A mineral fertilizer production quality management and control method based on big data

Abstract

The present application relates to big data analysis technical field, specifically to a kind of mineral fertilizer production quality control method based on big data, comprising the following steps: reading phosphorus ore diphosphorus pentoxide content and potassium ore impurity calcium magnesium content, generating incoming material quality correlation coefficient set, the numerical values in the incoming material quality correlation coefficient set are weighted combination mapping, and incoming material comprehensive characteristic parameter matrix is generated.The present application introduces finished product granulation pulverization rate, acid-base neutralization consumption determination value into the inversion process of significant dynamic threshold, so that front-end raw material fluctuation and middle section operation fluctuation can accept the reverse check of end quality boundary, thereby obtaining stronger quality causal through capability, abnormal positioning capability and interception specificity, which can not only compress the path length of raw material difference transmission to finished product defect, but also reduce the misjudgment risk caused by empirical threshold, and can also enhance the identification accuracy of batch fluctuation, working condition drift and quality instability precursor.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a method for quality control of mineral fertilizer production based on big data. Background Technology

[0002] Big data analytics technology refers to a comprehensive technology system that revolves around the collection, storage, processing, mining, and application of massive, multi-source, and heterogeneous data. Its core lies in identifying potential relationships and extracting patterns between data through the structured organization and unstructured analysis of high-dimensional data.

[0003] Current technologies focus on structuring high-dimensional data, parsing unstructured data, and identifying potential correlations. In practice, they tend to rely on general data processing frameworks, lacking a continuous characterization of the variable transmission chain in industrial production scenarios. Although data sources are numerous, different data often remain at a parallel summary level. There is a lack of hierarchical constraints between raw material indicators, production actions, and finished product indicators, resulting in correlation results that are closer to macroscopic statistical descriptions and difficult to directly support batch judgment and anomaly interception in the workshop. For example, when multiple data fluctuations are identified, it is difficult to determine whether the fluctuations originate from differences in incoming materials, changes in fan damper opening, or granulator tilt angle deviation. Furthermore, it is difficult to determine whether such fluctuations have approached the quality boundaries of finished product granulation pulverization rate and acid-base neutralization consumption. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a big data-based method for quality control in mineral fertilizer production.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for quality control of mineral fertilizer production based on big data, comprising the following steps: Read the phosphorus pentoxide content of phosphate rock and the calcium and magnesium impurity content of potassium rock to generate a set of incoming material quality correlation coefficients. Then, perform weighted combination mapping on the values ​​of each item in the set of incoming material quality correlation coefficients to generate a comprehensive feature parameter matrix of incoming material. The incoming material comprehensive feature parameter matrix is ​​called, and the fan damper opening and granulator tilt angle are read in combination with the production system log data. A feature vector matrix is ​​constructed, and the covariance value is calculated based on the feature vector matrix to generate a production operation feature vector set. The coordinate parameters of the center point of the production operation feature vector set are extracted to obtain the spatial reference center point parameters. The production operation feature vector set and the spatial reference center point parameter are called, and the distance calculation between the elements in the production operation feature vector set and the spatial reference center point parameter is performed to generate a distance deviation feature set. The distribution variance of each element in the distance deviation feature set is calculated to obtain the variance test parameter of the deviation variable. Based on the variance test parameters of the deviation variables, the measured values ​​of the granulation pulverization rate and acid-base neutralization consumption of the corresponding batch of finished products in the workshop system are extracted, and the corresponding significance dynamic threshold is obtained by inversion. The variance test parameters of the deviation variables are compared with the significance dynamic threshold to generate control interception rules.

[0006] Preferably, the steps for obtaining the comprehensive feature parameter matrix of the incoming material are as follows: Read the values ​​of phosphorus pentoxide content in phosphate rock and the values ​​of calcium and magnesium impurities in potash rock. Calculate the maximum, minimum, and mean deviation of the phosphorus pentoxide content values ​​in phosphate rock and the mean deviation of the calcium and magnesium impurities content values ​​in potash rock. Arrange the values ​​in the corresponding positions based on the range, variance, and values ​​of the phosphorus pentoxide content in phosphate rock and the calcium and magnesium impurities content in potash rock to form a judgment sequence matrix. Extract the ratio relationship of the values ​​in each row of the judgment sequence matrix, sort them row by row, lock the dominant change direction in the judgment sequence matrix, calculate the feature distribution order corresponding to the largest eigenvalue of the judgment sequence matrix, extract the proportion of each component according to the feature distribution order, and form a constant weight. The corresponding benchmark values ​​of the reference standard sequence data are called, and the degree of closeness of the index items corresponding to the constant weights is calculated by comparing them with the reference standard sequence data one by one. The incoming material quality correlation coefficient set is generated according to the index order. Then, the values ​​of each item in the incoming material quality correlation coefficient set are weighted and mapped according to the component proportions corresponding to the constant weights to obtain the incoming material comprehensive feature parameter matrix.

[0007] Preferably, the steps for obtaining the production operation feature vector set are as follows: The incoming material comprehensive feature parameter matrix is ​​invoked to parse the time stamp, fan damper opening record value, and granulator tilt angle record value in the production system log data. The fan damper opening record value is compared with adjacent items according to the time stamp. The number of times the fan damper opening record value changes within a unit recording period is counted, and the number of changes is mapped to continuous recording intervals to form the fan damper opening adjustment frequency. At the same time, the adjacent differences of the granulator tilt angle record value are extracted according to the time stamp. The absolute change of each adjacent difference is counted, and the distribution of absolute change in each recording interval is summarized to form the operation adjustment parameter group. According to the operation adjustment parameter group, the parameter values ​​in the incoming material comprehensive feature parameter matrix are called. The parameter values ​​in the incoming material comprehensive feature parameter matrix, the fan damper opening adjustment frequency, and the granulator tilt angle adjustment range are arranged one by one in a unified recording order to establish a multi-feature combination relationship under the same recording position. Then, the multi-feature combination relationship under each recording position is expanded into a column vector. The deviation of each column element from the mean is calculated. The collaborative change value between elements is calculated column by column to form a production operation feature vector set.

[0008] Preferably, the steps for obtaining the spatial reference center point parameters are as follows: Extract the coordinate components of each vector in the production operation feature vector set, summarize the component values ​​of each vector's corresponding position according to the coordinate dimension, calculate the concentrated position values ​​of each coordinate dimension, and combine the concentrated position values ​​of each coordinate dimension in the original coordinate order to obtain the spatial reference center point parameters.

[0009] Preferably, the step of obtaining the distance deviation feature set is as follows: The production operation feature vector set and the spatial reference center point parameter are called. The coordinate components of each dimension are extracted according to the arrangement order of each vector in the production operation feature vector set. The coordinate components of each dimension are matched with the corresponding coordinate components in the spatial reference center point parameter one by one. The discrete difference of each coordinate component relative to the spatial reference center point parameter is calculated. Then, combined with the cooperative change relationship of each coordinate component in the production operation feature vector set, the discrete difference of each dimension is merged and converted to determine the Mahalanobis distance value of each vector element to the spatial reference center point parameter, forming a deviation variable group. Extract the recording position, coordinate component position, and distance deviation order corresponding to each Mahalanobis distance value in the deviation variable group. Rearrange the Mahalanobis distance values ​​according to the order of the recording positions. Then, group the Mahalanobis distance values ​​of the same type according to the coordinate component positions to establish the distance deviation correspondence relationship under each recording position. Write the rearranged Mahalanobis distance values ​​into a unified sequence according to the distance deviation correspondence relationship to generate a distance deviation feature set.

[0010] Preferably, the steps for obtaining the variance test parameters of the deviation variable are as follows: Read the values ​​of each element in the distance deviation feature set, count the distribution position of each element in the unified sequence, calculate the deviation of each element from the distribution center, summarize the squared results of the deviation of each element, count the distribution variance of each element in the distance deviation feature set, extract the test statistic according to the distribution variance correspondence, and obtain the variance test parameter of the deviation variable.

[0011] Preferably, the step of obtaining the control interception rule item is as follows: The extraction workshop system retrieves the finished product granulation pulverization rate and acid-base neutralization consumption values ​​for the corresponding batches. It reads the upper and lower limits of the quality qualification boundary corresponding to the finished product granulation pulverization rate and the acid-base neutralization consumption values. It calculates the remaining interval and deviation interval of the finished product granulation pulverization rate relative to the upper and lower limits of the quality qualification boundary, respectively. It also calculates the remaining interval and deviation interval of the acid-base neutralization consumption values ​​relative to the upper and lower limits of the quality qualification boundary, respectively. The variance test parameters of the deviation variable are mapped to the boundary intervals of the finished product granulation pulverization rate and acid-base neutralization consumption values. The allowable fluctuation range is then calculated in reverse order of batches to form a significant dynamic threshold. The record position, variable identifier, and numerical value of each test quantity in the variance test parameter of the deviation variable are called. The corresponding threshold relationship between the variance test parameter of the deviation variable and the significance dynamic threshold is compared item by item according to the variable identifier. The variable identifiers of the variance test parameter of the deviation variable are screened out that are greater than the significance dynamic threshold. The start record position, end record position, continuous excess length, maximum deviation value in the interval, and minimum deviation value in the interval corresponding to the excess variable identifier are extracted and written into the interval arrangement table according to the record position order to generate a non-standard action abnormal interval matrix.

[0012] Preferably, the step of obtaining the control interception rule item further includes: Read the maximum deviation value, minimum deviation value, start record position, and end record position corresponding to each abnormal interval in the non-standard action abnormal interval matrix. Extract the maximum deviation value as the upper interception judgment boundary and the minimum deviation value as the lower interception judgment boundary. Combine the start record position and the end record position to limit the alarm triggering interval. Then, write the variable identifier, upper interception judgment boundary, lower interception judgment boundary, and alarm triggering interval into the rule entries one by one to form control interception rule items.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention first transforms the phosphorus pentoxide content of phosphate rock and the calcium and magnesium impurity content of potassium ore into a set of incoming material quality correlation coefficients, and then maps them into a comprehensive incoming material characteristic parameter matrix. This gives the originally scattered raw material indicators a unified expression basis. Furthermore, the fan damper opening, granulator tilt angle, and the comprehensive incoming material characteristic parameter matrix are incorporated into the feature vector matrix. By using covariance values ​​and center point coordinate parameters, the concentrated position of the production status in multidimensional space is extracted. This transforms production fluctuations from simple comparisons of individual records into measurable overall offset relationships. Further, distance calculations, distance deviation feature sets, and deviation... The variable variance test parameters continuously characterize the strength of production disturbances, the distribution of deviations, and the degree of abnormal aggregation. Then, the measured values ​​of finished product granulation pulverization rate and acid-base neutralization consumption are introduced into the inversion process of significance dynamic threshold. This allows the fluctuations of raw materials at the front end and the fluctuations of intermediate operations to be reverse-verified by the end quality boundary. This results in a stronger ability to connect quality causes and effects, anomaly location, and targeted interception. It can not only compress the path length of raw material differences to finished product defects, but also reduce the risk of misjudgment caused by relying solely on empirical thresholds, and enhance the accuracy of identifying batch fluctuations, operating condition drift, and precursors to quality instability. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 A graph showing the relationship between Mahalanobis distance and the significance dynamic threshold for non-standard action anomalies. Figure 3 The simulation diagram shows the boundary inversion of pulverization rate and the control interception. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1-3 This invention provides a technical solution: a method for quality control of mineral fertilizer production based on big data, comprising the following steps: Read the phosphorus pentoxide content of phosphate rock and the calcium and magnesium impurity content of potassium rock to generate a set of incoming material quality correlation coefficients. Then, perform weighted combination mapping on the values ​​of each item in the incoming material quality correlation coefficient set to generate a comprehensive feature parameter matrix of incoming material. The incoming material comprehensive feature parameter matrix is ​​called, and the fan damper opening and granulator tilt angle are read in combination with the production system log data. The feature vector matrix is ​​constructed, the covariance value is calculated based on the feature vector matrix, the production operation feature vector set is generated, and the coordinate parameters of the center point of the production operation feature vector set are extracted to obtain the spatial reference center point parameters. Call the production operation feature vector set and the spatial reference center point parameter, perform distance calculation on the elements in the production operation feature vector set and the spatial reference center point parameter to generate a distance deviation feature set, and calculate the distribution variance of each element in the distance deviation feature set to obtain the variance test parameter of the deviation variable. Based on the variance test parameters of the deviation variable, the measured values ​​of the granulation pulverization rate and acid-base neutralization consumption of the corresponding batch of finished products in the workshop system are extracted, and the corresponding significance dynamic threshold is obtained by inversion. The variance test parameters of the deviation variable are compared with the significance dynamic threshold to generate control interception rules.

[0017] The steps for obtaining the comprehensive feature parameter matrix of incoming materials are as follows: Read the values ​​of phosphorus pentoxide content in phosphate rock and the values ​​of calcium and magnesium impurities in potash rock. Calculate the maximum, minimum, and mean deviation of the phosphorus pentoxide content values ​​in phosphate rock and the mean deviation of the calcium and magnesium impurities content values ​​in potash rock. Arrange the values ​​in the corresponding positions based on the range, variance, and values ​​of the phosphorus pentoxide content in phosphate rock and the calcium and magnesium impurities content in potash rock to form a judgment sequence matrix. Extract the ratio relationship of values ​​in each row of the judgment sequence matrix, sort them row by row, lock the dominant change direction in the judgment sequence matrix, calculate the feature distribution order corresponding to the largest eigenvalue of the judgment sequence matrix, extract the proportion of each component according to the feature distribution order, and form constant weights. The corresponding benchmark values ​​of the reference standard sequence data are called, and the degree of closeness of the index items corresponding to the constant weights is calculated by comparing them with the reference standard sequence data. The incoming material quality correlation coefficient set is generated according to the index order. Then, the values ​​of each item in the incoming material quality correlation coefficient set are weighted and mapped according to the component proportions corresponding to the constant weights to obtain the incoming material comprehensive feature parameter matrix.

[0018] Specifically, after reading the phosphorus pentoxide content values ​​of phosphate rock and the calcium and magnesium impurity content values ​​of potash ore, for each batch of incoming materials, at least 100 samples of phosphorus pentoxide content data for phosphate rock and calcium and magnesium impurity content data for potash ore are collected. Statistical analysis is performed on these two sets of data. Taking the phosphorus pentoxide content data of phosphate rock as an example, the maximum and minimum values ​​are identified, and the difference between them is calculated to obtain the range value of phosphorus pentoxide content in phosphate rock. Simultaneously, the average value of the 100 samples in this set is calculated, and then the average of the squares of the differences between each sample value and the average value is calculated to obtain the variance value of phosphorus pentoxide content in phosphate rock. The same statistical steps are performed on the calcium and magnesium impurity content data of potash ore to obtain the range value of calcium and magnesium impurity content in potash ore and the variance value of potash ore. The variance values ​​of impurity calcium and magnesium content are then used to construct a 2x2 judgment sequence matrix. This matrix is ​​generated by arranging the four key statistical indicators: the range and variance values ​​of phosphorus pentoxide content in phosphate rock, the range and variance values ​​of impurity calcium and magnesium content in potash ore, and the range and variance values ​​of impurity calcium and magnesium content in potash ore. Specifically, the first row of this matrix contains the range and variance values ​​of phosphorus pentoxide content in phosphate rock, and the second row contains the range and variance values ​​of impurity calcium and magnesium content in potash ore. For example, if the calculated range of phosphorus pentoxide content in phosphate rock is 1.2% and the variance is 0.08, and the range of impurity calcium and magnesium content in potash ore is 0.5% and the variance is 0.02, then the resulting judgment sequence matrix is ​​as follows: The structure of this matrix is ​​fixed and is used to quantify the performance of incoming material quality in two dimensions: fluctuation range and stability, forming a judgment sequence matrix.

[0019] Based on the constructed judgment sequence matrix, the weights of each indicator are determined using the analytic hierarchy process (AHP). First, a 4x4 pairwise comparison matrix needs to be established to compare the relative importance of the four indicators: the range of phosphorus pentoxide content in phosphate rock, the variance of phosphorus pentoxide content in phosphate rock, the range of calcium and magnesium impurities in potash rock, and the variance of calcium and magnesium impurities in potash rock. The elements of this matrix... This is determined based on production experience and historical data, using a 1-9 scale. For example, if the variance in phosphorus pentoxide content in phosphate rock is considered to have a far greater impact on production quality than the variance in calcium and magnesium content, which are impurities in potash ore, then... It could be assigned the value 5, depending on... and The entire matrix is ​​filled according to the principle of pairwise comparisons, and after completing the pairwise comparisons, the largest eigenvalue of the matrix is ​​calculated. And its corresponding eigenvectors, this step is achieved by solving the characteristic equation. Implementation, in which Yes, it's a 4x4 pairwise comparison matrix. These are the eigenvectors to be found. These are eigenvalues; the largest eigenvalue obtained after solving for them is... The corresponding eigenvector Each component initially reflects the weights of the four indicators. To obtain the final weight coefficients, the feature vector needs to be normalized, that is, each component is divided by the sum of all components. The calculation formula is as follows: ,in It is the first Normalized weights of each indicator It is the sum of all components of the eigenvector. For example, if the obtained eigenvector is... Their sum is 3.0, then the normalized weight vector is The normalized vector is the constant weight we are looking for.

[0020] The system calls upon pre-established reference standard sequence data, derived from the statistical indicators of incoming materials from several historically highest-quality batches (i.e., "gold batches"). This data includes ideal benchmark values ​​for each indicator. For example, the benchmark value for the range of phosphorus pentoxide content in phosphate rock is 0.8%, and the benchmark value for variance is 0.05; the benchmark value for the range of calcium and magnesium impurities in potash ore is 0.3%, and the benchmark value for variance is 0.01. The system then compares each of the four statistical indicators for the current batch of incoming materials (range of phosphorus pentoxide content in phosphate rock, variance of phosphorus pentoxide content in phosphate rock, range of calcium and magnesium impurities in potash ore, and variance of calcium and magnesium impurities in potash ore) with the corresponding benchmark values ​​in the reference standard sequence data, calculating their degree of similarity. The calculation method is as follows ,in This is the current batch number The measured values ​​of each indicator It is the first The formula uses the baseline value of each indicator to ensure that the degree of closeness is 1 when the measured value is the same as the baseline value. The larger the deviation, the smaller the value. The four calculated degree of closeness values ​​are arranged in order of indicator to form a set of incoming material quality correlation coefficients. For example, if the calculated degree of closeness is as follows: This is the set of incoming material quality correlation coefficients. Next, using the constant weights obtained in the previous step (e.g., ...), ... The set of correlation coefficients is weighted and combined to obtain a comprehensive score. The calculation method is to multiply and sum the values ​​of each item in the set of correlation coefficients with the corresponding weight components to form a vector containing the measured values ​​of each individual indicator, the degree of closeness of each item, and the final comprehensive score. This vector is the incoming material comprehensive feature parameter matrix.

[0021] The steps for obtaining the production operation feature vector set are as follows: The system calls the incoming material comprehensive feature parameter matrix to parse the time stamp, fan damper opening record value, and granulator tilt angle record value in the production system log data. It compares adjacent items of the fan damper opening record value according to the time stamp, counts the number of times the fan damper opening record value changes within a unit recording period, and maps the number of changes to continuous recording intervals to form the fan damper opening adjustment frequency. At the same time, it extracts the adjacent differences of the granulator tilt angle record value according to the time stamp, counts the absolute change of each adjacent difference, summarizes the distribution of absolute change in each recording interval, and forms the operation adjustment parameter group. Based on the operation adjustment parameter group, the parameter values ​​in the incoming material comprehensive feature parameter matrix are called. The parameter values ​​in the incoming material comprehensive feature parameter matrix, the fan damper opening adjustment frequency, and the granulator tilt angle adjustment range are arranged one by one according to a unified recording order. The multi-feature combination relationship under the same recording position is established. Then, the multi-feature combination relationship under each recording position is expanded into a column vector. The deviation of each column element from the mean is calculated. The collaborative change value between elements is calculated column by column to form a production operation feature vector set.

[0022] Specifically, the incoming material comprehensive characteristic parameter matrix is ​​invoked, and time-series data automatically collected from the Production Control System (PCS) logs is parsed simultaneously. This data is recorded once per second and includes millisecond-accurate timestamps, fan damper opening values ​​ranging from 0% to 100%, and granulator tilt angle values ​​ranging from 0 to 15 degrees. These continuous records are divided into independent recording cycles of 60 seconds each. Within each cycle, the fan damper opening values ​​at adjacent time points are compared one by one. If the value recorded in the next second is not exactly the same as that in the previous second, it is considered that an adjustment has occurred. The number of adjustments within a cycle is accumulated to obtain the fan damper opening adjustment for that cycle. The adjustment frequency is calculated as follows: For example, if the baffle opening changes at the 10th, 25th, and 50th seconds within a 60-second cycle, the adjustment frequency for that cycle is 3 times per minute. Simultaneously, for the granulator tilt angle recording value, the absolute value of the difference between every two adjacent second-by-second recording values ​​is calculated within each 60-second cycle. These 59 absolute differences are then summed to obtain the total adjustment range of the granulator tilt angle within that cycle. For example, if the sum of the absolute values ​​of all adjacent tilt angle differences within a certain cycle is 1.2 degrees, then the total adjustment range for that cycle is 1.2 degrees. The fan baffle opening adjustment frequency calculated for each unit recording cycle and the total adjustment range of the granulator tilt angle are combined to form an operating adjustment parameter set.

[0023] The operation adjusts the parameter group and calls the incoming material comprehensive feature parameter matrix corresponding to the current production batch. This matrix contains several fixed parameter values, such as the range and variance of phosphorus pentoxide content in phosphate rock. For each unit recording period (e.g., per minute), all parameter values ​​in this fixed incoming material comprehensive feature parameter matrix are sequentially concatenated with the fan damper opening adjustment frequency and granulator tilt angle adjustment amplitude calculated for that period to construct a high-dimensional feature vector. For example, if the incoming material comprehensive feature parameter matrix has 4 parameters, then the feature vector for each recording period is a vector containing 6 elements (4 incoming material parameters + 1...). A vector (frequency + 1 amplitude) is used to aggregate all feature vectors generated during the entire production batch duration (e.g., 8 hours, or 480 recording cycles). This aggregates multiple feature combinations. For this feature set consisting of 480 6-dimensional vectors, a column vector expansion is first performed, treating it as a 480-row, 6-column matrix. Then, the average value of all elements in each column (representing a specific feature dimension) is calculated. Next, the difference between each element in each column and its column average is calculated to obtain the deviation. Finally, based on these deviations, the covariance between any two columns is calculated. The formula for calculating this covariance is... ,in, and These represent two different feature column vectors. and These two vectors are at the th... The value for each recording period, and It is the average of these two columns. The total number of recording cycles (e.g., 480) quantifies the synergistic relationship between different operating parameters and incoming material quality characteristics. The final set of 480 6-dimensional feature vectors is the production operation feature vector set.

[0024] The steps for obtaining the parameters of the spatial reference center point are as follows: Extract the coordinate components of each vector in the production operation feature vector set, summarize the component values ​​of each vector's corresponding position according to the coordinate dimension, calculate the central position values ​​of each coordinate dimension, and combine the central position values ​​of each coordinate dimension in the original coordinate order to obtain the spatial reference center point parameters.

[0025] Specifically, the production operation feature vector set generated in the previous step is extracted. This set contains multi-dimensional feature vectors corresponding to each unit recording cycle (e.g., per minute) in the entire production batch. Taking a set containing 480 6-dimensional vectors as an example, the processing is as follows: First, for the first coordinate dimension (e.g., the range of phosphorus pentoxide content in phosphate rock), all component values ​​located at the first position in the 480 vectors are extracted to form a sequence containing 480 values. Then, the average value of this sequence is calculated to obtain the central location value of the first dimension. Next, the same operation is performed on the second coordinate dimension (e.g., the variance of phosphorus pentoxide content in phosphate rock), extracting the components at the second position in all vectors and calculating their average value to obtain the central location value of the second dimension. This process is repeated for all 6 coordinate dimensions, that is, summarizing the values ​​of all dimensions such as the frequency of fan damper opening adjustment and the adjustment range of granulator tilt angle within 480 recording cycles and calculating their respective average values. Finally, 6 central location values ​​corresponding to each dimension are obtained. For example, the obtained values ​​may be: Then, these 6 calculated average values ​​are recombined into a vector with the same dimensions as the original feature vector according to their original coordinate dimension order (i.e., incoming material parameter 1, incoming material parameter 2, ..., frequency, amplitude), to obtain the spatial reference center point parameters.

[0026] The steps for obtaining the distance deviation feature set are as follows: The production operation feature vector set and spatial reference center point parameters are called. The coordinate components of each dimension are extracted according to the arrangement order of each vector in the production operation feature vector set. The coordinate components of each dimension are matched with the corresponding coordinate components in the spatial reference center point parameters one by one. The discrete difference of each coordinate component relative to the spatial reference center point parameters is calculated. Then, combined with the cooperative change relationship of each coordinate component in the production operation feature vector set, the discrete difference of each dimension is merged and converted to determine the Mahalanobis distance value of each vector element to the spatial reference center point parameters, forming a deviation variable group. Extract the recording position, coordinate component position, and distance deviation order corresponding to each Mahalanobis distance value in the deviation variable group. Rearrange the Mahalanobis distance values ​​according to the order of the recording positions. Then, group the Mahalanobis distance values ​​of the same type according to the coordinate component positions to establish the distance deviation correspondence under each recording position. Write the rearranged Mahalanobis distance values ​​into a unified sequence according to the distance deviation correspondence to generate a distance deviation feature set.

[0027] Specifically, the process involves calling the production operation feature vector set and the spatial reference center point parameters. First, the feature vector corresponding to each recording cycle is extracted sequentially from the production operation feature vector set, and its coordinate components in each dimension are obtained. Simultaneously, this feature vector is aligned with the spatial reference center point parameters, which serve as the geometric center of the multidimensional space. The differences between the two in each dimension are calculated, forming a discrete difference vector with the same dimensions as the original feature vector. Then, combined with the covariance matrix calculated in previous steps, which reflects the cooperative change relationship between the feature dimensions, this discrete difference vector is merged and transformed. This transformation process is accomplished by calculating the Mahalanobis distance, the formula of which is... ,in, Representing the Feature vector of each recording period The Mahalanobis distance to the center point. The first one extracted from the set of production operation feature vectors 1 eigenvector It is a spatial reference center point parameter that represents the overall central tendency of the data. It is the covariance matrix of the feature vector set of production operations. The inverse matrix is ​​used to overcome the inconsistency in the frequency of change of multidimensional feature data (such as the relatively stable parameters of incoming materials within a single batch), which leads to a different covariance matrix. The problem of singularity preventing inversion can be addressed by considering the covariance matrix before performing the inversion operation. A very small regularization perturbation constant is uniformly superimposed on the main diagonal to ensure the stability of the full-rank mapping of the low-rank matrix. The elements of this covariance matrix quantify the correlation strength between different characteristics, such as the frequency of fan damper opening adjustment and the variance of phosphorus pentoxide content in the incoming phosphate rock. The transpose operation of a vector is represented by repeatedly performing this Mahalanobis distance calculation on each feature vector in the set of feature vectors of production operations, resulting in a series of scalar distance values. For example, calculating the Mahalanobis distance on a set containing feature vectors of 480 recording periods will yield 480 Mahalanobis distance values, which together form the bias variable set.

[0028] Extract a series of Mahalanobis distance values ​​from the deviation variable group, and associate each value with its original production time information, i.e., the recording position. For example, the first Mahalanobis distance value corresponds to the first unit recording cycle after production begins (e.g., minute 1), the second value corresponds to the second cycle (e.g., minute 2), and so on. Organize these Mahalanobis distance values ​​with recording position identifiers and rearrange them according to the chronological order of the recording positions to form an ordered time series. This process does not change the Mahalanobis distance values ​​themselves, but only organizes them in chronological order. Next, to form a structured dataset, establish a one-to-one mapping relationship between each recording position and its corresponding Mahalanobis distance value. For example, (recording position...) (Record position 1, Mahalanobis distance value 2.5), (Record position 2, Mahalanobis distance value 3.1), ..., (Record position 480, Mahalanobis distance value 2.8). After establishing this correspondence, all rearranged Mahalanobis distance values ​​are written into a single sequence in chronological order. The first element of this sequence is the Mahalanobis distance of the first recording period, the second element is the Mahalanobis distance of the second recording period, and so on, until the Mahalanobis distance of the last recording period. Each element in this sequence only represents the overall operational deviation at its corresponding time point, without retaining the component information of the original feature vector. This final ordered list consisting of a series of Mahalanobis distance values ​​sorted by time is the distance deviation feature set.

[0029] The steps for obtaining the variance test parameters for the deviation variable are as follows: Read the values ​​of each element in the distance deviation feature set, count the distribution position of each element in the unified sequence, calculate the deviation of each element from the distribution center, summarize the squared results of the deviation of each element, count the distribution variance of each element in the distance deviation feature set, extract the test statistic according to the distribution variance correspondence, and obtain the variance test parameters of the deviation variable.

[0030] Specifically, the distance deviation feature set is read. This set is a uniform sequence containing, for example, 480 Mahalanobis distance values. First, the mean of all elements in the sequence is calculated, and this mean is defined as the distribution center of the sequence. Then, for each element value in the sequence, the difference between it and the previously calculated distribution center (mean) is calculated to obtain the deviation of each element. Next, each calculated deviation is squared, and the squared results of the deviations of all elements are summed to obtain the sum of squares of the deviations. Finally, to calculate the dispersion of the entire sequence, i.e., the distribution variance, this deviation is... The sum of squares of the deviations is divided by (the total number of elements in the sequence minus 1). For example, if the sequence contains 480 elements, the sum of squares is divided by 479. This result is the sample variance of the deviation feature set. As a key test metric, the magnitude of this sample variance reflects the overall volatility of the operating status relative to the ideal baseline state in the entire production batch. This sample variance value is ultimately determined as the deviation variable variance test parameter. For example, if the calculated sample variance is 2.73, then the value of the deviation variable variance test parameter is 2.73. This parameter will be used subsequently to assess the stability of the production process.

[0031] The steps to obtain the control and interception rule items are as follows: Extract the granulation pulverization rate and acid-base neutralization consumption values ​​of the corresponding batches in the extraction workshop system. Read the upper and lower limits of the quality qualification boundary corresponding to the granulation pulverization rate and acid-base neutralization consumption values. Calculate the remaining interval and deviation interval of the granulation pulverization rate relative to the upper and lower limits of the quality qualification boundary, respectively. Calculate the remaining interval and deviation interval of the acid-base neutralization consumption value relative to the upper and lower limits of the quality qualification boundary, respectively. Map the variance test parameters of the deviation variable to the boundary intervals of the granulation pulverization rate and acid-base neutralization consumption values. Calculate the allowable fluctuation range in reverse order of batches to form a significant dynamic threshold. The system calls up the record position, variable identifier, and value of each test quantity in the variance test parameter of the deviation variable. It compares the corresponding threshold relationship between the variance test parameter of the deviation variable and the significance dynamic threshold item by item according to the variable identifier. It screens out the variable identifiers of the variance test parameter of the deviation variable that are greater than the significance dynamic threshold. It extracts the start record position, end record position, continuous excess length, maximum deviation value in the interval, and minimum deviation value in the interval corresponding to the excess variable identifier. It writes them into the interval arrangement table according to the record position order to generate the non-standard action abnormal interval matrix. Read the maximum deviation value, minimum deviation value, start record position, and end record position corresponding to each abnormal interval in the non-standard action abnormal interval matrix. Extract the maximum deviation value as the upper interception judgment boundary and the minimum deviation value as the lower interception judgment boundary. Combine the start record position and the end record position to limit the alarm triggering interval. Then, write the variable identifier, upper interception judgment boundary, lower interception judgment boundary, and alarm triggering interval into the rule entries one by one to form the control interception rule items.

[0032] Specifically, the system extracts the pulverization rate and acid-base neutralization consumption values ​​for a specific production batch from the workshop quality inspection system. For example, a batch might have a pulverization rate of 4.5% and an acid-base neutralization consumption of 22 ml. Simultaneously, it retrieves the acceptable boundaries for these two indicators from a pre-set quality standard database. For instance, the acceptable quality boundaries for pulverization rate are set at [3.0%, 5.0%], and for acid-base neutralization consumption at [20 ml, ...]. [25 ml] Then, for the powdering rate, the remaining interval of its measured value relative to the upper limit is calculated, i.e., 5.0% minus 4.5% equals 0.5%, and its deviation interval relative to the lower limit is calculated, i.e., 4.5% minus 3.0% equals 1.5%. Similarly, the acid-base neutralization consumption is calculated, and the remaining interval relative to the upper limit is 3 ml, and the deviation interval relative to the lower limit is 2 ml. Next, a correlation model is established between the variance test parameter of the deviation variable and the finished product quality index. This model is based on historical data analysis, and calculates the ratio of the variance test parameter of the deviation variable of multiple batches to the remaining interval of the finished product quality of the corresponding batch, and takes the average value as the correlation coefficient. For example, if the historical average ratio is 5.0, then This coefficient is used to inversely calculate the upper limit of process fluctuations that the current batch can tolerate, i.e., to calculate separately. and The smaller value is selected as the most stringent control standard, i.e., 2.5, to form the significance dynamic threshold.

[0033] The test parameter for the variance of the deviation variable, determined in the previous step and representing the stability of the entire batch process fluctuation, is invoked. This parameter is essentially the sample variance of all Mahalanobis distance values ​​within the distance deviation feature set. Simultaneously, the distance deviation feature set itself is invoked; this set is a sequence of Mahalanobis distance values ​​arranged in chronological order, for example, a sequence containing 480 Mahalanobis distance values. Each Mahalanobis distance value (as the test statistic) in the sequence is compared one by one with a significance dynamic threshold (e.g., 2.5). Each Mahalanobis distance value has its corresponding recording position (e.g., minute 1, minute 2, etc.) and a unique variable identifier (in this scenario, the variable identifier is uniformly "Mahalanobis distance"). The entire sequence is traversed. The sequence identifies all recorded points with values ​​greater than 2.5 and finds consecutively exceeding limits intervals. For example, if the Mahalanobis distances at the 12th, 13th, and 14th minutes are 2.8, 3.1, and 2.6 respectively, they constitute a consecutive exceeding limit interval. For each such interval, the starting recording position (12th minute), ending recording position (14th minute), consecutive exceeding limit length (3 minutes), maximum deviation value (3.1), and minimum deviation value (2.6) within the interval are extracted. This information is combined into an entry, and all entries corresponding to the exceeding limit intervals are written into a list in chronological order. This list is the interval arrangement table, which ultimately forms a structured non-standard action abnormal interval matrix.

[0034] Read the non-standard action abnormality interval matrix generated in the previous step. Each row of this matrix describes a non-standard action interval in the production process in detail. Process each abnormal interval entry in the matrix to generate specific control rules. For example, process one entry, which records the start recording position as the 12th minute, the end recording position as the 14th minute, the maximum deviation value within the interval is 3.1, the minimum deviation value is 2.6, and the variable is identified as "Madara distance". According to the rule, the maximum deviation value of 3.1 in this interval is set as the upper interception judgment boundary of this rule item, and the minimum deviation value of 2.6 is set as the lower interception judgment boundary. At the same time, the alarm triggering time range is limited to the beginning of the 12th minute to the end of the 14th minute using the start recording position and the end recording position. Then, the variable identifier ("Madara distance"), the upper interception judgment boundary (3.1), the lower interception judgment boundary (2.6), and the alarm triggering interval ([12:00, [14:59]) These four pieces of information are integrated to form a complete rule entry. This operation is repeated for all entries in the non-standard action abnormal interval matrix. Each generated rule entry is then compiled to form a control interception rule item.

[0035] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for quality control of mineral fertilizer production based on big data, characterized in that, Includes the following steps: Read the phosphorus pentoxide content of phosphate rock and the calcium and magnesium impurity content of potassium rock to generate a set of incoming material quality correlation coefficients. Then, perform weighted combination mapping on the values ​​of each item in the set of incoming material quality correlation coefficients to generate a comprehensive feature parameter matrix of incoming material. The incoming material comprehensive feature parameter matrix is ​​called, and the fan damper opening and granulator tilt angle are read in combination with the production system log data. A feature vector matrix is ​​constructed, and the covariance value is calculated based on the feature vector matrix to generate a production operation feature vector set. The coordinate parameters of the center point of the production operation feature vector set are extracted to obtain the spatial reference center point parameters. The production operation feature vector set and the spatial reference center point parameter are called, and the distance calculation between the elements in the production operation feature vector set and the spatial reference center point parameter is performed to generate a distance deviation feature set. The distribution variance of each element in the distance deviation feature set is calculated to obtain the variance test parameter of the deviation variable. Based on the variance test parameters of the deviation variables, the measured values ​​of the granulation pulverization rate and acid-base neutralization consumption of the corresponding batch of finished products in the workshop system are extracted, and the corresponding significance dynamic threshold is obtained by inversion. The variance test parameters of the deviation variables are compared with the significance dynamic threshold to generate control interception rules.

2. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the comprehensive feature parameter matrix of the incoming material are as follows: Read the values ​​of phosphorus pentoxide content in phosphate rock and the values ​​of calcium and magnesium impurities in potash rock. Calculate the maximum, minimum, and mean deviation of the phosphorus pentoxide content values ​​in phosphate rock and the mean deviation of the calcium and magnesium impurities content values ​​in potash rock. Arrange the values ​​in the corresponding positions based on the range, variance, and values ​​of the phosphorus pentoxide content in phosphate rock and the calcium and magnesium impurities content in potash rock to form a judgment sequence matrix. Extract the ratio relationship of the values ​​in each row of the judgment sequence matrix, sort them row by row, lock the dominant change direction in the judgment sequence matrix, calculate the feature distribution order corresponding to the largest eigenvalue of the judgment sequence matrix, extract the proportion of each component according to the feature distribution order, and form a constant weight. The corresponding benchmark values ​​of the reference standard sequence data are called, and the degree of closeness of the index items corresponding to the constant weights is calculated by comparing them with the reference standard sequence data one by one. The incoming material quality correlation coefficient set is generated according to the index order. Then, the values ​​of each item in the incoming material quality correlation coefficient set are weighted and mapped according to the component proportions corresponding to the constant weights to obtain the incoming material comprehensive feature parameter matrix.

3. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the production operation feature vector set are as follows: The incoming material comprehensive feature parameter matrix is ​​invoked to parse the time stamp, fan damper opening record value, and granulator tilt angle record value in the production system log data. The fan damper opening record value is compared with adjacent items according to the time stamp. The number of times the fan damper opening record value changes within a unit recording period is counted, and the number of changes is mapped to continuous recording intervals to form the fan damper opening adjustment frequency. At the same time, the adjacent differences of the granulator tilt angle record value are extracted according to the time stamp. The absolute change of each adjacent difference is counted, and the distribution of absolute change in each recording interval is summarized to form the operation adjustment parameter group. According to the operation adjustment parameter group, the parameter values ​​in the incoming material comprehensive feature parameter matrix are called. The parameter values ​​in the incoming material comprehensive feature parameter matrix, the fan damper opening adjustment frequency, and the granulator tilt angle adjustment range are arranged one by one in a unified recording order to establish a multi-feature combination relationship under the same recording position. Then, the multi-feature combination relationship under each recording position is expanded into a column vector. The deviation of each column element from the mean is calculated. The collaborative change value between elements is calculated column by column to form a production operation feature vector set.

4. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the spatial reference center point parameters are as follows: Extract the coordinate components of each vector in the production operation feature vector set, summarize the component values ​​of each vector's corresponding position according to the coordinate dimension, calculate the concentrated position values ​​of each coordinate dimension, and combine the concentrated position values ​​of each coordinate dimension in the original coordinate order to obtain the spatial reference center point parameters.

5. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the distance deviation feature set are as follows: The production operation feature vector set and the spatial reference center point parameter are called. The coordinate components of each dimension are extracted according to the arrangement order of each vector in the production operation feature vector set. The coordinate components of each dimension are matched with the corresponding coordinate components in the spatial reference center point parameter one by one. The discrete difference of each coordinate component relative to the spatial reference center point parameter is calculated. Then, combined with the cooperative change relationship of each coordinate component in the production operation feature vector set, the discrete difference of each dimension is merged and converted to determine the Mahalanobis distance value of each vector element to the spatial reference center point parameter, forming a deviation variable group. Extract the recording position, coordinate component position, and distance deviation order corresponding to each Mahalanobis distance value in the deviation variable group. Rearrange the Mahalanobis distance values ​​according to the order of the recording positions. Then, group the Mahalanobis distance values ​​of the same type according to the coordinate component positions to establish the distance deviation correspondence relationship under each recording position. Write the rearranged Mahalanobis distance values ​​into a unified sequence according to the distance deviation correspondence relationship to generate a distance deviation feature set.

6. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the variance test parameters of the deviation variable are as follows: Read the values ​​of each element in the distance deviation feature set, count the distribution position of each element in the unified sequence, calculate the deviation of each element from the distribution center, summarize the squared results of the deviation of each element, count the distribution variance of each element in the distance deviation feature set, extract the test statistic according to the distribution variance correspondence, and obtain the variance test parameter of the deviation variable.

7. The method for quality control of mineral fertilizer production based on big data according to claim 1, characterized in that, The steps for obtaining the control interception rule item are as follows: The extraction workshop system retrieves the finished product granulation pulverization rate and acid-base neutralization consumption values ​​for the corresponding batches. It reads the upper and lower limits of the quality qualification boundary corresponding to the finished product granulation pulverization rate and the acid-base neutralization consumption values. It calculates the remaining interval and deviation interval of the finished product granulation pulverization rate relative to the upper and lower limits of the quality qualification boundary, respectively. It also calculates the remaining interval and deviation interval of the acid-base neutralization consumption values ​​relative to the upper and lower limits of the quality qualification boundary, respectively. The variance test parameters of the deviation variable are mapped to the boundary intervals of the finished product granulation pulverization rate and acid-base neutralization consumption values. The allowable fluctuation range is then calculated in reverse order of batches to form a significant dynamic threshold. The record position, variable identifier, and numerical value of each test quantity in the variance test parameter of the deviation variable are called. The corresponding threshold relationship between the variance test parameter of the deviation variable and the significance dynamic threshold is compared item by item according to the variable identifier. The variable identifiers of the variance test parameter of the deviation variable are screened out that are greater than the significance dynamic threshold. The start record position, end record position, continuous excess length, maximum deviation value in the interval, and minimum deviation value in the interval corresponding to the excess variable identifier are extracted and written into the interval arrangement table according to the record position order to generate a non-standard action abnormal interval matrix.

8. The method for quality control of mineral fertilizer production based on big data according to claim 7, characterized in that, The steps for obtaining the control interception rule items also include: Read the maximum deviation value, minimum deviation value, start record position, and end record position corresponding to each abnormal interval in the non-standard action abnormal interval matrix. Extract the maximum deviation value as the upper interception judgment boundary and the minimum deviation value as the lower interception judgment boundary. Combine the start record position and the end record position to limit the alarm triggering interval. Then, write the variable identifier, upper interception judgment boundary, lower interception judgment boundary, and alarm triggering interval into the rule entries one by one to form control interception rule items.

Images