the distribution and evaporation of water on the surface.

[0058] The present application also extracts data features from the internal composition structure information of the original phosphogypsum and the current phosphogypsum. The extraction process uses appropriate technical means (such as spectral analysis, tomographic imaging, etc., depending on the actual way of obtaining internal composition structure information) to convert the internal composition structure information into numerical feature data. For example, if the internal composition of the phosphogypsum contains multiple chemical substances, the types and contents of each substance are determined by a specific analysis method and converted into data to obtain original internal structure feature data and current internal structure feature data, thereby converting complex internal composition structure information into internal state data that can be processed and analyzed by a computer.

[0059] The internal state data refers to internal characteristics closely related to the water content that can be determined through the internal structure and composition of the phosphogypsum, including internal composition structure, internal water distribution, and internal pore structure.

[0060] According to the above rules, feature sets can be extracted from the appearance state data and the internal state data in subsequent steps.

[0061] Step S111: Extracting appearance feature data sets according to the appearance state data;

[0062] This step includes extracting data features from the appearance feature information of the phosphogypsum to obtain appearance surface feature data, identifying abnormal areas from the appearance humidity state information of the phosphogypsum to obtain appearance humidity abnormal data, and identifying abnormal areas from the appearance texture state information of the phosphogypsum to obtain appearance texture abnormal data.

[0063] The appearance humidity abnormal data is obtained by identifying abnormal areas of the appearance humidity information of the phosphogypsum, and refers to area-related data in which the appearance humidity is obviously different from the normal case.

[0064] For example, in a pile of phosphogypsum, the surface of most areas is dry, and the humidity value is between 5% and 10%, but there is an area in which the humidity value reaches more than 30% due to local water seepage. The position and humidity value data of this area are the appearance humidity abnormal data.

[0065] The appearance texture abnormal data is obtained by identifying abnormal areas of the appearance texture information of the phosphogypsum, and refers to area-related data in which the appearance texture is obviously different from the normal texture.

[0066] The appearance texture abnormal data is obtained by identifying abnormal areas of the appearance texture information of the phosphogypsum, and refers to area-related data in which the appearance texture is obviously different from the normal texture.

[0067] In this step, the above appearance state data of the phosphogypsum is subjected to feature extraction to obtain appearance extraction feature results; when feature extraction is performed through appearance feature information, the appearance feature calculation formula is used to calculate the appearance surface feature data ; wherein, is the feature data in the appearance extraction feature results, is a preset appearance feature threshold.

[0068] Finally, the appearance surface feature data, the appearance humidity abnormal data, and the appearance texture abnormal data are combined to form an appearance feature data set; each type of data in the appearance feature data set can be used to calculate the water content rate alone; therefore, each type of data in the appearance feature data set includes the corresponding water content rate.

[0069] Step S112: extracting an internal feature data set according to the internal state data;

[0070] The types of the internal feature data set include structure abnormal features, distribution abnormal features, and pore abnormal features.

[0071] When the internal state data is processed, the original internal structure feature data and the current internal structure feature data are compared. In the comparison process, differences (such as difference calculation, similarity algorithm, etc.) in composition types, content proportions, structure arrangements, etc. are calculated to find inconsistent or changed parts. These difference parts constitute the structure abnormal feature data. In this way, changes in the internal composition structure of the phosphogypsum relative to the original state can be identified.

[0072] The structural abnormality characteristic data is obtained by processing the internal composition structure information of the phosphogypsum, and reflects the characteristic data of the abnormal change of the internal composition structure relative to the original state.

[0073] For example, the original internal composition of a certain phosphogypsum contains 30% of component A, 50% of component B, and 20% of component C, and the distribution is uniform. At present, the proportion of component A in the phosphogypsum is reduced to 10%, the proportion of component B is 60%, and the proportion of component C is 30%, and local aggregation phenomenon occurs. The quantitative data reflecting the change of the composition ratio and distribution obtained by calculation is the structural abnormality characteristic data.

[0074] The distribution abnormality characteristic data is obtained by region matching the internal moisture distribution information of the phosphogypsum, and refers to the region related data of the abnormal internal moisture distribution.

[0075] The distribution abnormality characteristic data is formed by extracting the data of local excessive moisture, excessive low moisture, and abnormal distribution form from the internal moisture distribution information. The water content rate of the abnormal region deviates from the normal level, and the abnormal degree of the abnormal data reflects the overall water content rate. If the difference is large and the abnormal region is wide, it indicates that the overall water content rate deviates from the normal. For example, under normal circumstances, the internal moisture distribution of the phosphogypsum should present a certain gradient or uniform distribution. If the moisture content of a certain region suddenly increases and is much higher than the average moisture content of the surrounding region, the position, moisture content value, and distribution range data of the region are the distribution abnormality characteristic data.

[0076] The pore abnormality characteristic data is obtained by region matching the internal pore structure information of the phosphogypsum, and refers to the region related data of the abnormal internal pore structure.

[0077] Since the pore is a water storage and migration channel, the abnormal data of the pore quantity, size, and connectivity are extracted from the internal pore structure information to form the pore abnormality characteristic data. When the pore is abnormal, the water content rate is directly affected. When the pore increases in size, the water storage capacity is strong, that is, the overall water content rate of the corresponding region is easy to be high. For example, the internal pore of a certain phosphogypsum is usually between 10-50 μm in size and has good connectivity. If the pore size in a certain region suddenly becomes 100 μm or more, and the connectivity becomes poor, the pore size, quantity, and connectivity parameter data of the region are the pore abnormality characteristic data.

[0078] The extraction of the internal feature dataset specifically includes: processing the internal component structure information of phosphogypsum to obtain structural anomaly feature data; performing region matching on the internal moisture distribution information of phosphogypsum to obtain distribution anomaly feature data; and performing region matching on the internal pore structure information of phosphogypsum to obtain pore anomaly feature data. The structural anomaly feature data, distribution anomaly feature data, and pore anomaly feature data are combined to form the internal feature dataset. The moisture content of each type of data in the internal feature dataset can be calculated independently; therefore, each type of data in the internal feature dataset includes the corresponding moisture content.

[0079] The internal composition and structural information of phosphogypsum is processed to obtain structural anomaly feature data, specifically including the following steps:

[0080] 1) Extract data features from the original internal composition and structure information of the original phosphogypsum to obtain the original internal structure feature data, and extract data features from the current internal composition and structure information of the current phosphogypsum to obtain the current internal structure feature data;

[0081] 2) Calculation formula based on abnormal data Calculated structural anomaly characteristic data ; wherein, the For the first Original internal structural feature data of each component For the first Current internal structural feature data of each component, Indicates the total number of components. This represents the structural difference weighting coefficient.

[0082] The process involves region matching to obtain distribution anomaly feature data from the internal moisture distribution information of phosphogypsum, and region matching to obtain pore anomaly feature data from the internal pore structure information. This includes the following steps:

[0083] 1) Extracting information on the internal moisture distribution of phosphogypsum, including data on abnormal moisture distribution, yields abnormal distribution characteristic data;

[0084] 2) Extract the internal pore structure information of phosphogypsum, including data on internal pore anomalies, to obtain pore anomaly feature data.

[0085] Step S120: Construct a phosphogypsum moisture content prediction calculation model based on the appearance feature dataset and the internal feature dataset, and calculate the optimal phosphogypsum moisture content.

[0086] Specifically, the steps for constructing a calculation model for predicting the moisture content of phosphogypsum are as follows:

[0087] Step S121: Select one or more types of feature data from the appearance feature dataset and the internal feature dataset respectively, and combine them to form N sets of M-dimensional combination results; where N and M are natural numbers;

[0088] When choosing to combine one or more types of feature data, you can select one or more types of feature data from each of the two datasets (appearance feature dataset and internal feature dataset) for combination. Specifically, this includes: selecting one type of feature data from each dataset to form combination one; selecting two types of feature data from each dataset to form combination two; and combining all data from both datasets to form combination three. In this case, N=3; the dimension is determined by the number of types of feature data selected.

[0089] 1) Select one class from each of the appearance feature dataset and the internal feature dataset to combine them into a one-dimensional combination result. The combination matching method is as follows:

[0090] ,in, For matching degree, For appearance feature data to be concentrated Data at any given time For internal feature data to be concentrated Data at any given time This represents the number of sampling points;

[0091] The largest first match in the appearance feature dataset and the internal feature dataset. The corresponding data are combined to obtain a one-dimensional combination result, which has a dimension of one, i.e., M=1.

[0092] 2) Select two classes of data from both the appearance feature dataset and the internal feature dataset, and combine them to form a two-dimensional combination result. In this case, M=2, and the matching method is:

[0093] ,in, For matching degree, This indicates the proportion of abnormal features in the appearance feature dataset. This indicates the proportion of outlier features in the internal feature dataset. express The mean, express The mean, Indicates sample size;

[0094] Select the matching degree from the appearance feature dataset and the internal feature dataset. The M types of data that meet the requirements are combined to obtain a two-dimensional combination result.

[0095] The present application provides an embodiment: two types of data are selected from the appearance feature data set, such as appearance surface feature data and appearance humidity anomaly data; at the same time, two types of data are also selected from the internal feature data set, such as structure anomaly feature data and distribution anomaly feature data. The two types of data from the appearance and the internal are combined to obtain a two-dimensional combination result. This step is equivalent to integrating more different angles and different properties of the feature information about phosphogypsum to form a data set containing more information.

[0096] 3) All the two data sets are combined to form a third combination result, and M=3 at this time. In this step, the appearance feature data set (including appearance surface feature data, appearance humidity anomaly data, and appearance texture anomaly data) is combined with the internal feature data set (including structure anomaly feature data, distribution anomaly feature data, and pore anomaly feature data) to obtain a three-dimensional combination result with M=3.

[0097] In each combination method, there is an appearance feature data set containing moisture content and an internal feature data set containing moisture content.

[0098] Step S122: obtaining moisture content data of each group from the N combination results, including prediction type and average moisture content; the prediction type includes moisture content prediction result or pending prediction result;

[0099] The process of obtaining moisture content data of each group from the N combination results includes:

[0100] 1) Extracting X types of abnormal feature point coincident data from each combination result of the N groups to form a coincident feature group; wherein X<=2M.

[0101] The coincidence of the abnormal feature points in the coincident feature group means that the X types of features selected from the appearance and the internal have consistency in some aspects, such as possible coincidence in the abnormal area reflecting the moisture distribution; there is no consistent feature to form a difference feature group, which represents the difference between these types of features in reflecting the moisture content related information. Through such separation, the similarities and differences between the appearance and internal features in the moisture content correlation can be more clearly understood, and more targeted data can be provided for subsequent in-depth analysis.

[0102] The features of the coincident feature group are determined according to the dimension.

[0103] For example: when M=1, two types of abnormal feature point coincident data are extracted from the combination result as the coincident feature group; when M=2, three types of abnormal feature point coincident data are extracted from the combination result as the coincident feature group.

[0104] 2) determining a screening index from the coincident feature group, determining the average moisture content and the prediction type according to the screening index, and the prediction type including: a prediction result or a pending prediction result; the pending prediction result indicates that the moisture content cannot be directly used as a reliable moisture content prediction value and needs to be combined with other analysis for further judgment.

[0105] Specifically, the screening index supports the moisture content amplitude value, the data change relationship feature, and the moisture content compliance rate.

[0106] (1) The moisture content amplitude value represents the change range or fluctuation degree of the moisture content of phosphogypsum under the feature. For example, the difference between the maximum value and the minimum value of the moisture content under a certain appearance feature within a certain time or condition is the moisture content amplitude value of the appearance feature. By statistical amplitude value, the influence degree of different features on the moisture content change can be quantified, providing specific numerical basis for subsequent judgment of the relationship between features.

[0107] The moisture content amplitude value is used for the case where X=2: two types of abnormal feature points coincide in the coincident feature group; in specific implementation, the moisture content of one type of feature in the coincident feature group is counted to generate a first moisture content amplitude value; the moisture content of another type of feature is counted to generate a second moisture content amplitude value; the first moisture content amplitude value and the second moisture content amplitude value are matched to obtain a matching degree: if the amplitude variation curve matching degree is within the preset amplitude value matching degree interval value, i.e. the matching degree is within the specified threshold, it means that the two types of features have high synchronicity and correlation in the moisture content change, which means that the moisture content change trend observed from the two features is relatively consistent, and the prediction type of the average moisture content of the combined result is a prediction result; if the amplitude variation curve matching degree is not within the preset amplitude value matching degree interval value, i.e. the matching degree is not within the specified threshold, it means that the correlation between the two types of features in the moisture content change is not clear, there may be other interference factors or the relationship between the two is relatively complex, and the prediction type of the average moisture content of the combined result is a pending prediction result.

[0108] (2) The data change relationship feature refers to the data change relationship between the distribution proportion of each type of abnormal feature and the abnormal moisture content data in the coincident feature group, including direct proportion and inverse proportion (direct proportion means that the higher the distribution proportion of abnormal features, the higher the abnormal moisture content). If the data change relationship is direct proportion, the prediction type of the average moisture content in the combined result is defined as a prediction result; if the data change relationship is inverse proportion, the prediction type of the average moisture content in the combined result is defined as a pending prediction result.

[0109] When analyzing the data change relationship feature, abnormal moisture content needs to be defined: when there are abnormal feature data in the appearance or internal state information of phosphogypsum (such as abnormal area of appearance humidity, abnormal area of internal moisture distribution, abnormal area of pore structure, etc.), the moisture content associated with these abnormal features is defined as abnormal moisture content.

[0110] If it is determined that X = 3, that is, the coincident abnormal feature points of the coincident feature group are three categories, the distribution proportion of three categories of abnormal features (selected from the appearance humidity abnormal feature, appearance texture abnormal data, structure abnormal feature, distribution abnormal feature and pore abnormal feature) and abnormal moisture content data are obtained. The abnormal feature distribution proportion represents the proportion or frequency of the feature in the entire feature group, and the abnormal moisture content data directly reflects the moisture content related to the moisture content. The change relationship between the abnormal feature distribution proportion and the abnormal moisture content data is determined. For example, when the abnormal feature distribution proportion increases, it is observed whether the abnormal moisture content increases or decreases.

[0111] If it is found that the abnormal feature distribution proportion and the abnormal moisture content data present a proportional change relationship (that is, the higher the abnormal feature distribution proportion, the higher the abnormal moisture content), it indicates that there is a positive and relatively clear correlation between these features and the moisture content. Based on this, the average moisture content in the two-dimensional combination result is marked as the moisture content prediction result, and it is considered that this result can more reliably reflect the moisture content of phosphogypsum. On the contrary, if they present an inverse relationship, it indicates that the relationship between these features and the moisture content is relatively complex, and the moisture content cannot be accurately predicted based on these features. At this time, the average moisture content in the two-dimensional combination result is marked as a pending prediction result, that is, the result still needs to be further combined with other information for comprehensive judgment.

[0112] (Three) moisture content compliance rate refers to the ratio of the moisture content of each group of data reaching the moisture content threshold, which is used for the case where X > 3: for example, two data sets are all combined to form a third combination result.

[0113] When the moisture content compliance rate is used, a moisture content threshold is first set, which is determined by considering the type of phosphogypsum, processing requirements, environmental standards and other factors. Then the moisture content data in the coincident feature group is checked one by one.

[0114] If the moisture content of each feature in the coincident feature group exceeds the preset moisture content threshold, the prediction type of the average moisture content in the combination result is defined as the prediction result. If the moisture content in the coincident feature group does not exceed the preset moisture content threshold, the prediction type of the average moisture content in the combination result is defined as the pending prediction result.

[0115] Step S123: obtaining the phosphogypsum moisture content prediction result according to N groups of moisture content data.

[0116] If the data with the prediction type of moisture content prediction result in the N groups of moisture content data is a group, it indicates that the data has relatively clear prediction information, and the average moisture content corresponding to the prediction type is defined as the phosphogypsum moisture content prediction result;

[0117] If the data of the N groups of moisture content data with the prediction type of moisture content prediction result is greater than one group, in order to more accurately reflect the moisture content of phosphogypsum, at least two groups of average moisture content corresponding to the prediction type are calculated to generate an average value of the average moisture content, and the average moisture content is defined as the phosphogypsum moisture content prediction result;

[0118] If the prediction type of the N groups of moisture content data is all pending prediction result, that is, the data of the moisture content prediction result is 0, it is indicated that there is no directly available prediction result; all group moisture contents are averaged to obtain an average value, which is defined as the phosphogypsum moisture content prediction result.

[0119] Select the same kind of phosphogypsum as the test object, and respectively use the traditional drying and weighing method (traditional method) and the above phosphogypsum moisture content prediction method based on machine learning (patent method) to test the moisture content, and the comparison results of 5 groups of parallel experiments are shown in the following table:

[0120]

[0121] The present application comprehensively considers the appearance and internal information of phosphogypsum, and fuses and analyzes multi-dimensional information such as appearance features, humidity conditions, texture conditions, internal component structures, moisture distribution, pore structures and the like. The multi-dimensional information fusion can more comprehensively reflect the characteristics of phosphogypsum and the complex relationship between the characteristics and the moisture content, and can improve the prediction accuracy from the aspect of data embodiment. In the multi-dimensional data analysis process, a multi-level combination analysis method is used to perform combination analysis from one type of data, two types of data to each type of data. The multi-level analysis method can mine information in the data from different angles, adapt to phosphogypsum with different complexity, and determine the phosphogypsum moisture content prediction result by using different calculation methods according to the existence of different combinations of moisture content prediction results. The flexible result processing mechanism can adapt to various data situations, whether there is a clear prediction result or multiple pending prediction results need to be considered comprehensively, a relatively reasonable moisture content prediction value can be given, and the effectiveness and practicability of the method under different conditions are ensured, so that the present application can comprehensively and accurately predict the moisture content of phosphogypsum.

[0122] The above disclosure is only a few specific embodiments of the present application, but the present application is not limited thereto, and any changes that can be thought of by those skilled in the art shall fall within the protection scope of the present application.

Claims

1. A method for predicting the water content of ardealite based on multi-dimensional features, characterized in that, The method comprises the following steps: Obtaining appearance state data and internal state data of the phosphogypsum; wherein the appearance state data comprises appearance features, appearance humidity state and appearance texture state; the internal state data comprises internal component structure, internal moisture distribution condition and internal pore structure condition; Extracting appearance feature data set according to the appearance state data; the types of the appearance feature data set comprise appearance surface features, appearance humidity abnormal features and appearance texture abnormal data; wherein each type of data in the appearance feature data set comprises corresponding water content; Extracting internal feature data set according to the internal state data; the types of the internal feature data set comprise structure abnormal feature data, distribution abnormal feature data and pore abnormal feature data; wherein each type of data in the internal feature data set comprises corresponding water content; Constructing a phosphogypsum water content prediction calculation model, and calculating the optimal phosphogypsum water content according to the appearance feature data set and the internal feature data set; The construction of the phosphogypsum water content prediction calculation model comprises: Selecting one or more types of feature data in the appearance feature data set and the internal feature data set respectively to form N groups of M-dimensional combination results; wherein N and M are natural numbers; the dimension of the combination result is determined by the number of selected feature data types; Obtaining water content data from each group of combination results, comprising: extracting data of X types of abnormal feature points from each group of combination results to form a coincident feature group; determining a screening index from the coincident feature group, and determining water content data according to the screening index; wherein the water content data comprises a prediction type and an average water content; the prediction type comprises a water content prediction result or a pending prediction result; the screening index comprises a water content amplitude value, a data change relationship feature and a water content compliance rate; The features of the coincident feature group are determined according to the dimension, comprising: when M=1, X=2; when M=2, X=3; when M=3, X>3; The water content amplitude value represents the change range or fluctuation degree of the water content of the phosphogypsum, and is used to determine the average water content and the prediction type for the coincident feature group with X=2; the data change relationship feature refers to the data change relationship between the distribution proportion of each type of abnormal feature and the abnormal water content data in the coincident feature group, comprising a direct proportion relationship and an inverse proportion relationship, and is used to determine the average water content and the prediction type for the coincident feature group with X=3; the water content compliance rate refers to the ratio of the water content of each group of data reaching the water content threshold, and is used to determine the average water content and the prediction type for the coincident feature group with X>3; Obtaining a phosphogypsum water content prediction result according to N groups of water content data.

2. The method of claim 1, wherein, The N groups of M-dimensional combination results are obtained by selecting one or more types of feature data in the appearance feature data set and the internal feature data set respectively, comprising: Selecting one type of feature data in the appearance feature data set and the internal feature data set respectively to form one-dimensional combination results, wherein M=1; Selecting two types of feature data in the appearance feature data set and the internal feature data set respectively to form two-dimensional combination results, wherein M=2; The appearance feature dataset and the internal feature dataset are combined to form a third combination result, and M=3. 3.The method according to claim 1, wherein the method for predicting the water content of phosphogypsum is characterized in that, The average water content and the prediction type are determined according to the water content amplitude value, and the method comprises the following steps: The water content of one type of features in the coincident feature group is counted to generate a first water content amplitude value; The water content of another type of features is counted to generate a second water content amplitude value; The first water content amplitude value and the second water content amplitude value are matched to obtain a matching degree: if the amplitude variation curve matching degree is within a preset amplitude value matching degree interval, the prediction type of the average water content of the combination result is defined as a prediction result; otherwise, the prediction type of the average water content of the one-dimensional combination result is defined as a pending prediction result. 4.The method according to claim 1, wherein the method for predicting the water content of phosphogypsum is characterized in that, The average water content and the prediction type are determined according to the data variation relationship feature, and the method comprises the following steps: The proportion of the abnormal feature distribution is judged to have a direct proportional relationship with the data of the abnormal water content, and if the relationship is direct proportional, the prediction type of the average water content of the combination result is defined as a prediction result; otherwise, it is defined as a pending prediction result. 5.The method according to claim 1, wherein the method for predicting the water content of phosphogypsum is characterized in that, The average water content and the prediction type are determined according to the water content compliance rate, and the method comprises the following steps: A water content threshold is set, which is determined by comprehensively considering the type of phosphogypsum, processing requirements, and environmental standards; The water content data of each group in the coincident feature group are checked one by one; If the water content of each feature in the coincident feature group exceeds the preset water content threshold, the prediction type of the average water content of the combination result is defined as a prediction result; if the water content of the coincident feature group does not exceed the preset water content threshold, the prediction type of the average water content of the combination result is defined as a pending prediction result.

6. The method of claim 1, wherein, The method for obtaining the prediction result of the water content of phosphogypsum from the N groups of water content data comprises: If the data with the prediction type of a prediction result in the N groups of water content data is one group, the average water content corresponding to the prediction type is defined as the prediction result of the water content of phosphogypsum; If the data with the prediction type of a prediction result in the N groups of water content data is more than one group, at least two groups of average water content corresponding to the prediction type are calculated to obtain an average water content, and the average water content is defined as the prediction result of the water content of phosphogypsum; If the prediction type of all the N groups of water content data is a pending prediction result, the average value of all the groups of water content is calculated to obtain the prediction result of the water content of phosphogypsum.

7. The method of claim 2, wherein the phosphogypsum water content is predicted by the equation: ###00001### where: W = water content of the phosphogypsum, in percent; and T = temperature of the phosphogypsum, in degrees Fahrenheit. When one type of data is selected from each of the appearance feature dataset and the internal feature dataset to form a one-dimensional combination result, the combination method is as follows: wherein, is a matching degree, is appearance feature data, is data at a time point, is internal feature data, is data at a time point, is a number of sampling points; The maximum first matching degree in the appearance feature dataset and the internal feature dataset is determined The corresponding data are combined to obtain a one-dimensional combination result.

8. The method of claim 2, wherein the phosphogypsum water content is predicted by the equation: ###0001### where: W = water content of the phosphogypsum, in percent; and T = temperature of the phosphogypsum, in degrees Fahrenheit. When two types of data are selected from each of the appearance feature dataset and the internal feature dataset to form a two-dimensional combination result, the combination method is as follows: wherein, is the matching degree, represents the proportion of abnormal features in the appearance feature dataset, represents the proportion of abnormal features in the internal feature dataset, represents the mean value of, represents the mean value of, represents the sample size; Selecting matching degrees in the appearance feature dataset and the internal feature dataset The M-class data satisfying the requirement are combined to obtain a two-dimensional combination result.

Images