CN110609530A

CN110609530A - Data mining method and system for realizing working condition optimization based on DCS system edge

Info

Publication number: CN110609530A
Application number: CN201910901372.4A
Authority: CN
Inventors: 沈志刚; 沈娅芳
Original assignee: Xiamen Huaxia International Power Development Co Ltd
Current assignee: Xiamen Huaxia International Power Development Co Ltd
Priority date: 2019-09-23
Filing date: 2019-09-23
Publication date: 2019-12-24
Anticipated expiration: 2039-09-23
Also published as: CN110609530B

Abstract

The invention provides a data mining method and a data mining system for realizing working condition optimization based on DCS system edges. According to the data mining method for realizing the working condition optimization based on the DCS system edge, the real-time variable of the alternative fitting quantity X is acquired through the DCS system acquisition device, and the input alternative fitting quantity X is associated with the early warning decision, so that historical data mining and the construction of a fault detection decision tree are carried out, and the purpose of analyzing the potential fault of the real-time working condition without an additional detection device is achieved; the fault detection is timely, the detection precision is high, and the method has good market application prospect.

Description

Data mining method and system for realizing working condition optimization based on DCS system edge

Technical Field

The invention relates to the field of industrial system fault detection, in particular to a data mining method and a data mining system for realizing working condition optimization based on DCS system edges.

Background

In the current large-scale industrial enterprises such as power enterprises, petrochemical enterprises and the like, along with the improvement of internal and external requirements such as environment requirements, efficiency requirements and the like, production service systems of the large-scale industrial enterprises are developed to be extremely complex, the requirements on precise control are stricter, and a DCS (distributed control system) is used as a general control core in most industrial production industry fields and needs to be continuously optimized, upgraded and adapted to new requirements.

However, for a large industry such as an electric power production enterprise, the whole thermodynamic cycle system involves an extremely large number of collected variables such as pressure, temperature, material level and the like, due to the limitation of a variable programming language of a DCS control system part, sequential logic control can only be completed in an adjacent time sequence range, and analysis of high-dimensional variable data and long-time data is often difficult. The common method is to divide the service subsystem and analyze the whole thermodynamic cycle system in a segmented and real-time manner, the analysis result is often rough, and the analysis error of the front-end system is often brought into the rear-end system. In order to eliminate analysis errors of part of key business systems, variable acquisition equipment is often required to be added to improve analysis control accuracy, but the production cost is high due to the fact that the equipment is added without limitation, the installation space of new equipment is limited due to the high integration degree of part of business systems, the heat value of the coal as fired is in a low-precision technical level at present due to the fact that some more special variable acquisition equipment is real-timely, and the actual effect of adding investment is not obvious. For years, production experts identify, analyze and judge long-period production data through tools such as historical trend graphs and the like according to own experiences by depending on a human-computer interaction interface. For a long time, the production business data of large industrial enterprises are considered to lack information islands and data deserts for effective mining and utilization.

With the integration of statistical data analysis and the industrial field, analytical assistance is beginning to be provided for multivariable complex control problems of the industrial DCS control system. However, the analysis means needs to complete the prior construction investment of DCS production data transmission, which is measured in millions of yuan, and along with the social events such as large-scale power failure caused by network security problems, the investment cost of security facilities of the production service system in the key industrial field and the outer database is increased year by year, and is also greatly limited in daily use, and in recent years, reverse data transmission from the outer data analysis system to the production service system is basically prohibited. Therefore, the method of collecting and developing professional data analysis based on secondary transmission of production process data is often a temporary centralized construction, basically only has an analysis prompt function, does not have a control function, lacks continuity, and cannot be developed for multiple times in real time. The industrial production requirements for working condition fault detection under the condition that a production business system cannot be added with variable acquisition equipment cannot be met.

Disclosure of Invention

In order to solve the problems mentioned in the background art, the invention provides a data mining method and a system for realizing condition optimization based on a DCS (distributed control system) edge, wherein the data mining method for realizing the condition optimization based on the DCS edge specifically comprises the following steps:

s1, analyzing the alternative fitting quantity X of the under-acquired working condition according to the control requirements of the production service system and equipment; collecting real-time variables of the alternative fitting quantity X as input items through a DCS system collecting device;

s2, analyzing and selecting the controlled condition of the fitting quantity X in the production process, confirming each boundary condition of fitting quantity change caused by the fitting working condition, and carrying out real-time acquisition and/or secondary processing on each boundary condition through a DCS acquisition device to form a confirmation set K;

s3, establishing a training set, taking the fitting working condition period change of each complete production process as a learning period, expecting the convergence direction of the alternative fitting quantity X, and estimating a reverse initial value;

s4, designing an iterative learning loop in the DCS, after the alternative fitting quantity X in each learning period is judged to be effective by a confirmation set K, carrying out convergence direction preferential with the initial estimation value, registering the preferential value as an intermediate preferential value, accumulating the learning rate lambda, and if the K judgment is invalid, not learning;

s5, setting a sectional amplitude limiting interval of a learning iterative process of the DCS; performing large-amplitude iteration in the initial learning period, and reducing the amplitude limiting threshold value in the middle and later periods for fine iteration;

s6, setting a middle-valued register retrieval branch in a DCS learning loop, and judging the learning rate lambda threshold of the register retrieval branch according to the single learning absolute time of a fitting working condition;

s7, setting a learning rate lambda termination value according to the single learning absolute time of the fitting working condition, terminating the learning iteration of the DCS by reaching a threshold value, and obtaining a historical optimal value of the alternative fitting quantity X;

s8, establishing a decision tree by using the historical optimal value, analyzing the potential fault of the real-time working condition, and forming a DCS decision early warning and control loop.

Further, the method comprises the following steps:

s1, analyzing alternative fitting quantity X of loading and unloading working conditions according to the control requirement of air compressor equipment, and taking air compressor motor current and air compressor stability approaching time acquired by a DCS system acquisition device as input quantities;

s2, analyzing the controlled conditions of the current of the motor of the air compressor and the stability approaching time of the air compressor in the production process, confirming various boundary conditions of the current of the motor of the air compressor and the stability approaching time change of the air compressor caused by the fitting working condition, and carrying out real-time collection and/or secondary processing on the boundary conditions through a DCS (distributed control system) collection device to form a confirmation set K; the confirmation set K at least comprises an air compressor comprehensive alarm S and a loading instruction D_lAir compressor running signal R, air compressor current quality signal Q and exhaust pressure P₁Pressure P of the separator₂Pressure P of pipe network₃Air compressor single machine output M, air compressor regulating valve position Z and unloading instruction D_uAir compressor loading current I_lAir compressor unloading current I_u；

S3, establishing a training set, taking the fitting working condition period change of each complete production process as a learning period, expecting the convergence direction of the motor current of the air compressor and the stability approaching time of the air compressor, and estimating a reverse initial value;

s4, designing an iterative learning loop in the DCS, after the current of the air compressor motor in each learning period is judged to be effective through a first learning confirmation coefficient, carrying out convergence direction optimization on the current and an initial estimation value, registering the optimization value as an intermediate optimization value, accumulating the learning rate lambda, and if the first learning confirmation coefficient is judged to be invalid, not learning;

after the stability approaching time of the air compressor in each learning period is judged to be effective by the second learning confirmation coefficient, the optimization of the convergence direction is carried out with the initial estimation value, the preferred value is registered as an intermediate preferred value, the learning rate lambda is accumulated, and the air compressor is not learned if the second learning confirmation coefficient is judged to be ineffective;

s6, setting a middle-valued register retrieval branch in a DCS loop, and judging the learning rate lambda threshold of the register retrieval branch according to the single learning absolute time of a fitting working condition;

Further, in the loading process of the air compressor, the current of the motor of the air compressor is the loading current I of the motor of the air compressor_lnThe stability approaching time of the air compressor is the loading stability approaching time T of the air compressor_ln；

In the step S4, the first learning confirmation coefficient is K_l，K_lThe formula of (1) is as follows:

wherein:

the above-mentionedComprehensively alarming the air compressor S to be not a signal; the described Δ P₁Is the exhaust pressure P₁A stability approaching confirmation signal; the described Δ P₃For pipe network pressure P₃A stability approaching confirmation signal; the delta Z is a stability approaching confirmation signal of the regulating valve position; and the delta I is a current stability approaching confirmation signal.

In the unloading process of the air compressor, the current of the motor of the air compressor is the unloading current I of the motor of the air compressor_unThe stability approaching time of the air compressor is the unloading stability approaching time T of the air compressor_un；

The first learning confirmation coefficient is K in step S4_u，K_uThe formula of (1) is as follows:

wherein:

the above-mentionedComprehensively alarming the air compressor S to be not a signal; the described Δ P₂Is the separator pressure P₂The pressure stabilizing confirmation signal; and the delta I is a current stability approaching confirmation signal.

Further, during loading, the second learning confirmation coefficient is K_lT，K_lTThe formula of (1) is as follows:

wherein: the above-mentionedComprehensively alarming the air compressor S to be not a signal; r is an air compressor running signal; q is a current quality signal of the air compressor; the described Δ P₃For pipe network pressure P₃A stability approaching confirmation signal; the described Δ I_lConfirming a signal for the stability trend of the loading current; d is the loaded real-time current and the preferred value I of the loaded current_lThe values are close;

during unloading, the second learning confirmation coefficient is K_uT，K_uTThe formula of (1) is as follows:

wherein, theComprehensively alarming the air compressor S to be not a signal; r is an air compressor running signal; q is a current quality signal of the air compressor; the described Δ P₂Is the separator pressure P₂The pressure stabilizing confirmation signal; the described Δ I_uConfirming a signal for the stability trend of the unloaded current; d is the optimized value I of the unloading real-time current and the loading current_lThe values are close.

Further, the method for analyzing the potential fault of the real-time working condition and forming the DCS decision early warning comprises the following steps:

s71, activating an early warning decision loop when the number of times of learning the optimum value of the loading current of the air compressor, the number of times of learning the optimum value of the loading stability tending time, the number of times of learning the unloading current of the air compressor and the number of times of learning the optimum value of the unloading stability tending time of the air compressor reach set times according to the learning degree of the loading and unloading working conditions of the air compressor;

s72, inputting the real-time air compressor loading current I_lnAir compressor discharge current I_unTime to steady state loading T_lnUnloading stability approaching time T_unPreferred value of the load current I_lThe preferred value of the discharge current I_uLoading the optimized value T of driving stability time_lUnloading stabilization driving time optimal value T_u；

S73, taking an air compressor comprehensive alarm S non-signal, an air compressor running signal R, an air compressor current quality signal Q and an air compressor current driving stability signal delta I as a loop decision confirmation condition K_A；

S74, generating Delta I_l＝I_ln-I_l、ΔT_l＝T_ln-T_l、ΔI_u＝I_un-I_u、ΔT_u＝T_un-T_uCalculating judgment branches by four difference values, and establishing a plurality of judgment threshold values;

and S75, outputting a normal or abnormal working condition analysis signal according to the judgment threshold value.

According to the data mining method for realizing the working condition optimization based on the DCS system edge, the real-time variable of the alternative fitting quantity X is acquired through the DCS system acquisition device, and the input alternative fitting quantity X is associated with the early warning decision, so that historical data mining and the construction of a fault detection decision tree are carried out, and the purpose of analyzing the potential fault of the real-time working condition without an additional detection device is achieved; the fault detection is timely, the detection precision is high, and the method has good market application prospect.

The invention also provides a data mining system for realizing the working condition optimization based on the DCS edge, which is used for executing the data mining method for realizing the working condition optimization based on the DCS edge, and comprises a loading current optimization fitting module, a loading stability tending time optimization fitting module, an unloading current optimization fitting module, an unloading stability tending time optimization fitting module and an early warning module; wherein:

the output end of the loading current optimization fitting module is respectively connected with the input end of the loading stability tending time optimization fitting module and the early warning module; the output end of the loading stability approaching time optimal fitting module is connected with the input end of the early warning module;

the output end of the unloading current optimization fitting module is respectively connected with the input end of the unloading stability tending time optimization fitting module and the input end of the early warning module; the output end of the unloading stability tending time optimization fitting module is connected with the input end of the early warning module.

Further, the loading current preference fitting module AND the unloading current preference fitting module each include a first SFT switching module, a third SFT switching module, a fourth SFT switching module, a fifth SFT switching module, a sixth SFT switching module, a seventh SFT switching module, an eighth SFT switching module, a ninth SFT switching module, a tenth SFT switching module, a first DELAY hysteresis operation module, a third DELAY hysteresis operation module, a first AND logic AND module, a first hlcnt counting function, a first HLLMT clipping module, a first HLALM high-low limit judgment module, a second HLALM high-low limit judgment module, a third HLALM high-low limit judgment module, a fourth HLALM high-low limit judgment module, a fifth HLALM high-low limit judgment module, a sixth HLALM high-low limit judgment module, a first tristel selector module, a first ADD-subtract-q quality judgment module, a first NOT-type logic non-module, a second NOT-type logic non-function module, a first ADD-subtract-type logic module, a second NOT-type ADD-drop calculation module, a first ADD-type-q quality judgment module, a, A fifth NOT logical NOT module;

the output end of the first SFT switching module is connected with the input end of the first DELAY lag operation module and the input end of the first HLLMT amplitude limiting module; the output end of the first DELAY lag operation module AND the output end of the third DELAY lag operation module are respectively connected with the input end of a first AND logic AND module; the first TQ quality judgment module is connected with the first AND logic connection module through a first NOT logic negation module; the second NOT logical negation module is connected with the first AND logical negation module;

the output end of the first AND logic AND module is respectively connected with the input end of a fourth SFT switching module AND the input end of a first CNT counting module;

the output end of the fourth SFT switching module is connected with the input end of the first TRISEL selector module; the output end of the first HLLMT amplitude limiting module is connected with the input end of a fourth SFT switching module; the output end of the first TRISEL selector module is respectively connected with a sixth SFT switching module, a seventh SFT switching module, an eighth SFT switching module, a ninth SFT switching module, a tenth SFT switching module and a first ADD addition and subtraction calculation functional block;

the output end of the first CNT counting module is respectively connected with the input ends of a first HLALM high-low limit judging module, a second HLALM high-low limit judging module, a third HLALM high-low limit judging module, a fourth HLALM high-low limit judging module, a fifth HLALM high-low limit judging module and a sixth HLALM high-low limit judging module; the output end of the first HLALM high-low limit judging module sequentially passes through a third SFT switching module, a first ADD addition and subtraction calculation functional block and the input end of the first HLLMT amplitude limiting module; the sixth HLALM high-low limit judgment module is connected with the input end of the first AND logic AND module through a fifth NOT logic negation module;

the sixth HLALM high-low limit judging module is connected with the input end of the first SEL module through a tenth SFT switching module; the fifth HLALM high-low limit judgment module is connected with the input end of the first SEL module through the ninth SFT switching module; the fourth HLALM high-low limit judging module is connected with the input end of the first SEL module through an eighth SFT switching module; the third HLALM high-low limit judging module is connected with the input end of the first SEL module through a seventh SFT switching module; the second HLALM high-low limit judging module is connected with the input end of the first SEL module through a sixth SFT switching module;

the output end of the first SEL module is connected with the input end of the first TR I SEL selector module through a fifth SFT switching module;

the loading current preference fitting module further comprises a second SFT switching module, a second DELAY hysteresis operation module, a fourth DELAY hysteresis operation module and a first NOR logic NOR module; the second SFT switching module is connected with the input end of the first AND logic AND module through a second DELAY lag operation module; the fourth DELAY operation module AND the first NOR logic NOR module are respectively connected with the input end of the first AND logic AND module.

Further, the loading stability tending time preference fitting module AND the unloading stability tending time preference fitting module both include a second AND logic AND module, a third AND logic AND module, a fourth AND logic AND module, a first TSUMD module, a first DEV deviation comparison module, a first RS flip-flop module, a fifth DELAY operation module, a sixth DELAY operation module, a second CNT counting function, a seventh HLALM high-low limit judgment module, an eighth HLALM high-low limit judgment module, a ninth HLALM high-low limit judgment module, a tenth HLALM high-low limit judgment module, an eleventh SFT switching module, a twelfth SFT switching module, a thirteenth SFT switching module, a fourteenth SFT switching module, a fifteenth SFT switching module, a sixteenth SFT switching module, a seventeenth SFT switching module, an Add module, a second HLLMT clipping module, a second TRI SEL selector module, a second SEL module, a second TQ quality judgment module, A third NOT logical negation module, a fourth NOT logical negation module and a second ADD addition and subtraction calculation functional block;

the second AND logic AND module is connected with a set end of the first TSUMD module; the loading current preference fitting module or the unloading current preference fitting module is connected with the first DEV deviation comparison module; the first DEV deviation comparison module is connected with a signal end of the first TSUMD module sequentially through the fourth AND logic AND module AND the first RS trigger module; the first TSUMD module is connected with the second HLLMT amplitude limiting module, the eleventh SFT switching module and the second TRI SEL selector module in sequence; the output end of the second TRI SEL selector module is respectively connected with the twelfth SFT switching module and the first Add module;

the seventeen SFT switching module is connected with the input end of the third AND logic AND module through a sixth DELAY operation module; the fifth DELAY lag operation module is connected with the input end of the third AND logic AND module; the second TQ quality judgment module is connected with the input end of the third AND logic AND module through a third NOT logic negation module; the fourth NOT logical negation module is connected with the input end of the third AND logical AND module;

the output end of the third AND logic AND module is respectively connected with the input end of the fourth AND logic AND module, the second CNT counting function AND the eleventh SFT switching module;

the second CNT counting function is respectively connected with the input ends of a seventh HLALM high-low limit judgment module, an eighth HLALM high-low limit judgment module, a ninth HLALM high-low limit judgment module and a tenth HLALM high-low limit judgment module

The tenth HLALM high-low limit judgment module is connected with the second HLLMT amplitude limiting module sequentially through the thirteenth SFT switching module and the second ADD addition and subtraction calculation functional block;

the seventh HLALM high-low limit judging module is connected with the fourteenth SFT switching module; the eighth HLALM high-low limit judging module is connected with the fifteenth SFT switching module; the output end of the ninth HLALM high-low limit judging module is respectively connected with the first TSUMD module and the sixteenth SFT switching module; and the fourteenth SFT switching module, the fifteenth SFT switching module and the sixteenth SFT switching module are connected with the twelfth SFT switching module through the second SEL module.

Further, the early warning module includes a fifth AND logic AND module, a sixth AND logic AND module, a seventh AND logic AND module, an eighth AND logic AND module, a ninth AND logic AND module, a tenth AND logic AND module, an eleventh AND logic AND module, a twelfth AND logic AND module, a thirteenth AND logic AND module, a fourteenth AND logic AND module, a fifteenth AND logic AND module, a sixteenth AND logic AND module, a seventeenth AND logic AND module, a second TSUMD module, a third TSUMD module, a second RS flip-flop module, a third RS flip-flop module, a second DEV deviation comparison module, a third DEV deviation comparison module, a fourth DEV deviation comparison function block, a fifth DEV deviation comparison function block, an eleventh HLALM high-low limit determination module, a twelfth HLALM high-low limit determination module, a thirteenth HLALM high-low limit determination module, a fourteenth HLALM high-low limit determination module, a fourth DEV high-low limit determination module, a sixth AND logic AND module, a thirteenth AND logic AND module, a sixteenth AND logic AND module, A seventh NOT logic negation module, an eighth NOT logic negation module, a ninth NOT logic negation module and a tenth NOT logic negation module;

the loading current optimal fitting module AND the loading stability approaching time optimal fitting module are respectively connected with the second TSUMD module, the fifth AND logic AND module, the sixth AND logic AND module, the seventh AND logic AND module, the eighth AND logic AND module AND the fourteenth AND logic AND module through the fifteenth AND logic AND module;

the unloading current optimal fitting module AND the unloading stability tending time optimal fitting module are respectively connected with the third TSUMD module, the ninth AND logic AND module, the tenth AND logic AND module, the eleventh AND logic AND module, the twelfth AND logic AND module AND the thirteenth AND logic AND module through a sixteenth AND logic AND module; the third TSUMD module is connected with the third DEV deviation comparison module; the third DEV deviation comparison module is respectively connected with the eleventh function and module and the twelfth function and module;

the seventeenth AND logic AND module is respectively connected with the second RS trigger module, the fifth AND logic AND module, the fourteenth AND logic AND module AND the third RS trigger module; the second RS trigger module, the second TSUMD module and the second DEV deviation comparison module; the second DEV deviation comparison module is respectively connected with a seventh function and module and a sixth function and module; the third RS trigger module is connected with a third TSUMD module; the fifth AND logic AND module AND the sixth AND logic AND module are connected with the first OR logic module; the tenth AND logic AND module AND the eleventh AND logic AND module are connected with the second OR logic module;

the fourth DEV deviation comparison functional block is respectively connected with a ninth NOT logic negation module, a twelfth HLALM high-low limit judgment module, a fourteenth AND logic AND module AND a fifth AND logic AND module; the ninth NOT logical negation module is respectively connected with the sixth AND logical AND module AND the seventh AND logical AND module; the twelfth HLALM high-low limit judgment module is respectively connected with the fifth AND logic AND module AND the eighth AND logic AND module; a tenth NOT logical negation module is arranged between the twelfth HLALM high-low limit judging module AND the fifth AND logical AND module;

the fifth DEV deviation comparison functional block is respectively connected with a thirteenth high-low limit judgment module, an eighth NOT logical negation module, a ninth AND logical AND module AND a tenth AND logical AND module; the thirteenth high-low limit judging module is respectively connected with the ninth AND logic AND module AND the thirteenth AND logic AND module; a seventh NOT logical negation module is arranged between the thirteenth high-low limit judging module AND the ninth AND logical negation module; AND the eighth NOT logical negation module is respectively connected with the eleventh AND logical AND module AND the twelfth AND logical AND module.

The invention also provides a data mining system for realizing the working condition optimization based on the DCS system edge, and the connection of all modules realizes the purposes of acquiring the real-time variable of the alternative fitting quantity X by the DCS system acquisition device and associating the input alternative fitting quantity X with the early warning decision, thereby solving the problem that the real-time input variable of the alternative fitting quantity X acquired by the existing DCS system is difficult to associate with the early warning decision. The invention further provides a data mining system for realizing working condition optimization based on the DCS edge, fault detection is timely, and detection precision is high.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

FIG. 1 is a block diagram of a data mining method for implementing condition optimization based on DCS system edge;

FIG. 2 is a loading merit fitting decision tree of the data mining method for implementing condition optimization based on the edge of the DCS system according to the present invention;

FIG. 3 is an unloading figure of merit fitting decision tree of the data mining method for implementing condition optimization based on DCS system edge provided by the present invention;

FIG. 4 is a real-time condition decision tree of the data mining method for implementing condition optimization based on the DCS system edge;

FIG. 5 is a fitting logic diagram of air compressor unloading current optimization in the data mining system based on the DCS system edge to realize the working condition optimization;

FIG. 6 is a preferable fitting logic diagram of the unloading stability trending time of the air compressor in the data mining system for realizing the working condition optimization based on the DCS system edge;

FIG. 7 is an optimal fitting logic of the loading current of the air compressor in the data mining system for implementing the working condition optimization based on the edge of the DCS, provided by the invention;

FIG. 8 is a preferred fitting logic of the loading stability approaching time of the air compressor in the data mining system for implementing the working condition optimization based on the DCS system edge;

FIG. 9 is a decision logic diagram of the air compressor loading and unloading early warning in the data mining system for implementing the working condition optimization based on the DCS system edge.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In the description of the present invention, it should be noted that the terms "connected" or "coupled" and the like are not limited to physical or mechanical connections, but may include electrical connections, optical connections and the like, whether direct or indirect.

The invention provides a data mining method applied to working condition detection, a DCS (distributed control system) and an application, wherein the data mining method applied to the working condition detection specifically comprises the following steps:

In specific implementation, a common air compressor station is aimed at in the embodiment, a set of pneumatic units for controlling loading and unloading are generally designed in the equipment, the pneumatic execution units generally adopt 2-3 independent electromagnetic valves to control the on-off of loading and unloading air paths, and therefore the loading, compression, work doing and unloading idling of the whole air compressor equipment are realized. Being limited to the trend of current air compressor machine high integration, miniaturization, inside pneumatic unit does not design more pressure acquisition component, and the internal leakage that can't effective monitoring add uninstallation gas circuit component trouble and lead to, and then causes most air compressors to take place following trouble: the air compressor cannot do work or the compression output is limited due to the blockage of the loading air path, the separator cannot dissipate pressure and generate oil vomit due to the blockage of the unloading air path, the air inlet of the compressor does unnecessary work during the unloading idling due to the leakage of the loading air path, and the output of the compressor is reduced due to the bypass caused by the leakage of the unloading air path.

The invention provides a data mining method for realizing working condition optimization based on DCS system edge, which respectively completes loading optimal value fitting, unloading optimal value fitting and real-time working condition decision after fitting through three decision trees, wherein the following processes are respectively the loading optimal value fitting, unloading optimal value fitting and real-time working condition decision after fitting, and a loading current optimization fitting module, a loading stability tending time optimization fitting module, an unloading current optimization fitting module, an unloading stability tending time optimization fitting module and an early warning module are all independent modules:

firstly, loading a figure of merit fitting:

s1, analyzing the alternative fitting quantity of the loading working condition according to the control requirement of the air compressor equipment, acquiring the real-time variable of the alternative fitting quantity X as an input item through a DCS (distributed control system) acquisition device, and setting I as the input quantity by using the current of a loading motor of the air compressor and the stability approaching time of loading as the input quantity_lnAnd T_ln；

S2, analyzing the controlled condition in the loading process of the alternative fitting quantity, and listing the I caused by the fitting working condition_lnAnd T_lnThe method comprises the steps of collecting various changed learning boundary conditions in real time and/or carrying out secondary treatment on the boundary conditions through a DCS (distributed control system) collection device to form a confirmation set K, wherein the confirmation set K comprises an air compressor comprehensive alarm S and a loading instruction D_lAir compressor running signal R, air compressor current quality signal Q and exhaust pressure P₁Pressure P of pipe network₃The single machine output M of the air compressor, the position Z of an air compressor regulating valve and other signals;

s3, establishing a learning tree, taking each complete air compressor loading process as a learning period, and anticipating the fitting amount I_lnAnd T_lnEstimating a reverse initial extremum;

s4, designing a loading current iterative learning loop in the DCS system to enable the fitting quantity I in each learning period_lnAfter confirming the coefficient K through the first learning_lAfter the determination is valid, the convergence direction is preferred with the initial extreme value put into the data register stack, the preferred value is registered as an intermediate preferred value, and the learning rate lambda is accumulated plus 1, K_lIf the judgment is invalid, the learning is not carried out; first learning confirmation coefficient K_lThe formula is as follows:

I_lnthe iterative learning formula is as follows;

iterative limited momentum omega_lThe formula is as follows:

during specific implementation, the learning loop is activated when the value of the air compressor running signal R is 1, and the fitting quantity I is obtained_lnSending the data to a first DELAY operation module through a first SFT switching module, and progressively calculating the fitting quantity I within 10 seconds_lnAnd (3) the change amplitude, namely sending a current stability tendency confirmation signal delta I with the value of 1 after the change amplitude tends to be stable. Air compressor regulating valve position Z is in loading instruction D_lWhen the value is 1, the value is sent to a second SFT switching module, the change amplitude of the valve position Z within 10 seconds is calculated progressively through a second DELAY hysteresis operation module, and an adjustment stability approaching confirmation signal delta Z with the value of 1 is sent out after the change amplitude tends to be stable; exhaust pressure P₁Pressure P of pipe network₃The change amplitude within 10 seconds is progressively calculated through a third DELAY operation module and a fourth DELAY operation module, and after the change amplitude tends to be stable, pressure stability approaching confirmation signals delta P with the value of 1 are respectively sent out₁、ΔP₃；

All other air compressor running signals of the air compressor station are sent to a first NOR logic NOR module, and when all the other air compressor running signals are 0, a single machine output confirmation signal M with the value of 1 is sent out; fitting amount I_lnSending the signals to a first TQ quality judgment module, and sending an air compressor current quality signal Q with the value of 1 when the quality is NOT bad through a first NOT logic NOT module; sending the air compressor comprehensive alarm signal S into a second NOT logic NOT module, and sending out an air compressor comprehensive alarm NOT signal with the value of 1 when no alarm exists(ii) a The 7 Boolean signals and the load instruction D_lAir compressor running signal R and air compressor loading current learning termination signal I_ldJointly enter a first AND logic AND module AND send out a learning confirmation signal K with the value of 1 when the signals are all 1_l. Learning confirmation signal K_lSending the data to a first CNT counting module, accumulating the learning rate lambda and adding 1, and outputting a real-time fitting quantity I to a variable pin of a first HLLMT amplitude limiting module by a first SFT switching module_lnAnd outputs the iteratively limited momentum omega_lThe corrected stable loading current of the air compressor learns the fitting quantity, the fitting quantity enters a fourth SFT switching module, and a learning confirmation signal K₁After activation, the current is sent to a first pin of a first TRISEL selector module, an initial extreme value of the first TRISEL selector module, an initial extreme value of a second pin and a convergence lower limit value of a third pin are output as an optimal loading current value of the air compressor after passing through a selected median value, the optimal loading current value output by the first TRISEL selector module is sent back to the second pin of the first TRISEL selector module through a fifth SFT switching module and is used as a new iteration extreme value to be compared and optimized with a new fitting quantity sent to the first pin in next learning activation;

s5, setting amplitude limiting intervals of different learning periods in a DCS loading current iterative learning loop, switching according to a learning rate lambda, wherein the switching intervals are lambda less than or equal to 10 and lambda more than 10, the overall principle is that the initial learning period is iterated greatly, and amplitude limiting threshold values are reduced in the middle and later periods for fine iteration; in the step, the accumulated sum of the learning rate lambda is sent to a first HLALM high-low limit judgment module through a first CNT counting module, the interval where the learning rate is located is judged, a learning rate judgment Boolean quantity of 0 or 1 is sent out, the learning rate judgment Boolean quantity enters an iterative amplitude limiting momentum output by a switching module of a third SFT switching module for switching, and when the lambda is less than or equal to 10, omega is output_l1And when lambda is greater than 10, output omega_l2The iterative amplitude limiting momentum enters a second pin of the first ADD addition and subtraction calculation functional block, a first pin of the first ADD addition and subtraction calculation functional block receives a loading current preferred value output by the first TRISEL selector module, the first ADD addition and subtraction calculation functional block executes the loading current preferred value to subtract the iterative amplitude limiting momentum, then the loading current preferred value is sent to a low limit pin of the first HLLMT amplitude limiting module, and the real-time fit quantity I output to a variable pin of the first HLLMT amplitude limiting module by the first SFT switching module is subjected to real-time fit quantity I_lnThe output performs iterative convergence direction clipping.

S6, setting a middle optimal value register retrieval branch in a DCS loading current iterative learning loop to prevent overfitting caused by generalization errors in the learning process, and setting the learning rate lambda threshold of the register branch as 10 registers at each time according to single learning absolute time judgment of loading conditions; in the step, the optimized value of the loading current output by the first TRISEL selector module is simultaneously sent to pins to be activated of a sixth, seventh, eighth, ninth and tenth SFT switching module, meanwhile, the first CNT counting module sends the accumulated sum of the learning rate lambda to a second, third, fourth, fifth and sixth HLALM high-low limit judging module, and after respectively judging that the accumulated sum of the learning rate lambda reaches 10, 20, 30, 40 and 50, an activation signal with the value of 1 is sent to the sixth, seventh, eighth, ninth and tenth SFT switching modules, and the optimized value of the loading current at the moment is registered. The register values are controlled to enter pins to be activated of a fifth SFT switching module through a first SEL multi-input selection function block, one of the register values can be selected through a first manual operator DMA module according to needs and sent back to a second pin of a first TRISEL selector module through the fifth SFT switching module again, and an iteration extreme value is replaced.

S7, setting the learning rate lambda to reach 20 times to fit a suboptimal value for the loading current, designing an iterative learning loop for loading stability approaching time in the DCS, and loading the stability approaching time T_lThe calculation formula is { T_l＝dT(I_n→I_l)，D_l ^0→1And fitting the second-order merit value for 20 times by using the loading current as a judgment threshold value. Make the fitting quantity T in each learning period_lnAfter confirming the coefficient K through the second learning_TAfter the determination is valid, the convergence direction is preferred with the initial extreme value put into the data register stack, the preferred value is registered as an intermediate preferred value, and the learning rate lambda is accumulated plus 1, K_TIf the judgment is invalid, the learning is not carried out; second learning confirmation coefficient K_TThe formula is as follows:

T_lnthe iterative learning formula is:iteration amplitude limiting momentum xi_lIs given by the formula：In the step, the judgment Boolean quantity which is obtained by fitting AND learning the loading current of the air compressor for 20 times AND the running signal R of the air compressor enter a second AND logic AND module together, AND the Boolean quantity which is output to be 1 is sent to a setting end of a first TSUMD module to activate the loading current to drive the stable time fitting AND learning loop. At the moment, the air compressor loads current I_lnThe optimal value is learned and fitted with the air compressor loading current for 20 times and sent to a first DEV deviation comparison module, and a confirmation signal Boolean quantity dI of the real-time air compressor loading current approaching the optimal value is generated_l(ii) a Fitting amount I_lnSending the data to a sixth DELAY operation module through a seventeenth SFT switching module, and progressively calculating the fitting quantity I within 10 seconds_lnThe current stability approaching confirmation signal delta I with the value of 1 is sent out after the change amplitude tends to be stable; fitting amount I_lnSending the signals to a second TQ quality judgment module, and sending an air compressor current quality signal Q with the value of 1 when the quality is NOT bad through a third NOT logical NOT module; sending the air compressor comprehensive alarm signal S into a fourth NOT logic NOT module, and sending out an air compressor comprehensive alarm NOT signal with the value of 1 when no alarm exists(ii) a Pipe network pressure P₃The variation amplitude within 10 seconds is calculated progressively through a fifth DELAY operation module, and pressure stability approaching confirmation signals delta P with the value of 1 are respectively sent out after the variation amplitude tends to be stable₃(ii) a And load instruction D_lAND an air compressor loading driving stability time learning confirmation signal K generated after the two signals of the air compressor running signal R AND the like are judged by the third AND logic AND module_TSending the data to a fourth AND logic AND module for judgment, sending the data to a reset end of the first RS trigger module, AND sending a loading instruction D to a position end of the first RS trigger module_lMutually arranging, generating a Boolean quantity which is continuously output as 1 when a loading signal is changed from 0 to 1, acquiring a driving stability confirmation signal again at a reset end and changing the driving stability confirmation signal into a pulse output Boolean quantity of 0 again when the loading signal is close to a good value, sending the Boolean quantity into a signal end of the first TSUMD module, and calculating the time length of the Boolean quantity value as 1 after a set end of the first TSUMD module is activated;

calculating the time length with the Boolean value of 1 after the first TSUMD module set end is activated, and outputting the obtained loading current driving stability time to the real-time simulation quantity T of the second HLLMT amplitude limiting module variable pin_lnAnd outputs the iterative amplitude limiting momentum xi_lThe corrected loading and stability driving time of the air compressor is the fitting quantity of the study, the fitting quantity enters an eleventh SFT switching module, and a signal K is confirmed through the study_TAnd the initial extreme value of the second pin and the convergence lower limit value of the third pin are output as a preferred value of the loading and stabilizing time of the air compressor after being activated, and the preferred value of the loading and stabilizing time output by the second TRISEL selector module is sent back to the second pin of the second TRISEL selector module through a twelfth SFT switching module and is used as a new iteration extreme value to be compared and selected again with a new fitting quantity which is input into the first pin in the next learning activation. Learning confirmation signal K_TSent to a second CNT counting module for accumulating the learning rate λ plus 1.

S8, loading a stability approaching time iterative learning loop in a DCS, setting amplitude limiting intervals of different learning periods, switching according to a learning rate lambda, wherein the switching interval is lambda is less than or equal to 10 and lambda is greater than 10, the overall principle is that the initial learning period is iterated greatly, and amplitude limiting threshold values are reduced in the middle and later periods for fine iteration; in the step, the accumulated sum of the learning rate lambda is sent to a tenth HLALM high-low limit judgment module through a second CNT counting module, the interval where the learning rate is located is judged, 0 or 1 learning rate is sent out to judge Boolean quantity, the Boolean quantity is judged by the learning rate to enter an iterative amplitude limiting momentum output by a thirteenth SFT switching module for switching the module, and when the lambda is less than or equal to 10, xi is output_l1Output xi when lambda > 10_l2The iterative amplitude limiting momentum enters a second pin of the second ADD addition and subtraction calculation functional block, a first pin of the second ADD addition and subtraction calculation functional block receives a loading and stabilizing time optimization value output by the second TRISEL selector module, and the second ADD addition and subtraction calculation functional block sends the loading and stabilizing time optimization value minus the iterative amplitude limiting momentum to a low limit pin of the second HLLMT amplitude limiting module to carry out real-time fitting quantity T_lnThe output performs iterative convergence direction clipping.

S9, setting a middle optimal value register retrieval branch in a DCS loading stability-approaching iterative learning loop to prevent overfitting caused by generalization errors in the learning process, and setting the learning rate lambda threshold of the register branch as register for every 10 times according to single learning absolute time judgment of loading conditions; in the step, the loading and stabilizing time optimal value output by the second TRISEL selector module is simultaneously sent to pins to be activated of a fourteenth, a fifteenth and a sixteenth SFT switching module, meanwhile, the second CNT counting module sends the accumulated sum of the learning rate lambda to a seventh, an eighth and a ninth HLALM high-low limit judging module, after the accumulated sum of the learning rate lambda is respectively judged to reach 10, 20 and 30, an activation signal with the value of 1 is sent to the fourteenth, the fifteenth and the sixteenth SFT switching module, and the loading and stabilizing time optimal value at the moment is registered. The register values are controlled to enter pins to be activated of a twelfth SFT switching module through a second SEL multi-input selection function block, one of the register values can be selected through a second manual operator DMA module according to needs and sent back to a second pin of a second TRISEL selector module through the twelfth SFT switching module again, and an iteration extreme value is replaced;

s10, according to the single learning absolute time of the loading working condition, designing the learning termination threshold value of the loading current iterative learning loop of the DCS to be 50 times, and designing the learning termination threshold value of the loading stability-approaching iterative learning loop to be 30 times. Stopping learning iteration when reaching a threshold value to obtain a direct dependent variable I of a loading working condition_lAnd T_l(iv) a historical fit figure of merit; in the step, a learning effective time signal with an output of 1 is obtained through a sixth HLALM high-low limit judgment module, AND a first AND logic AND module is locked through a fifth NOT logic NOT module; and obtaining a learning effective time signal with an output of 1 by the tenth HLALM high-low limit judgment module, sending the learning effective time signal into the reset end of the first TSUMD module, and locking the first TSUMD module.

The SFT is a two-input variable switching module.

Second, unloading figure of merit fitting:

s1, analyzing alternative fitting quantity of unloading working conditions according to control requirements of air compressor equipment, using current of unloading motor of air compressor and unloading stability approaching time as input quantities, and setting I as I_unAnd T_un；

S2, analyzing the controlled condition in the unloading process of the alternative fitting quantityFitting conditions cause I_unAnd T_unThe method comprises the steps of collecting various changed learning boundary conditions in real time and/or carrying out secondary processing on the boundary conditions through a DCS (distributed control system) collection device to form a confirmation set K, wherein the confirmation set K comprises an air compressor comprehensive alarm S and an unloading instruction D_uAir compressor running signal R, air compressor current quality signal Q and separator pressure P₂Etc.;

s3, establishing a learning tree, taking each complete air compressor unloading process as a learning period, and anticipating the fitting amount I_unAnd T_unEstimating a reverse initial extremum;

s4, designing an unloading current iterative learning loop in the DCS system to enable the fitting quantity I in each learning period_unThe learning confirms the coefficient K_uAfter the determination is valid, the convergence direction is preferred with the initial extreme value put into the data register stack, the preferred value is registered as an intermediate preferred value, and the learning rate lambda is accumulated plus 1, K_uIf the judgment is invalid, the learning is not carried out; learning confirmation coefficient K_uThe formula is as follows:

I_unthe iterative learning formula is:

iterative limited momentum omega_uThe formula is as follows:

when the step is implemented specifically, the learning loop is activated when the running signal R value of the air compressor is 1, and the fitting quantity I is obtained_unSending the data to a first DELAY operation module through a first SFT switching module, and progressively calculating the fitting quantity I within 10 seconds_unThe current stability approaching confirmation signal delta I with the value of 1 is sent out after the change amplitude tends to be stable; separator pressure P₂The change amplitudes within 10 seconds are progressively calculated through a third DELAY lag operation module, and the change amplitudes are respectively sent to the DELAY lag operation module after the change amplitudes tend to be stableA pressure stability tendency confirmation signal delta P with the value of 1₂(ii) a Fitting amount I_unSending the signals to a first TQ quality judgment module, and sending an air compressor current quality signal Q with the value of 1 when the quality is NOT bad through a first NOT logic NOT module; sending the air compressor comprehensive alarm signal S into a second NOT logic NOT module, and sending out an air compressor comprehensive alarm NOT signal with the value of 1 when no alarm existsThe 4 Boolean signals and the unload instruction D_uAir compressor running signal R and air compressor unloading current learning termination signal I_udJointly enter a first AND logic AND module AND send out a learning confirmation signal K with the value of 1 when the signals are all 1_u(ii) a Learning confirmation signal K_uSending the data to a first CNT counting module, accumulating the learning rate lambda and adding 1, and outputting a real-time fitting quantity I to a variable pin of a first HLLMT amplitude limiting module by a first SFT switching module_unAnd outputs the iteratively limited momentum omega_u2The corrected stable unloading current of the air compressor learns the fitting quantity, the fitting quantity enters a fourth SFT switching module, and a learning confirmation signal K_uAfter activation, the current is sent to a first pin of a first TRISEL selector module, an initial extreme value of the first TRISEL selector module, an initial extreme value of a second pin and a convergence lower limit value of a third pin are output as an air compressor unloading current optimized value after passing through a selected median value, the unloading current optimized value output by the first TRISEL selector module is sent back to the second pin of the first TRISEL selector module through a fifth SFT switching module and is used as a new iteration extreme value to be compared and optimized with a new fitting quantity sent to the first pin in next learning activation;

s5, amplitude limiting intervals of different learning periods are set in a DCS unloading current iterative learning loop, switching is carried out according to a learning rate lambda, the switching intervals are lambda is smaller than or equal to 10 and lambda is larger than 10, the overall principle is that the initial learning period is iterated greatly, and amplitude limiting threshold values are reduced in the middle and later periods for fine iteration;

when the step is implemented specifically, the first CNT counting module sends the accumulated sum of the learning rate lambda to the first HLALM high-low limit judgment module, judges the interval where the learning rate is located and sends out the learning rate of 0 or 1 to judge the Boolean quantity, and the learning rate judges that the Boolean quantity enters the iteration limit of the third SFT switching module for switching the output of the moduleAmplitude momentum, lambda is less than or equal to 10, and omega is output_u1And when lambda is greater than 10, output omega_u2The iterative amplitude limiting momentum enters a second pin of the first ADD addition and subtraction calculation functional block, a first pin of the first ADD addition and subtraction calculation functional block receives an unloading current optimized value output by the first TRISEL selector module, the first ADD addition and subtraction calculation functional block executes the unloading current optimized value to subtract the iterative amplitude limiting momentum, then the unloading current optimized value is sent to a low limit pin of the first HLLMT amplitude limiting module, and the real-time fit quantity I output to a variable pin of the first HLLMT amplitude limiting module by the first SFT switching module is subjected to_unOutputting and executing iterative convergence direction amplitude limiting;

s6, setting a middle-optimal-value register retrieval branch in a DCS unloading current iterative learning loop to prevent overfitting caused by generalization errors in the learning process, and setting the learning rate lambda threshold of the register branch as register for every 10 times according to single learning absolute time judgment of unloading working conditions;

when the step is implemented specifically, the optimized value of the unloading current output by the first TRISEL selector module is simultaneously sent to pins to be activated of the sixth, seventh, eighth, ninth and tenth SFT switching modules, meanwhile, the first CNT counting module sends the accumulated sum of the learning rate lambda to the first, second, third, fourth, fifth and sixth HLALM high-low limit judging modules, and after respectively judging that the accumulated sum of the learning rate lambda reaches 10, 20, 30, 40 and 50, the first CNT counting module sends an activation signal with the value of 1 to the sixth, seventh, eighth, ninth and tenth SFT switching modules, and registers the optimized value of the unloading current at the moment. The register values are controlled to enter pins to be activated of a fifth SFT switching module through a first SEL multi-input selection function block, one of the register values can be selected through a DMA (direct memory access) module of the first manual operator as required and sent back to a second pin of the first TRISEL selector module through the fifth SFT switching module again, and an iteration extreme value is replaced;

s7, setting the learning rate lambda to reach 20 times to fit a suboptimal value for the unloading current, designing an unloading stability tending time iterative learning loop in the DCS, and unloading stability tending time T_uThe calculation formula is { T_u＝dT(I_n→I_u)，D_u ^0→1Adopting the 20-time fitting suboptimal value of the unloading current as a judgment threshold value; make the fitting quantity T in each learning period_unThe learning confirms the coefficient K_TAfter the determination is valid, the convergence direction is preferred with the initial extreme value put into the data register stack, the preferred value is registered as an intermediate preferred value, and the learning rate lambda is accumulated plus 1, K_TIf the judgment is invalid, the learning is not carried out; learning confirmation coefficient K_TThe formula is as follows:

T_unthe iterative learning formula is as follows;iteration amplitude limiting momentum xi_uThe formula is as follows:when the step is implemented specifically, the judgment Boolean quantity obtained by the air compressor unloading current fitting learning for 20 times AND the air compressor running signal R enter a second AND logic AND module together, AND the output Boolean quantity of 1 is sent to a setting end of a first TSUMD module to activate the unloading current driving stability time fitting learning loop. At this time, the air compressor unloads the current I_unThe optimal value is learned and fitted with the air compressor unloading current for 20 times and sent to a first DEV deviation comparison module, and a confirmation signal Boolean quantity dI of the air compressor real-time unloading current approaching the optimal value is generated_uFitting quantity I_unSending the data to a sixth DELAY operation module through a seventeenth SFT switching module, and progressively calculating the fitting quantity I within 10 seconds_unThe current stability approaching confirmation signal delta I with the value of 1 is sent out after the change amplitude tends to be stable; fitting amount I_unSending the signals to a second TQ quality judgment module, and sending an air compressor current quality signal Q with the value of 1 when the quality is NOT bad through a third NOT logical NOT module; sending the air compressor comprehensive alarm signal S into a fourth NOT logic NOT module, and sending out an air compressor comprehensive alarm NOT signal with the value of 1 when no alarm exists(ii) a Separator pressure P₂The change amplitude within 10 seconds is progressively calculated through a fifth DELAY lag operation module, and the change amplitude tends to tend toAfter stabilization, respectively sending out a pressure stability approaching confirmation signal delta P with the value of 1₂(ii) a And an unload instruction D_uAND an air compressor unloading driving stability time learning confirmation signal K generated after the two signals such as the air compressor running signal R AND the like are judged by the third AND logic AND module_TSending the unloading instruction D into the reset end of the first RS trigger module AND the position end of the first RS trigger module after the unloading instruction D is sent into the fourth AND logic AND module for judgment_uAnd mutually setting, generating a Boolean quantity which is continuously output as 1 when the unloading signal is changed from 0 to 1, acquiring the driving stability confirmation signal again at a reset end and changing the driving stability confirmation signal into a pulse output Boolean quantity of 0 again when the unloading signal is close to the optimal value, sending the Boolean quantity into the signal end of the first TSUMD module, and calculating the time length of the Boolean quantity value as 1 after the setting end of the first TSUMD module is activated. Outputting the obtained unloading current driving stability time to a real-time analog quantity T of a variable pin of a second HLLMT amplitude limiting module_unAnd outputs the iterative amplitude limiting momentum xi_uThe corrected unloading driving stability time of the air compressor is the fitting quantity of the study, the fitting quantity enters an eleventh SFT switching module, and a signal K is confirmed through the study_TAnd the optimized value of the unloading and stabilizing time output by the second TRISEL selector module is sent back to the second pin of the second TRISEL selector module through a twelfth SFT switching module and is used as a new iteration extreme value to be compared and optimized with a new fitting quantity sent to the first pin in the next learning activation again. Learning confirmation signal K_TSending the result to a second CNT counting module, and adding 1 to the accumulated learning rate lambda;

s8, setting amplitude limiting intervals of different learning periods in a DCS unloading stability tending time iterative learning loop, switching according to a learning rate lambda, wherein the switching intervals are lambda less than or equal to 10 and lambda more than 10, the overall principle is that the initial learning period is iterated greatly, and amplitude limiting threshold values are reduced in the middle and later stages for fine iteration;

in the step, the accumulated sum of the learning rate lambda is sent to a tenth HLALM high-low limit judgment module through a second CNT counting module, the interval where the learning rate is located is judged, 0 or 1 learning rate is sent out to judge the Boolean quantity, and the learning rate judges the Boolean quantity to enter a thirteenth SFT switching moduleThe iterative amplitude limiting momentum output by the module is switched, and xi is output when lambda is less than or equal to 10_u1Output xi when lambda > 10_u2The iterative amplitude limiting momentum enters a second pin of the second ADD addition and subtraction calculation functional block, a first pin of the second ADD addition and subtraction calculation functional block receives the unloading and stabilizing driving time optimal value output by the second TRISEL selector module, and the second ADD addition and subtraction calculation functional block sends the unloading and stabilizing driving time optimal value minus the iterative amplitude limiting momentum to a low limit pin of the second HLLMT amplitude limiting module to carry out real-time fitting quantity T_unOutputting and executing iterative convergence direction amplitude limiting;

s9, setting a middle optimal value register retrieval branch in a DCS unloading stability-approaching iterative learning loop to prevent overfitting caused by generalization errors in the learning process, and setting the learning rate lambda threshold of the register branch as register for every 10 times according to single learning absolute time judgment of unloading working conditions;

in the step, the unloading and stabilizing time optimal value output by the second TRISEL selector module is simultaneously sent to pins to be activated of a fourteenth, a fifteenth and a sixteenth SFT switching module, meanwhile, the second CNT counting module sends the accumulated sum of the learning rate lambda to a seventh, an eighth and a ninth HLALM high-low limit judging module, after the accumulated sum of the learning rate lambda is respectively judged to reach 10, 20 and 30, an activation signal with the value of 1 is sent to the fourteenth, the fifteenth and the sixteenth SFT switching module, and the unloading and stabilizing time optimal value at the moment is registered. The register values are controlled to enter pins to be activated of a twelfth SFT switching module through a second SEL multi-input selection function block, one of the register values can be selected through a second manual operator DMA module according to needs and sent back to a second pin of a second TRISEL selector module through the twelfth SFT switching module again, and an iteration extreme value is replaced;

s10, designing the learning termination threshold value of the unloading current iterative learning loop of the DCS to be 50 times and the learning termination threshold value of the unloading stability-approaching iterative learning loop to be 30 times according to the single learning absolute time of the unloading working condition. Stopping learning iteration when a threshold value is reached to obtain an unloading working condition direct dependent variable I_uAnd T_uThe historical fit figure of merit of.

In this step, a learning effective time signal with an output of 1 is obtained by the ninth HLALM high-low limit judgment module, and is sent to the reset terminal of the first TSUMD module, so as to lock the first TSUMD module.

Thirdly, real-time working condition early warning decision:

s701, according to learning degrees of loading and unloading working conditions of the air compressor, learning the optimum value of the loading current of the air compressor for 50 times, learning the optimum value of the loading stability tending time for 30 times, learning the optimum value of the unloading current of the air compressor for 50 times, learning the optimum value of the unloading stability tending time for 30 times, operating signals R of the air compressor and loading instructions D_lUnload instruction D_uThe equal signal is used as an early warning decision loop activation signal;

when the step is implemented specifically, the judgment Boolean quantity that the optimal value of the loading current of the air compressor is learned for 50 times AND the optimal value of the loading stability approaching time is learned for 30 times is accessed into a fifteenth AND logic AND module, AND the judgment Boolean quantity AND the operation signal R AND the loading instruction D of the air compressor are accessed together_lAND when the output value is 1, the early warning decision loading branch circuit activation signal with the output value of 1 is accessed to the second TSUMD module position end AND the fifth, sixth, seventh, eighth AND fourteenth AND logic AND module.

The judgment Boolean quantity of the air compressor with the unloading current optimal value learning reaching 50 times AND the unloading stability approaching time optimal value learning reaching 30 times is accessed into the sixteenth AND logic AND module, AND an air compressor running signal R AND an unloading instruction D which are accessed together_uWhen the output value is 1, outputting an early warning decision unloading branch activation signal with the value of 1, wherein the signal is accessed to a third TSUMD module position end AND ninth, tenth, eleventh, twelfth AND thirteenth AND logic AND modules;

s702, inputting an air compressor loading instruction D_lAir compressor loading current T_lnAir compressor discharge current I_unTime to steady state loading T_lnUnloading stability approaching time T_unPreferred value of the load current I_lUnload instruction D_uThe preferred value of the discharge current I_uLoading the optimized value T of driving stability time_lUnloading stabilization driving time optimal value T_u；

When the step is implemented, the air compressor is loaded with current I_lnPreferred value of loading current I_lInputs into a fourth DEV bias comparison functionPin 1, input pin 2. Air compressor unloading current I_unPreferred value of the discharge current I_uRespectively enter an input pin 1 and an input pin 2 of a fifth DEV deviation comparison functional block. Loading stabilization driving time optimal value T_lEntering an input pin 2 of a second DEV deviation comparison functional block, and unloading the optimized value T of the driving stability time_uEnter the input pin 2 of the third DEV bias compare block. Load instruction D_lEntering a set end of a second RS trigger functional block, enabling an output signal of a seventeenth AND logic AND functional block to enter a reset end of the second RS trigger functional block, connecting an output end of the second RS trigger functional block to a signal end of a second TSUMD functional block, AND obtaining loading stability approaching time T at the output end of the second TSUMD functional block_lnAnd enters the input pin 1 of the second DEV offset comparison function. Unload instruction D_uEntering a set end of a third RS trigger functional block, enabling an output signal of a seventeenth AND logic AND functional block to enter a reset end of the third RS trigger functional block, connecting an output end of the third RS trigger functional block to a signal end of a third TSUMD functional block, AND obtaining unloading stability approaching time T at the output end of the third TSUMD functional block_unEntering an input pin 1 of a third DEV deviation comparison functional block;

s703, adopting an air compressor comprehensive alarm S non-signal, an air compressor current quality signal Q and an air compressor current driving stability signal delta I as a loop decision confirmation condition K_A；

When the step is implemented specifically, the air compressor comprehensive alarm S enters a seventeenth AND logic AND module together with an air compressor current quality signal Q AND an air compressor current driving stability signal delta I through a sixth NOT non-logic module, AND a loop decision confirmation signal K with the value of 1 is sent out when the air compressor comprehensive alarm S AND the air compressor current quality signal Q AND the air compressor current driving stability signal delta I are both 1_ALoop decision acknowledge signal K_ARespectively entering reset ends of the second RS trigger module AND the third RS trigger module, AND eighth, fifth, fourteenth, ninth, tenth AND thirteenth AND logic AND modules;

s704, generating Delta I_l＝I_ln-I_l、ΔT_l＝T_ln-T_l、ΔI_u＝I_un-I_u、ΔT_u＝T_un-T_uCalculating and judging branches by four difference values, and establishing 10 judgment threshold values in four categories;

when the step is implemented, the second DEV deviation comparison module calculates Delta I_lThe third DEV deviation comparison module calculates Δ I_uThe fourth DEV deviation comparison module calculates Δ T_lThe fifth DEV deviation comparison module calculates Δ T_u. The output end of a fifth DEV deviation comparison module deviation out-of-limit D passes through an eighth NOT non-logic module AND enters an eleventh AND logic AND module AND a twelfth AND logic AND module, the output end of the fifth DEV deviation comparison module deviation out-of-limit D1 is accessed into the ninth AND logic AND module, the output end of a second DEV deviation comparison module deviation out-of-limit D2 is accessed into the tenth AND logic AND module, the output end of a fifth DEV deviation comparison module deviation Y is accessed into a thirteenth HLALM high-low limit judgment module, AND the low-limit output end of the thirteenth HLALM high-low limit judgment module D2 is accessed into a seventh NOT non-logic module AND a thirteenth AND logic AND module; AND the seventh NOT non-logic module is connected to the ninth AND logic AND module. The output end of the deviation out-of-limit D1 of the third DEV deviation comparison module is connected to a twelfth AND logic AND module, AND the output end of the deviation out-of-limit D2 of the third DEV deviation comparison module is connected to an eleventh AND logic AND module. The output end of a fourth DEV deviation comparison module deviation out-of-limit D passes through a ninth NOT non-logic module AND enters a sixth AND logic AND module AND a seventh AND logic AND module, the output end of the fourth DEV deviation comparison module deviation out-of-limit D1 is accessed into a fourteenth AND logic AND module, the output end of the fourth DEV deviation comparison module deviation out-of-limit D2 is accessed into a fifth AND logic AND module, the output end of the fourth DEV deviation comparison module deviation Y is accessed into a twelfth HLALM high-low limit judgment module, the twelfth HLALM high-low limit judgment module D2 low limit output end is accessed into a tenth NOT non-logic module AND an eighth AND logic AND module, AND the tenth NOT non-logic module is accessed into the fifth AND logic AND module. The output end of the deviation out-of-limit D1 of the second DEV deviation comparison module is connected to a seventh AND logic AND module, AND the output end of the deviation out-of-limit D2 of the second DEV deviation comparison module is connected to a sixth AND logic AND module.

S705, outputting 10 types of working condition analysis signals of normal loading, normal unloading, loading fitting value updating, unloading fitting value updating, abnormal increase of output, slow loading, loading blockage, reduction of unloading efficiency, slow unloading, unloading blockage and the like.

When the step is implemented, the normal loading and unloading signals are not output. Signals are respectively led out from the output ends of the fifth AND sixth AND logic AND modules to the first OR logic OR module, the updated loading fitting value signal is output, the fourteenth AND logic AND module outputs an abnormally increased output signal, the seventh AND logic AND module outputs a slow loading signal, AND the eighth AND logic AND module outputs a loading blocking signal. Signals are respectively led out from the output ends of the tenth AND eleventh AND logic AND modules to the second OR logic OR module to output the signals for updating the unloading fitting value, the ninth AND logic AND module outputs the unloading efficiency reduction signals, the twelfth AND logic AND module outputs the unloading slow signals, AND the thirteenth AND logic AND module outputs the unloading blocking signals.

Of course, the modules such as AND logic, OR logic OR module may also adopt the existing devices such as diode AND triode.

Preferably, the AND logic, OR logic OR module AND the like may be made of semiconductor OR diode materials.

In order to test the application of the method provided by the invention in practice, tests are carried out on four air compressors in an air compressor station, wherein the numbers of the air compressors are #1, #2, #3 and #4 respectively; after each air compressor is implemented for 15 days by adopting the working condition detection method provided by the invention, the air compressor station obtains 4 convergent fitting optimal values, and after the optimal values are substituted into a fault early warning decision loop, internal leakage of the unloading electromagnetic valve of the #1 air compressor and blockage of the unloading electromagnetic valve of the #3 air compressor are successively found, so that the no-load current 40A of the #1 air compressor is effectively reduced, the long-term near-empty carrying running state of the #3 air compressor is eliminated, the single-machine carrying capacity is recovered, and the operation of 1 air compressor is reduced under the normal pipe network load condition, thereby greatly saving the plant power consumption.

In addition, by adopting the method provided by the invention, the quality defect that the coil suction force of the newly purchased electromagnetic valve of the air compressor is insufficient is also found, the defect cannot be detected by the conventional means, but the problems of slow unloading and insufficient output are analyzed through the decision of unloading fitting figure of merit during the unloading test after installation, and the direct economic loss can be avoided.

Although terms such as air compressor, decision tree, iterative learning loop, etc. are used more often herein, the possibility of using other terms is not excluded. These terms are used merely to more conveniently describe and explain the nature of the present invention; they are to be construed as being without limitation to any additional limitations that may be imposed by the spirit of the present invention.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; that is, the above method can be adopted in a production service system with finite variables, and the training set, the confirmation set, the weight, the decision tree function and the like are changed through the correlation of the variables, so that the historical optimization and the control optimization of the key weight are realized by the method, and the modification or the replacement does not make the essence of the corresponding technical scheme depart from the scope of the technical scheme of each embodiment of the invention.

Claims

1. The data mining method for realizing the working condition optimization based on the DCS system edge is characterized by comprising the following steps of: the method comprises the following steps:

2. The data mining method for realizing the working condition optimization based on the DCS system edge is characterized by comprising the following steps of: the method comprises the following steps:

3. The data mining method for implementing condition optimization based on DCS system edge according to claim 2, characterized in that: in the loading process of the air compressor, the current of the air compressor motor is the air compressor motor loading current I_lnThe stability approaching time of the air compressor is the loading stability approaching time T of the air compressor_ln；

wherein:

4. The data mining method for implementing condition optimization based on DCS system edge according to claim 3, wherein:

during loading, the second learning confirmation coefficient is K_lT，K_lTThe formula of (1) is as follows:

wherein: the above-mentionedComprehensively alarming the air compressor S to be not a signal; r is an air compressor running signal; q is a current quality signal of the air compressor; the described Δ P₃For pipe network pressure P₃A stability approaching confirmation signal; the delta I is a current stability approaching confirmation signal; the dI_lOptimized value I for loading real-time current and loading current_lThe values are close;

wherein, theComprehensively alarming the air compressor S to be not a signal; r is an air compressor running signal; q is a current quality signal of the air compressor; the described Δ P₂Is the separator pressure P₂The pressure stabilizing confirmation signal; the delta I is a current stability approaching confirmation signalNumber; the dI_uOptimizing value I for unloading real-time current and loading current_uThe values are close.

5. The data mining method for implementing condition optimization based on DCS system edge according to claim 2, characterized in that: the method for analyzing the potential fault of the real-time working condition and forming the DCS decision early warning comprises the following steps:

6. The data mining system for realizing condition optimization based on DCS system edge is used for executing the data mining method for realizing condition optimization based on DCS system edge according to any one of claims 2-5, and is characterized in that: the system comprises a loading current optimization fitting module, a loading stability tending time optimization fitting module, an unloading current optimization fitting module, an unloading stability tending time optimization fitting module and an early warning module; wherein:

7. The data mining system for implementing condition optimization based on DCS system edge according to claim 6, wherein: the loading current preference fitting module AND the unloading current preference fitting module respectively comprise a first SFT switching module, a third SFT switching module, a fourth SFT switching module, a fifth SFT switching module, a sixth SFT switching module, a seventh SFT switching module, an eighth SFT switching module, a ninth SFT switching module, a tenth SFT switching module, a first DELAY hysteresis operation module, a third DELAY hysteresis operation module, a first AND logic AND module, a first CNT counting function, a first HLLMT amplitude limiting module, a first HLALM high-low limit judgment module, a second HLALM high-low limit judgment module, a third HLALM high-low limit judgment module, a fourth HLALM high-low limit judgment module, a fifth HLALM high-low limit judgment module, a sixth HLALM high-low limit judgment module, a first TRISEL selector module, a first TQ quality judgment module, a first NOT logic non-module, a second NOT logic non-module, a first addition AND subtraction calculation function block, A fifth NOT logical NOT module;

the output end of the first SEL module is connected with the input end of the first TRISEL selector module through a fifth SFT switching module;

8. The data mining system for implementing condition optimization based on DCS system edge according to claim 6, wherein: the loading stability tending time preference fitting module AND the unloading stability tending time preference fitting module respectively comprise a second AND logic AND module, a third AND logic AND module, a fourth AND logic AND module, a first TSUMD module, a first DEV deviation comparison module, a first RS trigger module, a fifth DELAY hysteresis operation module, a sixth DELAY hysteresis operation module, a second CNT counting function, a seventh HLALM high-low limit judgment module, an eighth HLALM high-low limit judgment module, a ninth HLALM high-low limit judgment module, a tenth HLALM high-low limit judgment module, an eleventh SFT switching module, a twelfth SFT switching module, a thirteenth SFT switching module, a fourteenth SFT switching module, a fifteenth SFT switching module, a sixteenth SFT switching module, a seventeen SFT switching module, an Add module, a second HLLMT clipping module, a second TRITRISEL selector module, a second SEL module, a second TQ quality judgment module, a third TQ logic non-logic module, A fourth NOT logical negation module and a second ADD addition and subtraction calculation functional block;

the second AND logic AND module is connected with a set end of the first TSUMD module; the loading current preference fitting module or the unloading current preference fitting module is connected with the first DEV deviation comparison module; the first DEV deviation comparison module is connected with a signal end of the first TSUMD module sequentially through the fourth AND logic AND module AND the first RS trigger module; the first TSUMD module is connected with the second HLLMT amplitude limiting module, the eleventh SFT switching module and the second TRISEL selector module in sequence; the output end of the second TRISEL selector module is respectively connected with the twelfth SFT switching module and the first Add module;

9. The data mining system for implementing condition optimization based on DCS system edge according to claim 6, wherein: the early warning module comprises a fifth AND logic AND module, a sixth AND logic AND module, a seventh AND logic AND module, an eighth AND logic AND module, a ninth AND logic AND module, a tenth AND logic AND module, an eleventh AND logic AND module, a twelfth AND logic AND module, a thirteenth AND logic AND module, a fourteenth AND logic AND module, a fifteenth AND logic AND module, a sixteenth AND logic AND module, a seventeenth AND logic AND module, a second TSUMD module, a third TSUMD module, a second RS trigger module, a third RS trigger module, a second DEV deviation comparison module, a third DEV deviation comparison module, a fourth DEV deviation comparison function block, a fifth DEV deviation comparison function block, an eleventh HLALM high-low limit judgment module, a twelfth HLALM high-low limit judgment module, a thirteenth HLALM high-low limit judgment module, a fourteenth HLALM high-low limit judgment module, a seventh NOT logic non-module, An eighth NOT logical negation module, a ninth NOT logical negation module and a tenth NOT logical negation module;