Dynamic prediction method for screen residue of cement negative pressure screen analyzer

Abstract

The invention relates to the technical field of industrial data processing, in particular to a cement negative pressure screen analyzer screen residue dynamic prediction method, which comprises the following steps: constructing a production stage database, collecting a to-be-detected sample through distributed nodes, mining and establishing the relevance between the to-be-detected sample and a historical screening map by utilizing an association rule, and generating a weighted sample data cube; analyzing multi-dimensional parameters in the dynamic screening log library through a data processing model, and constructing a screen state topological relation; a multi-dimensional index algorithm is adopted to calculate similarity measurement of the sensing data flow and the particle distribution database, and a data association factor is generated; a state judgment rule base is established according to the screen state topological relation and the data association factors, and prediction of the screen residue is executed through a corresponding data processing method; and carrying out weight vector aggregation on the predicted value, and outputting a screen residue prediction result set. According to the method, high-precision screen residue prediction under complex production scenes such as insufficient sample representativeness, frequent screen blockage and environmental factor disturbance is realized.

Description

Technical Field

[0001] The invention relates to the technical field of industrial data processing, in particular to a method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer. Background Art

[0002] Currently, methods for predicting the residue of cement negative pressure sieve analyzers are generally based on static structured data architectures. These methods primarily construct indexed relationships between attribute data tables (including particle size distribution and chemical composition) of a finite sample and historical screening records, using traditional database query methods to achieve predictions. Existing systems often store production sample data, screening equipment operation logs, and environmental sensor parameters in separate database tables. These systems only support single-dimensional, static information retrieval, making it difficult to conduct dynamic data analysis across variables and stages.

[0003] Existing methods often use relational databases to match and conditionally filter different data sources using structured query language (SQL). Some systems rely on static weight models to achieve a basic mapping between sample attributes and screening results, but lack the ability to deeply explore the implicit relationships between high-dimensional, multivariate data. This isolated, static data organization model limits the accuracy and adaptability of data prediction in complex production scenarios. In particular, traditional models struggle to ensure the stability and real-time performance of prediction results in situations with insufficient sample representativeness, frequent screen blockages, and environmental disturbances.

[0004] Therefore, existing technologies have yet to establish a unified multi-dimensional data fusion and dynamic modeling mechanism, making it impossible to fully collect, index, and adaptively analyze key information during the screening process. To address the issues of process nonlinearity, dynamic state evolution, and external environmental interference in cement screening, an information processing method focused on data structure optimization and dynamic association mining is urgently needed to support high-precision screen residue prediction and linked judgment of complex screening states.

[0005] Therefore, a dynamic prediction method for the sieve residue of cement negative pressure sieve analyzer was proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a dynamic prediction method for the sieve residue of a cement negative pressure sieve analyzer, so as to achieve high-precision sieve residue prediction in complex production scenarios such as insufficient sample representativeness, frequent screen blockage and environmental disturbance.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer comprises:

[0009] Build a production stage database, collect samples to be tested through distributed nodes, and obtain the attribute data table of the samples to be tested; use association rule mining to establish the correlation between the attribute data table and the historical screening map, and generate a weighted sample data cube;

[0010] Create a dynamic screening log library based on the weighted sample data cube, analyze the multi-dimensional parameters in the dynamic screening log library through the data processing model, and construct the topological relationship of the screening state;

[0011] The sensor data stream is acquired through multi-source sensors, and the similarity measure between the sensor data stream and the particle distribution database is calculated using a multidimensional indexing algorithm to generate a data correlation factor.

[0012] The data relationship model is called to output the first predicted value, and a state discrimination rule base is established based on the screen state topological relationship and the data association factor: when the first rule is met, the first predicted value is directly output; when the second rule is met, the first compensation calibration is performed based on the screen state topological relationship to generate the second predicted value; when the third rule is met, the second compensation calibration is performed through the environmental impact model to generate the third predicted value;

[0013] Perform weight vector aggregation on the first predicted value, the second predicted value, and the third predicted value, and output a sieve residue prediction result set.

[0014] Preferably, the process of generating the weighted sample data cube includes:

[0015] The cement production process is divided into multiple production stages, and N distributed sampling nodes are configured for each production stage to collect samples to be tested;

[0016] Obtain attribute data of each sample to be tested, including particle size distribution, particle morphology, particle spacing and humidity, construct attribute data table, and store it in the production stage database;

[0017] Use association rules to mine and analyze the potential associations between attribute data tables and historical screening graphs, and establish association rule paths;

[0018] Calculating the importance of each sample to be tested to the preset screening index based on the association rule path, and assigning a representative weight to each sample to be tested;

[0019] The samples to be detected are weighted and combined according to the representative weights to generate the weighted sample data cube.

[0020] Preferably, the process of constructing the sieve state topological relationship includes:

[0021] Dynamically screening the weighted sample data cube, monitoring multi-dimensional parameters in real time, including sieve flow rate, sieve blockage degree and vibration frequency, and creating a dynamic screening log library;

[0022] Analyzing the multi-dimensional parameters, solving the congestion rate by simultaneous equations, and identifying and predicting the congestion trend;

[0023] A screen state topology relationship is constructed according to the clogging rate and the clogging trend.

[0024] Preferably, the process of generating the data correlation factor includes:

[0025] Normalizing the sensor data stream acquired by the multi-source sensors, the sensor data stream including humidity, temperature, air pressure, and dust concentration;

[0026] A multidimensional feature vector is constructed based on the sensor data stream, and a multidimensional indexing algorithm is used to perform similarity matching on a particle distribution database to obtain a similarity metric value, and K reference samples with the highest similarity metric values are extracted; the particle distribution database contains a plurality of paired samples of historical environmental characteristics and corresponding particle distribution curves;

[0027] The data association factor reflecting the relationship between the current environmental state and the historical particle distribution is generated in combination with the similarity measurement value.

[0028] Preferably, the state discrimination rule base specifically includes:

[0029] When the congestion trend does not meet the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the first rule is met;

[0030] When the congestion trend meets the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the second rule is met;

[0031] When the congestion trend does not meet the preset congestion condition and the data correlation factor is higher than the preset correlation threshold, it is determined to be in compliance with the third rule.

[0032] Preferably, the specific process of the first compensation calibration includes:

[0033] Combined with the topological relationship of the screen state, the deviation influence of the multi-dimensional parameters on the prediction of the screen residue is analyzed;

[0034] calculating a compensation amount for correcting the second predicted value according to the deviation influence;

[0035] The calculated compensation amount is combined with the first predicted value to generate a second predicted value that has undergone first compensation calibration.

[0036] Preferably, the specific process of the second compensation calibration includes:

[0037] Through the environmental impact model, the coupling mapping relationship between environmental factors and particle movement characteristics is constructed, including the temperature-diffusion rate function, humidity-aggregation coefficient function, air pressure-particle migration probability function, and dust concentration-sieve permeability function;

[0038] Applying the coupling mapping relationship to the current sensor data stream to dynamically calculate the environmental interference influence coefficient matrix;

[0039] Correcting the particle distribution model according to the influence coefficient matrix and inverting the trend of the sieve residue deviation during the screening process;

[0040] The first predicted value is offset-adjusted in combination with the inversion result to generate a third predicted value that has undergone a second compensation calibration.

[0041] Preferably, the output process of the sieve residue prediction result set includes:

[0042] integrating the first prediction value, the second prediction value, and the third prediction value generated according to different rules;

[0043] Based on the reliability and importance of each set of prediction values, weights are assigned, and the three sets of prediction values are weightedly fused to obtain the comprehensive screening residue prediction value;

[0044] The comprehensive screen residue prediction value is output in a structured format as the final screen residue prediction result set.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention deploys distributed sampling nodes at multiple stages of cement production and constructs a weighted sample data cube based on multiple attributes such as particle size distribution, particle morphology, and humidity, thereby improving the comprehensiveness and representativeness of the collected samples. Compared with existing static sample models, this method no longer relies on a finite sample library for static inference, but instead strengthens the dynamic association between samples and historical screening behaviors through a representative weight mechanism. This method significantly improves the adaptability and generalization ability of the prediction model in cross-batch and multi-environment production scenarios, avoids prediction bias caused by single or distorted samples, and is suitable for high-frequency variable operating conditions in actual working conditions.

[0047] 2. This invention proposes a topological relationship modeling method for screen mesh states based on a dynamic screening log library. This method monitors and analyzes key parameters such as mesh flow rate and blockage level in real time, enabling timely identification of screen blockage trends and their evolutionary paths. This method incorporates a simultaneous parameter solution and trend prediction mechanism, which not only improves the accuracy of blockage event detection but also proactively triggers a compensation calibration mechanism before blockage occurs. This mechanism enhances the predictive model's ability to perceive and respond to changes in device status, effectively reducing the problem of misjudgment of screen residue due to abnormal device status.

[0048] 3. The present invention comprehensively considers the impact of environmental variables such as temperature, humidity, air pressure, and dust concentration on the screening process by constructing a coupling mapping model between environmental factors and particle movement behavior. When the screening result is judged to be abnormal, the system can call this model to perform environmental compensation calibration on the screening prediction result, thereby correcting problems such as particle aggregation and sieve hole blockage caused by environmental disturbances. Compared with the lack of dynamic modeling of environmental factors in existing methods, this method significantly improves the robustness and prediction reliability of the model under complex working conditions, meeting the needs of accurate prediction under highly volatile working conditions in industrial sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic flow chart of a method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer provided in an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of a process for generating a weighted sample data cube according to an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of a process for constructing a topological relationship of a screen state according to an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of a process for generating a data correlation factor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] Predicting the residual volume of cement negative pressure sieve analyzers is crucial for ensuring the quality stability of cement products and the continuity of production processes. Accurately predicting the residual volume not only reflects the changing trends in cement particle distribution in real time, allowing for the timely detection of abnormal fluctuations in the production process, but also assists in adjusting mill operating parameters and raw material ratios, effectively avoiding rework and resource waste caused by unqualified screening. Furthermore, accurate residual volume prediction helps improve the automation and intelligence of screening tests, providing critical data support for achieving green, efficient, and controllable cement production.

[0055] This paper proposes a method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer. This method enables high-precision sieve residue prediction in complex production scenarios, such as insufficient sample representation, frequent screen clogging, and environmental disturbances. To demonstrate the effectiveness of the method for high-precision sieve residue prediction, the following two examples will illustrate the effectiveness of the method.

[0056] Example 1

[0057] In the embodiment of the present application, the high-precision screen residue prediction process of the method proposed in the present invention in complex production scenarios such as insufficient sample representativeness, frequent screen blockage and environmental disturbance is described in detail. The embodiment of the present application is aimed at the high-precision screen residue prediction in the cement production process of a certain factory A. Figure 1 The content describes in detail the high-precision screen residue prediction process in the cement production process; Figure 1 The specific flow chart of the method proposed in the present invention includes: constructing a production stage database, collecting samples to be tested through distributed nodes, and obtaining the attribute data table of the samples to be tested; using association rule mining to establish the correlation between the attribute data table and the historical screening map, and generating a weighted sample data cube; creating a dynamic screening log library based on the weighted sample data cube, analyzing the multi-dimensional parameters in the dynamic screening log library through the data processing model, and constructing the sieve state topology relationship; obtaining the sensor data stream through multi-source sensors, and using the multi-dimensional indexing algorithm to calculate the similarity measure between the sensor data stream and the particle distribution database, and generating a data association factor; establishing a state discrimination rule library based on the sieve state topology relationship and the data association factor, and performing the prediction of the sieve residue through the corresponding data processing method; performing weight vector aggregation on the first prediction value, the second prediction value and the third prediction value, and outputting the sieve residue prediction result set. Combined with Figure 1 and Figure 2 The following describes the contents:

[0058] A method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer comprises:

[0059] Build a production stage database, collect samples to be tested through distributed nodes, and obtain the attribute data table of the samples to be tested; use association rule mining to establish the correlation between the attribute data table and the historical screening map, and generate a weighted sample data cube, such as Figure 2 As shown;

[0060] The process of generating the weighted sample data cube includes:

[0061] The cement production process is divided into multiple production stages, and N distributed sampling nodes are configured for each production stage to collect samples to be tested;

[0062] Obtain attribute data of each sample to be tested, including particle size distribution, particle morphology, particle spacing and humidity, construct attribute data table, and store it in the production stage database;

[0063] Use association rules to mine and analyze the potential associations between attribute data tables and historical screening graphs, and establish association rule paths;

[0064] Calculating the importance of each sample to be tested to the preset screening index based on the association rule path, and assigning a representative weight to each sample to be tested;

[0065] The samples to be detected are weighted and combined according to the representative weights to generate the weighted sample data cube.

[0066] Specifically, the cement production process is divided into multiple stages, such as raw material batching, clinker calcination, mill grinding, and finished product screening. N distributed sampling nodes (N = 10 in this example) are deployed at each stage. Automated sampling equipment (such as robotic arms or conveyor belt samplers) collects samples to be tested. The sampling frequency is set to twice per hour to ensure that samples cover dynamic production changes.

[0067] For each sample, a laser particle size analyzer, a microscope imaging system, and a humidity sensor are used to measure particle size distribution, particle morphology, interparticle spacing, and moisture. This attribute data is stored in a tabular format to generate an attribute data table, which is then stored in a production database managed using a relational database such as MySQL.

[0068] The Apriori algorithm is used to analyze potential associations between the attribute data table and historical screening maps (including the screening residue and screening conditions of historical samples). The historical screening maps are stored in a database and contain information such as screening residue, screen aperture, and production parameters. The mining process generates association rule paths.

[0069] Based on the association rule path, the importance of each sample to be tested to the preset screening index (in this example, the standard deviation of the screen residue) is calculated. The weight is calculated using the entropy method: the information entropy is calculated for each attribute, and the lower the entropy value, the greater the influence of the attribute on the screen residue. The entropy value is normalized to obtain the weight coefficient. Each sample is assigned a representative weight, with the weight value range being [0, 1].

[0070] Based on representative weights, the attribute data of the samples to be tested is weighted and combined to generate a three-dimensional data cube. The dimensions include sample ID, production stage, and attribute type. The data cube is stored as a multidimensional array for subsequent analysis.

[0071] Table 1 compares the sieve residue prediction performance of dynamic samples in the present invention and traditional static samples. The sample representativeness score is based on the overlap between the feature distributions of different sample sets (static and dynamic samples) and the full production dataset, while the cross-batch adaptability score is based on the difference in prediction accuracy between different sample models (static and dynamic samples) on different production batches that were not trained.

[0072] Table 1 Comparison of the prediction performance of dynamic samples of the present invention and traditional static samples

[0073] Prediction performance indicators Traditional static samples Dynamic sample of the present invention Cross-batch fitness scoring 9.1 9.2 Sample representativeness score 4.6 5.1 Screen residue prediction error 3.4％ 9.7％

[0074] By generating a weighted sample data cube, this method effectively quantifies sample representativeness and reduces prediction errors caused by sample bias. Compared to traditional uniform sampling methods, this method improves the pertinence and information utilization of sample data through association rule mining and weight assignment, significantly improving the accuracy of screen residue prediction in complex production scenarios.

[0075] Preferably, a dynamic screening log library is created based on the weighted sample data cube, and the multi-dimensional parameters in the dynamic screening log library are analyzed by the data processing model to construct the topological relationship of the screen state, such as Figure 3 As shown;

[0076] The process of constructing the topological relationship of the screen state includes:

[0077] Dynamically screening the weighted sample data cube, monitoring multi-dimensional parameters in real time, including sieve flow rate, sieve blockage degree and vibration frequency, and creating a dynamic screening log library;

[0078] Analyzing the multi-dimensional parameters, solving the congestion rate by simultaneous equations, and identifying and predicting the congestion trend;

[0079] A screen state topology relationship is constructed according to the clogging rate and the clogging trend.

[0080] Specifically, a weighted sample data cube is fed into a negative pressure sieve analyzer for dynamic screening. Multi-dimensional parameters, including sieve flow rate (the mass of particles passing through the sieve per unit time) and sieve blockage (the percentage of blockage area to total sieve area), are monitored in real time. The monitoring equipment includes a flow sensor and an image analysis system. The monitoring data is stored in a time series format, creating a dynamic screening log library.

[0081] Based on the sieve flow rate, screen blockage degree and vibration frequency, the following simultaneous equations are constructed:

[0082]

[0083] in, is the clogging rate; Q is the sieve hole flow rate; B is the sieve clogging degree; V is the vibration frequency; α i and β i is the adjustment coefficient;

[0084] will be different The state node corresponding to the vibration frequency V is defined as:

[0085] normal: And V>f min ;

[0086] Blockage:

[0087] In a graph database (such as Neo4j), the evolutionary topology of "normal→congested" is described using nodes and directed edges, and the transition probability of each edge is stored.

[0088] By building a high-precision dynamic screening log library and combining it with a simultaneous equation model, the degree of screen blockage and its evolution trend can be reflected in real time and accurately, providing refined and traceable status information for subsequent screen residue prediction through the state discrimination rule library, thereby significantly improving the stability and reliability of screen residue prediction.

[0089] Preferably, the sensor data stream is acquired through multi-source sensors, and a multi-dimensional indexing algorithm is used to calculate the similarity measure between the sensor data stream and the particle distribution database to generate a data correlation factor, such as Figure 4 As shown;

[0090] The process of generating the data association factor includes:

[0091] Normalizing the sensor data stream acquired by the multi-source sensors, the sensor data stream including humidity, temperature, air pressure, and dust concentration;

[0092] A multidimensional feature vector is constructed based on the sensor data stream, and a multidimensional indexing algorithm is used to perform similarity matching on a particle distribution database to obtain a similarity metric value, and K reference samples with the highest similarity metric values are extracted; the particle distribution database contains a plurality of paired samples of historical environmental characteristics and corresponding particle distribution curves;

[0093] The data association factor reflecting the relationship between the current environmental state and the historical particle distribution is generated in combination with the similarity measurement value.

[0094] Specifically, the sensor data streams including temperature, humidity, air pressure and dust concentration are standardized using the min-max method or the Z-score method to ensure that each feature has the same dimension.

[0095] The normalized sensor data streams are spliced into multidimensional feature vectors. Each environmental feature-particle distribution curve sample in the historical particle distribution database is also pre-constructed into a four-dimensional feature vector set. A multidimensional R-tree structure is used to store historical samples. During online queries, the similarity between the multidimensional feature vectors and the particle distribution database is quickly calculated based on Euclidean distance or cosine similarity, and the similarity is sorted to obtain a similarity metric value. The K reference samples with the highest similarity metric values are then extracted.

[0096] The data association factor is obtained by performing a weighted average on the retrieved K reference samples; the weight is obtained through multiple experiments.

[0097] Through efficient similarity comparison between multi-source sensor data and historical environmental particle distribution, the most representative reference samples can be dynamically extracted and quantifiable correlation factors can be formed, providing an accurate basis for environmental disturbance compensation and improving the predictive model's adaptability to external changes.

[0098] Preferably, the data relationship model is called to output the first prediction value, and a state discrimination rule base is established according to the screen state topological relationship and the data association factor; the state discrimination rule base specifically includes:

[0099] When the congestion trend does not meet the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the first rule is met;

[0100] When the congestion trend meets the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the second rule is met;

[0101] When the congestion trend does not meet the preset congestion condition and the data correlation factor is higher than the preset correlation threshold, it is determined to be in compliance with the third rule.

[0102] Specifically, the data relationship model uses a machine learning algorithm (such as random forest or gradient boosting tree) to establish a mapping relationship between the attribute data table and the sieve residue;

[0103] The attribute data table is standardized (mean is 0, variance is 1) and key features are extracted, such as D50, circularity, humidity, etc.

[0104] Use the random forest algorithm (number of trees = 100, maximum depth = 10) with the goal of minimizing the mean squared error;

[0105] The training process optimizes model parameters through 5-fold cross-validation to ensure generalization ability.

[0106] Call the trained data relationship model and perform the following steps to output the first predicted value:

[0107] Input the attribute data table in the weighted sample data cube into the model and process multiple samples in batches to improve efficiency;

[0108] The model calculates the predicted value of the sieve residue for each sample based on the key features of the input.

[0109] The predicted values are weighted averaged according to the representative weights of the samples to generate the first predicted value.

[0110] The specific judgment conditions of the state judgment rule base include:

[0111] First rule (slight perturbation):

[0112] Condition: Blockage rate and the data correlation factor is <0.3;

[0113] Judgment: Directly output the first predicted value.

[0114] Second rule (blocking dominant):

[0115] Condition: Blockage rate and the data correlation factor is <0.3;

[0116] Judgment: Perform the first compensation calibration, adjust the blockage deviation and then output the second prediction value.

[0117] The third rule (environmental disturbance):

[0118] Condition: Blockage rate And the data correlation factor is > 0.3;

[0119] Determination: Perform a second compensation calibration and output a third prediction value through the environmental impact model.

[0120] Based on the branch rule judgment of different dominant factors (mechanical blockage vs. environmental disturbance), targeted models can be accurately selected for prediction and calibration according to different production environments, overcoming the problem of insufficient robustness of a single model, improving the adaptability of the environment, and thus improving the overall prediction accuracy.

[0121] Preferably, when the first rule is met, the first predicted value is directly output; when the second rule is met, the first compensation calibration is performed based on the topological relationship of the screen state to generate the second predicted value;

[0122] The specific process of the first compensation calibration includes:

[0123] Combined with the topological relationship of the screen state, the deviation influence of the multi-dimensional parameters on the prediction of the screen residue is analyzed;

[0124] calculating a compensation amount for correcting the second predicted value according to the deviation influence;

[0125] The calculated compensation amount is combined with the first predicted value to generate a second predicted value that has undergone first compensation calibration.

[0126] Specifically, when the second rule is triggered, the multi-dimensional parameter deviations within the last 10 minutes are extracted;

[0127] Define the compensation function based on multi-dimensional parameter deviation:

[0128] C=k1·(ΔQ) 2 +k2·(ΔB) 2 +k3·(ΔV) 2 +k4·ΔQ·ΔB+k5·ΔQ·ΔV+k6·ΔB·ΔV;

[0129] Where C is the compensation amount; ΔQ is the sieve flow deviation; ΔB is the sieve blockage deviation; ΔV is the vibration frequency deviation; k1, k2, k3, k4, k5 and k6 are weights;

[0130] The compensation amount is added to the first predicted value to generate a second predicted value.

[0131] Table 2 is a comparison table of the ability of introducing blockage compensation to predict screen residue.

[0132] Table 2 Comparison of the ability of blockage compensation to predict screen residue

[0133] Prediction performance indicators No congestion compensation Introducing congestion compensation Congestion early recognition rate (%) Unable to identify in advance 90.6 Compensation trigger timeliness (ms) Unable to trigger compensation 38 Screen residue prediction error (%) 25.3 4.6

[0134] The first compensation calibration quantitatively analyzes and compensates for the blockage deviation in the screening process in real time. The second prediction value can effectively correct the model deviation, reduce the system error caused by blockage, and achieve more reliable near-term prediction.

[0135] Preferably, when the third rule is met, a second compensation calibration is performed through the environmental impact model to generate a third predicted value;

[0136] The specific process of the second compensation calibration includes:

[0137] Through the environmental impact model, a coupling mapping relationship between environmental factors and particle movement characteristics is constructed, including the temperature-diffusion rate function, humidity-aggregation coefficient function, air pressure-particle migration probability function, and dust concentration-sieve permeability function;

[0138] Applying the coupling mapping relationship to the current sensor data stream to dynamically calculate the environmental interference influence coefficient matrix;

[0139] Correcting the particle distribution model according to the influence coefficient matrix and inverting the trend of the sieve residue deviation during the screening process;

[0140] The first predicted value is offset-adjusted in combination with the inversion result to generate a third predicted value that has undergone a second compensation calibration.

[0141] Specifically, experimental fitting was performed on temperature-diffusion rate, humidity-aggregation coefficient, air pressure-migration probability, and dust concentration-permeability to obtain coupling mapping relationships;

[0142] The coupling mapping relationship is applied to the current sensor data stream to calculate the influence coefficient matrix. The matrix elements represent the weight of each environmental factor on the particle distribution. The matrix dimension is 4×N (N is the number of particle distribution characteristics). The calculation efficiency is optimized through matrix decomposition algorithm.

[0143] The particle distribution model is modified according to the influence coefficient matrix, and the offset trend of the sieve residue during the screening process is inverted using Monte Carlo simulation to obtain the offset;

[0144] Perform an offset adjustment on the first predicted value to generate a third predicted value.

[0145] Table 3 is a comparison table of the prediction performance of screen residue under environmental disturbance.

[0146] The environmental fluctuation intensity level is obtained by calculating the standard deviation and fluctuation coefficient of each environmental parameter in the current sensor data stream, and performing z-score normalization and weighting on the standard deviation and fluctuation coefficient to obtain the environmental fluctuation index;

[0147] The intensity of environmental fluctuations is divided into levels according to the environmental fluctuation index: an environmental fluctuation index between 0 and 1 is medium to high; an environmental fluctuation index between 1 and 2 is high; and an environmental fluctuation index greater than 2 is extremely high.

[0148] Table 3 Comparison of sieve residue prediction performance under environmental disturbance

[0149] Environmental fluctuation intensity level Prediction error of existing methods (%) Prediction error of the method of the present invention (%) high 11.4 4.8 medium to high 10.2 4.2 Very high 14.7 3.9 high 12.1 4.7 Very high 15.3 4.1

[0150] The second compensation calibration introduces environmental coupling mapping, which effectively quantifies the comprehensive impact of multiple environmental factors on particle motion and screening efficiency by coupling environmental factors with particle motion characteristics, making the prediction model have good environmental adaptability and generalization ability.

[0151] Preferably, weight vector aggregation is performed on the first predicted value, the second predicted value and the third predicted value, and a sieve residue prediction result set is output.

[0152] The output process of the sieve residue prediction result set includes:

[0153] integrating the first prediction value, the second prediction value, and the third prediction value generated according to different rules;

[0154] Weights are assigned based on the reliability and importance of each set of prediction values, and the three sets of prediction values are weightedly fused to obtain the comprehensive screening residue prediction value.

[0155] The comprehensive screen residue prediction value is output in a structured format as the final screen residue prediction result set.

[0156] Specifically, weights are assigned based on the reliability and importance of each set of prediction values, where the reliability is calculated as the inverse of the historical prediction error and the importance is determined according to the rule priority;

[0157] The first prediction value, the second prediction value and the third prediction value are weighted and summed according to the assigned weights to obtain a comprehensive screening residue prediction value.

[0158] Through multi-model weighted fusion, the advantages of each model are complemented, and congestion calibration and environmental compensation are comprehensively considered to achieve the fusion of multi-scenario prediction results, and finally output the sieve residue prediction results with higher accuracy and stability.

[0159] Example 2

[0160] In Example 1, the method proposed in the present invention successfully achieved high-precision screen residue prediction in complex production scenarios such as insufficient sample representativeness, frequent screen clogging, and environmental disturbances. To further verify the effectiveness of the present invention, the present example also performed high-precision screen residue prediction during cement production in a certain factory B.

[0161] A method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer comprises:

[0162] Build a production stage database, collect samples to be tested through distributed nodes, and obtain the attribute data table of the samples to be tested; use association rule mining to establish the correlation between the attribute data table and the historical screening map, and generate a weighted sample data cube;

[0163] The process of generating the weighted sample data cube includes:

[0164] The cement production process is divided into multiple production stages, and N distributed sampling nodes are configured for each production stage to collect samples to be tested;

[0165] Obtain attribute data of each sample to be tested, including particle size distribution, particle morphology, particle spacing and humidity, construct attribute data table, and store it in the production stage database;

[0166] Use association rules to mine and analyze the potential associations between attribute data tables and historical screening graphs, and establish association rule paths;

[0167] Calculating the importance of each sample to be tested to the preset screening index based on the association rule path, and assigning a representative weight to each sample to be tested;

[0168] The samples to be detected are weighted and combined according to the representative weights to generate the weighted sample data cube.

[0169] Preferably, a dynamic screening log library is created based on the weighted sample data cube, and the multi-dimensional parameters in the dynamic screening log library are analyzed by a data processing model to construct a topological relationship of the screen state;

[0170] The process of constructing the topological relationship of the screen state includes:

[0171] Dynamically screening the weighted sample data cube, monitoring multi-dimensional parameters in real time, including sieve flow rate, sieve blockage degree and vibration frequency, and creating a dynamic screening log library;

[0172] Analyzing the multi-dimensional parameters, solving the congestion rate by simultaneous equations, and identifying and predicting the congestion trend;

[0173] A screen state topology relationship is constructed according to the clogging rate and the clogging trend.

[0174] Preferably, a sensor data stream is acquired through multi-source sensors, and a similarity measure between the sensor data stream and the particle distribution database is calculated using a multidimensional indexing algorithm to generate a data correlation factor;

[0175] The process of generating the data association factor includes:

[0176] Normalizing the sensor data stream acquired by the multi-source sensors, the sensor data stream including humidity, temperature, air pressure and dust concentration;

[0177] A multidimensional feature vector is constructed based on the sensor data stream, and a multidimensional indexing algorithm is used to perform similarity matching on a particle distribution database to obtain a similarity metric value, and K reference samples with the highest similarity metric values are extracted; the particle distribution database contains a plurality of paired samples of historical environmental characteristics and corresponding particle distribution curves;

[0178] The data association factor reflecting the relationship between the current environmental state and the historical particle distribution is generated in combination with the similarity measurement value.

[0179] Preferably, the data relationship model is called to output the first prediction value, and a state discrimination rule base is established according to the screen state topological relationship and the data association factor;

[0180] The state discrimination rule base specifically includes:

[0181] When the congestion trend does not meet the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined to be in compliance with the first rule;

[0182] When the congestion trend meets the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the second rule is met;

[0183] When the congestion trend does not meet the preset congestion condition and the data correlation factor is higher than the preset correlation threshold, it is determined to be in compliance with the third rule.

[0184] Preferably, when the first rule is met, the data relationship model is called to output a first prediction value; when the second rule is met, a first compensation calibration is performed based on the screen state topological relationship to generate a second prediction value;

[0185] The specific process of the first compensation calibration includes:

[0186] Combined with the topological relationship of the screen state, the deviation influence of the multi-dimensional parameters on the prediction of the screen residue is analyzed;

[0187] calculating a compensation amount for correcting the second predicted value according to the deviation influence;

[0188] The calculated compensation amount is combined with the first predicted value to generate a second predicted value that has undergone first compensation calibration.

[0189] Preferably, when the third rule is met, a second compensation calibration is performed through the environmental impact model to generate a third predicted value;

[0190] The specific process of the second compensation calibration includes:

[0191] Through the environmental impact model, a coupling mapping relationship between environmental factors and particle movement characteristics is constructed, including the temperature-diffusion rate function, humidity-aggregation coefficient function, air pressure-particle migration probability function, and dust concentration-sieve permeability function;

[0192] Applying the coupling mapping relationship to the current sensor data stream to dynamically calculate the environmental interference influence coefficient matrix;

[0193] Correcting the particle distribution model according to the influence coefficient matrix and inverting the trend of the sieve residue deviation during the screening process;

[0194] The first predicted value is offset-adjusted in combination with the inversion result to generate a third predicted value that has undergone a second compensation calibration.

[0195] Preferably, weight vector aggregation is performed on the first predicted value, the second predicted value and the third predicted value, and a sieve residue prediction result set is output.

[0196] The output process of the sieve residue prediction result set includes:

[0197] integrating the first prediction value, the second prediction value, and the third prediction value generated according to different rules;

[0198] Based on the reliability and importance of each set of prediction values, weights are assigned, and the three sets of prediction values are weightedly fused to obtain the comprehensive screening residue prediction value;

[0199] The comprehensive screen residue prediction value is output in a structured format as the final screen residue prediction result set.

[0200] Table 4 is a comparison table of the sieve residue prediction performance of the overall solution of the present invention.

[0201] Among them, the sample representativeness score is obtained by calculating the proportion of the effective coverage dimension of the sample space in the theoretical feature space dimension;

[0202] The multi-operating-condition adaptability score is obtained by calculating the mean error across operating conditions;

[0203] The accuracy score of screen clogging perception was assessed by the clogging identification accuracy;

[0204] The effectiveness score of environmental disturbance compensation is obtained by comparing the difference in prediction error between uncompensated and compensated values;

[0205] The overall prediction accuracy score was calculated using the standard MAE formula.

[0206] Table 4 is a comparative table of the performance evaluation of the screen residue prediction of the overall solution of the present invention

[0207] Evaluation Metrics Existing methods Solution of the present invention Sample representativeness score 5.2 9.1 Multi-condition adaptability score 4.9 8.8 Screen clogging perception accuracy score 5.7 9.2 Environmental disturbance compensation effectiveness score 4.8 8.9 Overall prediction accuracy score 5.3 9.3

[0208] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamically predicting the sieve residue of a cement negative pressure sieve analyzer, characterized in that: include: Build a production stage database, collect samples to be tested through distributed nodes, and obtain the attribute data table of the samples to be tested; Use association rule mining to establish the correlation between the attribute data table and the historical screening map, and generate a weighted sample data cube; Create a dynamic screening log library based on the weighted sample data cube, analyze the multi-dimensional parameters in the dynamic screening log library through the data processing model, and construct the topological relationship of the screening state; The sensor data stream is acquired through multi-source sensors, and the similarity measure between the sensor data stream and the particle distribution database is calculated using a multidimensional indexing algorithm to generate a data correlation factor. Output the first predicted value through the data relationship model, and establish a state discrimination rule base based on the screen state topology relationship and data association factors: when the first rule is met, directly output the first predicted value; When the second rule is met, a first compensation calibration is performed based on the topological relationship of the screen state to generate a second predicted value; When the third rule is met, a second compensation calibration is performed through the environmental impact model to generate a third predicted value; Perform weight vector aggregation on the first predicted value, the second predicted value, and the third predicted value, and output a sieve residue prediction result set.

2. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The process of generating the weighted sample data cube includes: The cement production process is divided into multiple production stages, and N distributed sampling nodes are configured for each production stage to collect samples to be tested; Obtain attribute data of each sample to be tested, including particle size distribution, particle morphology, particle spacing and humidity, construct attribute data table, and store it in the production stage database; Use association rules to mine and analyze the potential associations between attribute data tables and historical screening graphs, and establish association rule paths; Calculating the importance of each sample to be tested to the preset screening index based on the association rule path, and assigning a representative weight to each sample to be tested; The samples to be detected are weighted and combined according to the representative weights to generate the weighted sample data cube.

3. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The process of constructing the topological relationship of the screen state includes: Dynamically screening the weighted sample data cube, monitoring multi-dimensional parameters in real time, including sieve flow rate, sieve blockage degree and vibration frequency, and creating a dynamic screening log library; Analyzing the multi-dimensional parameters, solving the congestion rate by simultaneous equations, and identifying and predicting the congestion trend; A screen state topology relationship is constructed according to the clogging rate and the clogging trend.

4. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The process of generating the data association factor includes: Normalizing the sensor data stream acquired by the multi-source sensors, the sensor data stream including humidity, temperature, air pressure, and dust concentration; A multidimensional feature vector is constructed based on the sensor data stream, and a multidimensional indexing algorithm is used to perform similarity matching on a particle distribution database to obtain a similarity metric value, and K reference samples with the highest similarity metric values are extracted; the particle distribution database contains a plurality of paired samples of historical environmental characteristics and corresponding particle distribution curves; The data association factor reflecting the relationship between the current environmental state and the historical particle distribution is generated in combination with the similarity measurement value.

5. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The state discrimination rule base specifically includes: When the congestion trend does not meet the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the first rule is met; When the congestion trend meets the preset congestion condition and the data correlation factor is lower than the preset correlation threshold, it is determined that the second rule is met; When the congestion trend does not meet the preset congestion condition and the data correlation factor is higher than the preset correlation threshold, it is determined to be in compliance with the third rule.

6. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The specific process of the first compensation calibration includes: Combined with the topological relationship of the screen state, the deviation influence of the multi-dimensional parameters on the prediction of the screen residue is analyzed; calculating a compensation amount for correcting the second predicted value according to the deviation influence; The calculated compensation amount is combined with the first predicted value to generate a second predicted value that has undergone first compensation calibration.

7. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The specific process of the second compensation calibration includes: Through the environmental impact model, the coupling mapping relationship between environmental factors and particle movement characteristics is constructed, including the temperature-diffusion rate function, humidity-aggregation coefficient function, air pressure-particle migration probability function, and dust concentration-sieve permeability function; Applying the coupling mapping relationship to the current sensor data stream to dynamically calculate the environmental interference influence coefficient matrix; Correcting the particle distribution model according to the influence coefficient matrix and inverting the trend of the sieve residue deviation during the screening process; The first predicted value is offset-adjusted in combination with the inversion result to generate a third predicted value that has undergone a second compensation calibration.

8. The method for dynamic prediction of sieve residue of cement negative pressure sieve analyzer according to claim 1, characterized in that: The output process of the screening residue prediction result set includes: integrating the first prediction value, the second prediction value and the third prediction value generated according to different rules; assigning weights based on the reliability and importance of each group of prediction values, and weightedly fusing the three groups of prediction values to obtain a comprehensive screening residue prediction value; outputting the comprehensive screening residue prediction value in a structured format as the final screening residue prediction result set.

Images