An artificial intelligence-based coal preparation plant production data management system

Abstract

The present application relates to the technical field of data processing, and more particularly to a coal preparation plant production data management system based on artificial intelligence, comprising: a data acquisition module for real-time acquisition of multi-modal data in the production process of the coal preparation plant; a time sequence fusion module for time alignment of the multi-modal data to construct a multi-modal data matrix; a feature analysis module for feature analysis of the multi-modal data matrix; a feature fusion module for analysis and fusion of feature vectors; a model construction module for construction of an anomaly scoring model based on the fused feature vectors to obtain an anomaly score; a data management module for judging data status and giving early warning according to the anomaly score; and a statistical analysis module for acquisition of abnormal event types after the early warning, statistical analysis of abnormal event probabilities, and updating of the judgment process of the data status according to the abnormal event probabilities and the multi-modal data. The present application realizes abnormal monitoring and processing of the production data of the coal preparation plant.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a coal preparation plant production data management system based on artificial intelligence. BACKGROUND

[0002] The main processing of coal preparation plant production data involves industrial big data, Internet of Things, cloud computing and other technologies. These technologies provide a full-process solution for data collection, cleaning, storage, analysis and visualization for coal preparation plants, helping them achieve intelligent management and decision-making, and improve production efficiency and economic benefits.

[0003] Chinese Patent Publication No. CN118446433A discloses an artificial intelligence coal production monitoring method based on an industrial internet platform, which includes: S1, collecting real-time coal production data based on an industrial internet platform; S2, analyzing and processing the real-time coal production data to establish a coal production correlation database; S3, establishing an artificial intelligence coal production monitoring and evaluation model based on the coal production correlation database; and S4, obtaining artificial intelligence coal production monitoring results using the artificial intelligence coal production monitoring and evaluation model. This invention realizes the construction of a correlation database for coal preparation plant production data to evaluate whether the data is abnormal, but does not realize the monitoring of production data in multiple dimensions such as comprehensive vibration data, raw material data and environmental data, and cannot dynamically analyze and judge the production data of the coal preparation plant in real time. It has the problems of low analysis efficiency of coal preparation plant production data and inaccurate monitoring of abnormal coal preparation plant production data. SUMMARY

[0004] The present application aims to provide a coal preparation plant production data management system based on artificial intelligence to solve at least one of the problems in the prior art.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] A coal preparation plant production data management system based on artificial intelligence, comprising:

[0007] A feature analysis module is used to analyze the features of the multi-modal data matrix to obtain vibration time domain features, vibration frequency domain features, raw material spectral features and process environment parameters;

[0008] A feature fusion module is used to analyze the fusion features of the multi-modal data, vibration time domain features, vibration frequency domain features, spectral features and process environment parameters to obtain a fusion feature vector;

[0009] A data management module is used to determine the data state and issue an early warning according to the abnormal score;

[0010] The statistical analysis module is used to collect the abnormal event type after the early warning, and statistically analyze the abnormal event probability, and update the judgment process of the data state according to the abnormal event probability and the multi-modal data.

[0011] Further, the system further comprises a data acquisition module used to acquire multi-modal data in the production process of the coal preparation plant in real time.

[0012] The time sequence fusion module is used to time align the multi-modal data to construct a multi-modal data matrix.

[0013] The model construction module is used to construct an abnormal score model based on the fusion feature vector to obtain an abnormal score.

[0014] Further, the time sequence fusion module time aligns the equipment data and the raw material data, the time sequence fusion module processes the equipment data using down-sampling, calculates the mean value, the standard deviation and the peak value of the equipment data within A seconds using A seconds as a window; the time sequence fusion module processes the raw material data using cubic spline interpolation to add A Hz time stamp in the raw material data; the time sequence fusion module arranges the time-aligned equipment data, the raw material data and the process data in time sequence to obtain a multi-modal data matrix, wherein A represents an alignment parameter.

[0015] Further, the feature analysis module is provided with a time domain analysis unit which is used to analyze the peak factor and the kurtosis according to the mean value, the variance and the peak value of the vibration signal to obtain the peak factor C and the kurtosis K.

[0016] The feature analysis module is further provided with a frequency domain analysis unit which is used to take the 1 / 3 octave energy proportion in the vibration signal as the vibration frequency domain feature.

[0017] Further, the feature analysis module is further provided with a raw material analysis unit which is used to take the ratio of the sulfur area to the raw coal ash area as the raw material spectrum feature.

[0018] The feature analysis module is further provided with a parameter construction unit which is used to construct a process environment parameter according to the medium density and the workshop humidity in the multi-modal data matrix, and the expression of the process environment parameter is S = p x l n(H + 1), wherein p represents the medium density, and H represents the workshop humidity.

[0019] Further, the feature fusion module performs fusion feature analysis on the vibration time domain feature, the vibration frequency domain feature, the spectrum feature and the process environment parameter according to the alignment parameter, the feature fusion module sets 16 different alignment parameters with a gradient of a times, analyzes the corresponding vibration time domain feature, vibration frequency domain feature, spectrum feature and process environment parameter, sets the feature vector corresponding to each alignment parameter as f, and sets f = [A, C, K, M, R, S, V, I].T wherein M represents the vibration frequency domain feature, R represents the raw material spectrum feature, V represents the flotation reagent flow, I represents the amplitude of the current signal, and 16 groups of feature vectors are fused to obtain a 128-dimensional fused feature vector F.

[0020] Further, the data management module compares and analyzes the abnormal probability to determine the data state. If the abnormal probability is greater than a first state threshold, the data management module determines that the data state is abnormal and performs a first-level early warning. If the abnormal probability is less than a second state threshold, the data management module determines that the data state is normal. If the abnormal probability is greater than or equal to the second state threshold and less than or equal to the first state threshold, the data management module determines that the data state is abnormal and performs a second-level early warning.

[0021] Further, the statistical analysis module is provided with an event collection unit, which is used to count the number of early warnings in Y time as the comprehensive abnormal quantity, and count the number of different abnormal event types as the abnormal event quantity. The ratio of the abnormal event quantity to the comprehensive abnormal quantity is taken as the abnormal event probability.

[0022] Further, the statistical analysis module is further provided with an event analysis unit, which is used to analyze the type loss degree according to the abnormal event probability to obtain a type loss degree B.

[0023] The event analysis unit updates the data state determination process according to the abnormal event probability and the type loss degree. If the abnormal event probability of the abnormal time type is no abnormality is greater than or equal to a no-abnormality threshold, the event analysis unit updates the data state determination process. When comparing and analyzing the abnormal probability, the event analysis unit sums the abnormal probability and the type loss degree of the abnormal time type of no abnormality, and compares the summed data with the first state threshold and the second state threshold to determine the data state. Otherwise, the event analysis unit does not update the data state determination process.

[0024] Further, the statistical analysis module is also provided with an environment compensation unit, which is used to extract the abnormal score of the data state as normal within 24 hours as a normal score, take the sum of the mean value of the normal score and 3 times the standard deviation of the normal score as an expected score, and calculate a compensation score according to the expected score and the second state threshold, set the compensation score = (the second state threshold - the expected score) / 2; the environment compensation unit triggers a threshold compensation mechanism after the judgment process of the data state is updated for J times in succession, the threshold compensation mechanism is to compensate the first state threshold and the second state threshold, when the workshop humidity is less than the humidity threshold, the first state threshold is compensated to the first state threshold + the compensation score, and the second state threshold is compensated to the second state threshold + the compensation score, when the workshop humidity is greater than or equal to the humidity threshold, the first state threshold is compensated to the first state threshold + y times the compensation score, and the second state threshold is compensated to the second state threshold + y times the compensation score; wherein, J represents a continuous parameter, and y represents a compensation parameter.

[0025] The beneficial effects of the present application are as follows: through real-time acquisition of multi-modal data by the data acquisition module and comprehensive analysis of the acquired data by other modules, multi-modal data fusion, nonlinear feature construction and adaptive decision mechanism are realized, the system realizes comprehensive multi-dimensional monitoring of abnormal conditions of the production data of the coal preparation plant in the dimensions of vibration signal analysis accuracy, raw material quality correlation and environmental interference robustness, improves the analysis efficiency of the system for the production data of the coal preparation plant, and improves the accuracy of the abnormal monitoring of the production data of the coal preparation plant. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0027] Figure 1 It is a structural schematic diagram of the coal preparation plant production data management system based on artificial intelligence of the present embodiment.

[0028] Figure 2 It is a structural schematic diagram of the feature analysis module of the present embodiment.

[0029] Figure 3 It is a structural schematic diagram of the model construction module of the present embodiment.

[0030] Figure 4 It is a structural schematic diagram of the statistical analysis module of the present embodiment. DETAILED DESCRIPTION

[0031] For more clearly illustrating the present application, the present application is further described below in conjunction with preferred embodiments and the accompanying drawings. Similar components are denoted by the same reference signs in the drawings. It should be understood by those skilled in the art that the following specific description is illustrative rather than limiting, and should not be used to limit the scope of protection of the present application.

[0032] It should be noted that although the terms first, second, third, etc. may be used in the embodiments of the present application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, the first can also be referred to as the second, and similarly, the second can also be referred to as the first, without departing from the scope of the embodiments of the present application.

[0033] Please refer to Figure 1 As shown in the figure, the coal preparation plant production data management system based on artificial intelligence in the embodiment comprises:

[0034] The data acquisition module is used to acquire multi-modal data in the production process of the coal preparation plant in real time. The multi-modal data comprises process data, equipment data, raw material data and environmental data. The process data comprises medium density and flotation reagent flow. The unit of the separation speed is meter per second, the unit of the medium density is gram per cubic centimeter, and the unit of the flotation reagent flow is liter per minute. The process data is the process data in the coal separation process using the dense medium coal separation method. The process data is uploaded by user interaction. The equipment data comprises vibration signals of the vibrating screen and current signals of the motor. The raw material data comprises raw coal ash area and sulfur area. The raw material data is a percentage value. The raw coal ash area is the characteristic peak area of the wavelength of 550 nm in the spectral data. The sulfur area is the characteristic peak area of the wavelength of 921 nm in the spectral data. The raw material data is the result of real-time detection of the LIBS spectrum. The environmental data is workshop humidity. The workshop humidity is relative humidity, which is a percentage value. The equipment data and the environmental data are acquired by sensors.

[0035] Specifically, the multi-modal data is synchronously acquired by the data acquisition module in the embodiment to construct a full-dimensional monitoring system, so as to solve the problem of response lag of traditional single parameter monitoring to raw material quality mutation.

[0036] Please continue to refer to Figure 1 As shown in the figure, the coal preparation plant production data management system based on artificial intelligence further comprises:

[0037] The time sequence fusion module is used to time-align the multi-modal data to construct a multi-modal data matrix. The time sequence fusion module is connected with the data acquisition module.

[0038] Specifically, the time sequence fusion module in the embodiment performs time alignment on the equipment data and the raw material data, the time sequence fusion module processes the equipment data using down-sampling, and calculates the mean value, the standard deviation and the peak value of the equipment data within A seconds using A seconds as a window; the time sequence fusion module processes the raw material data using cubic spline interpolation to add A Hz timestamps in the raw material data; and the time sequence fusion module arranges the time-aligned equipment data, the raw material data and the process data in time sequence to obtain a multi-modal data matrix, where A represents an alignment parameter.

[0039] Specifically, the time sequence fusion module in the embodiment performs time alignment (down-sampling, interpolation) on the multi-source data to generate a multi-modal data matrix, so as to retain key features and reduce computational load in the analysis process, avoid time faults of the raw material quality data, and reduce misjudgments caused by time sequence misalignment.

[0040] Specifically, the data in each column of the multi-modal data matrix in the embodiment represent time-aligned multi-modal data, which can be arranged in the order of "timestamp, equipment data, raw material data, process data".

[0041] Specifically, the alignment parameter in the embodiment is set to 1, and the value of the alignment parameter is not specifically limited in the embodiment, which can be freely set by those skilled in the art. The alignment parameter is a parameter for uniformly aligning the timestamps of the collected multi-modal data, so as to ensure that the corresponding time is consistent when the multi-modal data is analyzed.

[0042] Specifically, the embodiment is applied to a cloud end of a coal preparation plant data processing system, and is used for monitoring and processing process data abnormal problems caused by equipment wear, raw material component mutation and environmental parameter fluctuation in the production process of the coal preparation plant.

[0043] Please continue to refer to Figure 1 As shown in the figure, the artificial intelligence-based coal preparation plant production data management system further includes:

[0044] The feature analysis module is configured to perform feature analysis on the multi-modal data matrix to obtain vibration time domain features, vibration frequency domain features, raw material spectral features and process environmental parameters, and is connected to the time sequence fusion module.

[0045] Please refer to Figure 2 As shown in the figure, the feature analysis module includes:

[0046] The time domain analysis unit is configured to perform time domain feature analysis on the vibration signal in the multi-modal data matrix to obtain vibration time domain features, and the vibration time domain features include a peak factor and kurtosis.

[0047] Specifically, the time domain analysis unit in this embodiment analyzes the peak factor and kurtosis according to the mean, variance and peak value of the vibration signal, the expression of the peak factor is C = Max(L) / Avg(Avg(L)) 2 ) 1 / 2 ; the expression of the kurtosis is K = Avg({[L-Avg(L)] / sigma(L)} 4 ); wherein, L represents the amplitude of the vibration signal, Avg(L) represents the mean value of the vibration signal, sigma(L) represents the standard deviation of the vibration signal, and Max(L) represents the peak value of the vibration signal. It is defined that Avg() represents calculating the average value of the data in the parentheses, Max() represents extracting the maximum value of the data in the parentheses, and sigma() represents calculating the standard deviation of the data in the parentheses.

[0048] Please continue to refer to Figure 2 , the feature analysis module further comprises:

[0049] The frequency domain analysis unit is configured to perform frequency domain feature analysis on the vibration signal in the multi-modal data matrix to obtain vibration frequency domain features, and the frequency domain analysis unit is connected with the time domain analysis unit.

[0050] Specifically, the frequency domain analysis unit in this embodiment takes the 1 / 3 octave energy ratio in the vibration signal as the vibration frequency domain feature.

[0051] Please continue to refer to Figure 2 , the feature analysis module further comprises:

[0052] The raw material analysis unit is configured to perform spectral feature analysis on the raw material data in the multi-modal data matrix to obtain raw material spectral features, so as to reflect the quality fluctuation of raw coal in real time, and the raw material analysis unit is connected with the frequency domain analysis unit.

[0053] Specifically, the raw material analysis unit in this embodiment takes the ratio of sulfur area to raw coal ash area as the raw material spectral feature.

[0054] The parameter construction unit is configured to construct process environment parameters according to the process data and environmental data in the multi-modal data matrix, and the parameter construction unit is connected with the raw material analysis unit.

[0055] Specifically, the parameter construction unit in this embodiment constructs process environment parameters according to the medium density and workshop humidity in the multi-modal data matrix, and the expression of the process environment parameters is S = p x l n(H + 1), wherein p represents the medium density, and H represents the workshop humidity.

[0056] Specifically, the parameter construction unit in this embodiment quantifies the influence of humidity on medium viscosity to solve the failure problem of the traditional linear model in a high-humidity environment.

[0057] Please continue to refer to Figure 1 As shown in the figure, the coal preparation plant production data management system based on artificial intelligence further comprises:

[0058] The feature fusion module is configured to perform fusion feature analysis on the multi-modal data, the vibration time domain features, the vibration frequency domain features, the spectral features, and the process environment parameters to obtain a fusion feature vector. The feature fusion module is connected to the feature analysis module.

[0059] Specifically, the feature fusion module performs fusion feature analysis on the vibration time domain features, the vibration frequency domain features, the spectral features, and the process environment parameters according to the alignment parameters. The feature fusion module sets 16 different alignment parameters with a gradient of a times, analyzes the corresponding vibration time domain features, vibration frequency domain features, spectral features, and process environment parameters, sets the feature vector corresponding to each alignment parameter as f, and sets f = [A, C, K, M, R, S, V, I] T , where M represents the vibration frequency domain features, R represents the raw material spectral features, V represents the flotation reagent flow, and I represents the amplitude of the current signal. The 16 groups of feature vectors are fused to obtain a 128-dimensional fusion feature vector F. It can be understood that the value of a is not specifically limited in this embodiment, and can be freely set by those skilled in the art, such as 0.1, 0.2, 0.3, etc. a should be greater than 0 and less than or equal to 0.3.

[0060] Specifically, when the feature fusion module fuses the 16 groups of feature vectors, the feature vectors are horizontally spliced to obtain an 8*16 matrix, and the spliced feature vectors are taken as the fusion feature vectors.

[0061] Specifically, the feature fusion module in this embodiment fuses 16 groups of multi-dimensional features under the alignment parameters, sets 16 time windows with a gradient of a, to capture the coupling effect of short-time impact and long-time trend, and improve the recognition accuracy of complex faults.

[0062] Please continue to refer to Figure 1 As shown in the figure, the coal preparation plant production data management system based on artificial intelligence further comprises:

[0063] The model construction module is configured to construct an anomaly scoring model based on the fusion feature vector to obtain an anomaly score. The model construction module is connected to the feature fusion module.

[0064] Please refer to Figure 3 As shown in the figure, the model construction module comprises:

[0065] The preprocessing unit is configured to perform normalization processing on the fusion feature vector.

[0066] Specifically, the preprocessing unit in this embodiment performs normalization processing on each item of data in the fusion feature vector respectively, and normalizes each item of data in the fusion feature vector to the interval [0, 1].

[0067] Specifically, when normalizing the data, the preprocessing unit in this embodiment processes the data using a normalization formula, which is: X = (x - x min ) / (x max -x min ); in the formula, X represents the normalized data, x represents the data before normalization, x min represents the minimum value of the data before normalization, and x max represents the maximum value of the data before normalization.

[0068] Please continue to refer to Figure 3 As shown in FIG. 1, the model construction module further includes:

[0069] A model construction unit is configured to construct an anomaly score model based on a deep residual network to obtain an anomaly score, and the model construction unit is connected to the preprocessing unit.

[0070] Specifically, the model construction unit in this embodiment uses a deep residual network structure as the model structure, sets four residual blocks, each of which contains a one-dimensional convolution with a kernel size of 5 and a filter size of 64 and a sigmoid as an activation function, to construct an anomaly score model, and inputs the fusion feature vector as training data into the anomaly score model to output an anomaly probability p, p ∈ [0, 1].

[0071] Please continue to refer to Figure 1 As shown in FIG. 1, the coal preparation plant production data management system based on artificial intelligence further includes:

[0072] A data management module is configured to determine a data state and give an early warning according to the anomaly score, and trigger a hierarchical early warning based on the anomaly score, and the data management module is connected to the model construction module.

[0073] Specifically, the data management module in this embodiment performs comparison analysis on the anomaly probability to determine the data state. If the anomaly probability is greater than a first state threshold, the data management module determines that the data state is abnormal and gives a first-level early warning. If the anomaly probability is less than a second state threshold, the data management module determines that the data state is normal. If the anomaly probability is greater than or equal to the second state threshold and less than or equal to the first state threshold, the data management module determines that the data state is abnormal and gives a second-level early warning.

[0074] Specifically, in the embodiment, the first state threshold is set to 0.85, and the second state threshold is set to 0.6. The values of the first state threshold and the second state threshold are not specifically limited in the embodiment, and can be freely set by those skilled in the art, as long as the judgment on the data state is met.

[0075] Specifically, in the embodiment, when the data management module performs the first level early warning, it indicates that the production data in the coal preparation plant is seriously abnormal, and the staff should be warned to immediately detect each device in the production process to ensure production safety. When the second level early warning is performed, it indicates that the production data in the coal preparation plant may be abnormal, and the staff should reasonably arrange the inspection to ensure the production safety.

[0076] Please continue to refer to Figure 1 As shown in the figure, the coal preparation plant production data management system based on artificial intelligence further comprises:

[0077] The statistical analysis module is used to collect the type of abnormal events after early warning, statistically analyze the probability of abnormal events, and update the judgment process of the data state according to the probability of abnormal events and the multi-modal data. The statistical analysis module is connected with the data management module.

[0078] Please refer to Figure 4 As shown in the figure, the statistical analysis module comprises:

[0079] The event collection unit is used to collect the type of abnormal events after early warning, and statistically analyze the probability of abnormal events. The type of abnormal events includes but is not limited to medium leakage, screen clogging, and no abnormality, etc.

[0080] Specifically, in the embodiment, the event collection unit counts the number of early warnings in Y time as the number of comprehensive abnormalities, and counts the number of different abnormal event types as the number of abnormal events. The ratio of the number of abnormal events to the number of comprehensive abnormalities is taken as the probability of abnormal events.

[0081] Specifically, in the embodiment, the historical abnormal type distribution is counted by the time collection unit to optimize the sensitivity of the model to high-frequency faults.

[0082] Specifically, in the embodiment, the value of the statistical time parameter Y is not specifically limited, and can be freely set by those skilled in the art, such as setting Y to 1 month and 2 months, etc. The statistical time parameter should be at least 1 month.

[0083] Please continue to refer to Figure 4 As shown in the figure, the statistical analysis module further comprises:

[0084] The event analysis unit is used to analyze the type loss degree according to the probability of abnormal events, and update the judgment process of the data state according to the type loss degree. The event analysis unit is connected with the event collection unit.

[0085] Specifically, the event analysis unit in the embodiment analyzes the type loss degree according to the abnormal event probability, and the expression of the type loss degree is B = -β × (1 - Pb) 2 × lg(Pb), Pb represents a probability parameter, Pb = p × P(i) + (1 - p) × (1 - P(i)), P(i) represents an abnormal time probability, i represents an abnormal event type, β represents a loss rate, and 0.2 ≤ β ≤ 0.3. It can be understood that the value of the loss rate is not specifically limited in the embodiment, and a person skilled in the art can freely set it, as long as the analysis of the type loss degree is met. The optimal value of the loss rate is β = 0.25.

[0086] Specifically, the event analysis unit in the embodiment updates the judgment process of the data state according to the abnormal event probability and the type loss degree. If the abnormal time type is an abnormal event probability that is greater than or equal to a no-abnormal threshold, the event analysis unit updates the judgment process of the data state. When the abnormal probability is compared and analyzed, the abnormal probability and the type loss degree of the no-abnormal abnormal time type are summed, and the summed data is compared with a first state threshold and a second state threshold to judge the data state. Otherwise, the event analysis unit does not update the judgment process of the data state.

[0087] Specifically, the no-abnormal threshold is set to 0.3 in the embodiment. The value of the no-abnormal threshold is not specifically limited in the embodiment, and a person skilled in the art can freely set it, as long as the update of the judgment process of the data state is met.

[0088] Specifically, the type loss degree is introduced by the time analysis unit in the embodiment to weight low-probability but high-hazard events and reduce false negatives.

[0089] Please continue to refer to Figure 4 As shown in the figure, the statistical analysis module further includes:

[0090] The environmental compensation unit is configured to trigger a threshold compensation mechanism according to environmental data. The environmental compensation unit is connected with the event analysis unit.

[0091] Specifically, the environment compensation unit extracts the abnormal score when the data state is normal within 24 hours as the normal score, takes the sum of the mean of the normal score and 3 times the standard deviation of the normal score as the expected score, and calculates the compensation score according to the expected score and the second state threshold, and sets the compensation score = (the second state threshold - the expected score) / 2; the environment compensation unit triggers the threshold compensation mechanism when the original data state is the same as the updated data state after the judgment process of updating the data state for J times in succession, and the threshold compensation mechanism is to compensate the first state threshold and the second state threshold, when the workshop humidity is less than the humidity threshold, the first state threshold is compensated to the first state threshold + the compensation score, and the second state threshold is compensated to the second state threshold + the compensation score, when the workshop humidity is greater than or equal to the humidity threshold, the first state threshold is compensated to the first state threshold + y times the compensation score, and the second state threshold is compensated to the second state threshold + y times the compensation score; wherein J represents the continuous parameter, 5≤K≤10, y represents the compensation parameter, 0.7≤y≤0.9. In the embodiment, the humidity threshold is set to 0.7, and it can be understood that the values of the humidity threshold, the continuous parameter and the compensation parameter are not specifically limited in the embodiment, and a person skilled in the art can freely set them as long as the triggering of the compensation mechanism is met, and the best values of the continuous parameter and the compensation parameter are: J=7, y=0.8.

[0092] Specifically, the environment compensation unit updates the expected score based on the normal data every 24 hours in the embodiment, so as to dynamically adapt to the baseline drift caused by equipment aging, thereby improving the system stability.

[0093] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the embodiments of the present application. For those skilled in the art, other different forms of changes or variations can be made on the basis of the above description, and it is impossible to enumerate all the embodiments here, and any obvious changes or variations derived from the technical solutions of the present application still fall within the protection scope of the present application.

Claims

1. A coal preparation plant production data management system based on artificial intelligence, characterized in that, include: The data acquisition module is used to collect multimodal data in real time during the coal preparation plant's production process; The time-series fusion module is used to align multimodal data over time in order to construct a multimodal data matrix; The feature analysis module is used to perform feature analysis on the multimodal data matrix to obtain vibration time-domain features, vibration frequency-domain features, raw material spectral features, and process environment parameters; The feature fusion module is used to perform fusion feature analysis on multimodal data, vibration time-domain features, vibration frequency-domain features, spectral features, and process environment parameters to obtain a fused feature vector; The model building module is used to construct an anomaly scoring model based on fused feature vectors to obtain anomaly scores; The data management module is used to determine the data status and issue warnings based on anomaly scores; The statistical analysis module is used to collect the types of abnormal events after the warning, statistically analyze the probability of abnormal events, and update the judgment process of data status based on the probability of abnormal events and multimodal data. The feature analysis module is equipped with a time-domain analysis unit, which is used to analyze the peak factor and kurtosis based on the mean, variance and peak value of the vibration signal to obtain the peak factor C and kurtosis K. The feature analysis module is also equipped with a frequency domain analysis unit, which is used to take the 1 / 3 octave band energy ratio in the vibration signal as the vibration frequency domain feature. The feature analysis module also includes a raw material analysis unit, which is used to take the ratio of sulfur area to ash area of ​​raw coal as the spectral feature of raw material. The feature analysis module also includes a parameter construction unit, which is used to construct process environment parameters based on the medium density and workshop humidity in the multimodal data matrix. The expression for the process environment parameters is S=ρ×ln(H+1), where ρ represents the medium density and H represents the workshop humidity. The time-series fusion module performs time alignment on equipment data and raw material data. It processes the equipment data using downsampling, calculating the mean, standard deviation, and peak value of the equipment data within a window of A seconds. It also processes the raw material data using cubic spline interpolation to add an A Hz timestamp. Finally, it arranges the time-aligned equipment data, raw material data, and process data in chronological order to obtain a multimodal data matrix. The feature fusion module performs fusion feature analysis on vibration time-domain features, vibration frequency-domain features, spectral features, and process environment parameters based on alignment parameters. The module sets 16 different alignment parameters with a gradient of α, and analyzes their corresponding vibration time-domain features, vibration frequency-domain features, spectral features, and process environment parameters. The feature vector corresponding to each alignment parameter is set as f, and f = [A, C, K, M, R, S, V, I]. T Where M represents the vibration frequency domain characteristics, R represents the raw material spectral characteristics, V represents the flotation reagent flow rate, I represents the amplitude of the current signal, and A represents the alignment parameter. Sixteen sets of feature vectors are fused to obtain a 128-dimensional fused feature vector F.

2. The artificial intelligence-based coal preparation plant production data management system according to claim 1, characterized in that, The data management module compares and analyzes the anomaly probability to determine the data status. If the anomaly probability is greater than the first status threshold, the data management module determines the data status as abnormal and issues a first-level warning. If the anomaly probability is less than the second status threshold, the data management module determines the data status as normal. If the anomaly probability is greater than or equal to the second status threshold and less than or equal to the first status threshold, the data management module determines the data status as abnormal and issues a second-level warning.

3. The artificial intelligence-based coal preparation plant production data management system according to claim 2, characterized in that, The statistical analysis module is equipped with an event acquisition unit, which is used to count the number of warnings within a time period Y as the comprehensive number of anomalies, and to count the number of different types of abnormal events as the number of abnormal events. The ratio of the number of abnormal events to the comprehensive number of anomalies is used as the probability of abnormal events.

4. The artificial intelligence-based coal preparation plant production data management system according to claim 3, characterized in that, The statistical analysis module also includes an event analysis unit, which is used to analyze the type loss degree based on the probability of abnormal events to obtain the type loss degree B. The event analysis unit updates the data status judgment process based on the probability of abnormal events and the type loss degree. If the probability of an abnormal event with no abnormality for the abnormal time type is greater than or equal to the no-abnormality threshold, the event analysis unit updates the data status judgment process. When comparing and analyzing the abnormal probability, the abnormal probability is summed with the type loss degree of the abnormal time type of no abnormality, and the summed data is compared with the first state threshold and the second state threshold to determine the data status; otherwise, the event analysis unit does not update the data status judgment process.

5. The artificial intelligence-based coal preparation plant production data management system according to claim 4, characterized in that, The statistical analysis module also includes an environmental compensation unit, which extracts abnormal scores when the data status is normal within 24 hours as normal scores. The sum of the mean of the normal scores and three times the standard deviation of the normal scores is used as the expected score. A compensation score is calculated based on the expected score and a second state threshold, with the compensation score set as (second state threshold - expected score) / 2. After J consecutive data status updates, if the original data status is the same as the updated data status, the environmental compensation unit triggers a threshold compensation mechanism. This mechanism compensates for both the first and second state thresholds. When the workshop humidity is less than the humidity threshold, the first state threshold is compensated to the first state threshold + compensation score, and the second state threshold is compensated to the second state threshold + compensation score. When the workshop humidity is greater than or equal to the humidity threshold, the first state threshold is compensated to the first state threshold + y times the compensation score, and the second state threshold is compensated to the second state threshold + y times the compensation score. Here, J represents a continuous parameter, and y represents a compensation parameter.

Images