CN112808348B - Multi-target rice milling unit dispatching optimization system based on ACO-BP - Google Patents
Multi-target rice milling unit dispatching optimization system based on ACO-BP Download PDFInfo
- Publication number
- CN112808348B CN112808348B CN202110099893.XA CN202110099893A CN112808348B CN 112808348 B CN112808348 B CN 112808348B CN 202110099893 A CN202110099893 A CN 202110099893A CN 112808348 B CN112808348 B CN 112808348B
- Authority
- CN
- China
- Prior art keywords
- rice
- neural network
- rice mill
- unit
- parameters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
- B02B—PREPARING GRAIN FOR MILLING; REFINING GRANULAR FRUIT TO COMMERCIAL PRODUCTS BY WORKING THE SURFACE
- B02B3/00—Hulling; Husking; Decorticating; Polishing; Removing the awns; Degerming
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
- B02B—PREPARING GRAIN FOR MILLING; REFINING GRANULAR FRUIT TO COMMERCIAL PRODUCTS BY WORKING THE SURFACE
- B02B7/00—Auxiliary devices
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Abstract
The invention discloses an ACO-BP (anaerobic-anoxic-oxic-BP) based multi-target rice milling unit scheduling optimization system, wherein a 4 x 4 rice milling unit control system is designed, multi-target processing can be realized by scheduling optimization among units, processing parameters of each rice milling are optimized, the rice crushing rate of a rice mill can be reduced, and the processing efficiency of the units is improved. The unit control system optimizes the utilization algorithm inside the unit, can regulate and control the processing parameters of each rice mill according to the monomer processing target, realizes the intelligent control of the online whitening precision of the unit, and reduces the running cost of the rice mill. A rice mill unit control system is built in a mode of optimizing a BP neural network by using an Ant Colony Optimization (ACO), the ant colony algorithm can accelerate the learning rate of the neural network, converge on the optimal parameters more quickly, and a neural network model for optimally regulating and controlling the rice mill unit is built. The set up rice mill group database can carry out iterative optimization on processing parameters and processing schemes through product evaluation and learning capacity of a neural network, and self-learning of the database is learned.
Description
Technical Field
The invention relates to the technical field of rice processing, in particular to an ACO-BP (anaerobic-anoxic-oxic-BP) based multi-target rice milling unit scheduling optimization system.
Background
In recent years, the automation of production equipment in the rice processing industry is rapidly developed, wherein a rice milling equipment has a single rice milling machine control system, and the flow of a feeding port or the rotating speed of a spindle motor is controlled by monitoring the rotating speed of a roller shaft of a single rice milling machine and the temperature rise state of a whitening room and then utilizing regulation modes such as PID (proportion integration differentiation) control or fuzzy control. The rice milling machine set at the present stage has simple processing function, the control system regulates and controls the single rice milling machine with low processing efficiency, the machine set has high production energy consumption, and the polished rice quality is different after processing.
Disclosure of Invention
The invention aims to solve the problems and provide an ACO-BP-based multi-target rice mill unit scheduling optimization system.
The invention achieves the above purpose through the following technical scheme:
the invention relates to an ACO-BP (access control-back propagation) -based multi-target rice mill unit scheduling optimization system, which comprises a plurality of groups of rice mill units, a data center and an SCADA (supervisory control and data acquisition) system, wherein each group of rice mill units can read data of the data center, a real-time data cache region of each group of rice mill units is established on the data center, the SCADA system analyzes and screens the data, an OPC (OLE for process control) server exchanges information with the SCADA system, an intelligent rice mill group database is established in the data center, each technical and economic data of a rice milling process is analyzed through the SCADA system, an optimal rice mill group data operation scheme is obtained, and operation control parameters of each rice mill unit are written into the data center for calling and execution by terminal equipment. A set of rice mill group is constituteed to 4 individual intelligent rice mills, and is a set of individual rice mill in the rice mill group is as single intelligent body, and every intelligent body all can be regulated and control on line through touch-sensitive screen manual regulation and unit control system, and is a set of two adjacent rice mills utilize the analog quantity to regulate and control unit internal flow through the detection data to discharge gate and feed inlet in the rice mill group.
The rice mill is by feeder hopper, whitening chamber, main axle, discharge gate, the device that induced drafts, flow control device, wind-jet device and frame constitution, the feeder hopper with the whitening chamber is connected, the main axle that grinds is located in the whitening chamber, the discharge gate of whitening chamber with the device that induced drafts is connected, the feeder hopper with be provided with flow control device between the whitening chamber, the entrance of whitening chamber is provided with the wind-jet device, the whitening chamber sets up in the frame.
The multi-core task scheduling algorithm is adopted for the multiple groups of rice mill unit control systems to conduct regulation and control, the main list scheduling algorithm comprises two parts of task list ordering and list scheduling, the task priority is judged by utilizing a task priority computing method, the execution sequence of the tasks is determined, and then the tasks are distributed to appropriate computing units according to the sequence to complete scheduling, the rice mill unit control system mainly inputs detection quantities which are respectively detected by flow Q1 at a feed inlet, temperature T1 and pressure P1 in a milling chamber, roller shaft speed V1, air blowing pressure P2 of an air blowing device, air pressure P3 of an air suction device and flow detection Q2 of a cool rice bucket at a discharge outlet, wherein the rice milling rate and whiteness detection serve as evaluation standards; the output regulating quantity comprises a feed inlet opening D1 for regulating the feed flow, a roller shaft rotating speed regulating V1X, an air-jet air pressure regulating P2X and an air-suction air pressure regulating P3X, the length of the longest path from an inlet node to a task nk is calculated as the weight of the task nk, and the longest path comprises the sum of the weights of all task nodes and communication edges on the path;
6 nodes of a BP neural network input layer of the rice milling unit control system are mainly flow Q1, the temperature T1 of a whitening chamber, pressure P1, the rotating speed V1 of a roller shaft, air injection pressure P2 and air suction pressure P3, 4 nodes of an output layer are mainly regulated and controlled to be a feed inlet opening D1, the rotating speed of the roller is corrected to be V1X, the air injection pressure is adjusted to be P2X, and the air suction pressure is adjusted to be P3X;
when the neural network is initialized, wherein X (i) represents the input of the neural network, Y (i) represents the expected output of the neural network, the number of nodes of an input layer is set to be n, the number of nodes of a hidden layer is set to be l, the number of output layers is set to be m, and the connection weight omega between the network input layer and the hidden layer ij Connection weight ω between hidden layer and output layer jk Hidden layer threshold a j Output layer threshold b k Learning rate is η, and the excitation function is
Output of the hidden layer:
output of the output layer:
and (3) calculating an error: take the error formula as
Wherein Y is k -O k ＝e k ，
Weight value updating formula:
ω jk ＝ω jk +ηH j e k (8)
the specific optimization steps of the ACO-BP are as follows:
(1) initializing parameters, setting BP neural network parameters, training error epsilon, learning efficiency eta and maximum iteration number N max (ii) a Set of settings C i The initial pheromone value of the middle element j is tau j (C i ) Determining the ant colony algorithm time as t and the iteration number as N in the initial set t 0, initial pheromone value is τ 0 Initial concentration is the same, information volatilization coefficient rho and maximum iteration number is N max 。
(2) Updating selection parameters of ant state, wherein m ants are located at initial positions of ant holes, and ant i is randomly selected from a set C i Selecting a set of weights and thresholds in a set, and then according to the probability P i Selecting a next set by a formula and randomly selecting a group of weight values and threshold values until all elements selected by each ant are combined to form a group of initial weight values and threshold values of the neural network, wherein tau j (C i ) Is set C i The pheromone value of a certain j group of weight values and threshold value combination;
(3) after the circulation of m ants is finished, substituting the selected m groups of initial weights and threshold values into the network to calculate the BP godRecording the minimum error epsilon through the error epsilon of the actual output and the expected output of the network min And comparing ε min With the magnitude of the expected error E, if
ε min <E, if the corresponding ant parameters are the optimal solution, executing the step 5, otherwise, continuing the next step;
(4) the pheromone is updated, the distribution of the pheromone is more reasonable, the updating rule of the pheromone is improved, and the pheromone is updated according to the following formula after all ants complete one-time circulation of the whole set;
τ j (C i )(t+1)＝(1-ρ)τ j (C i )(t)+ρ△τ j (C i ) (10)
in the formula:
q is a constant, typically taken as 1;
(5) repeating the steps 2 and 3 until the minimum error meets the condition or all ants converge to the same group of paths, and the iteration number meets the maximum iteration number N max Ending the circulation;
(6) the ant parameters obtained by algorithm optimization are used as the initial weight and the threshold of the BP neural network, the iterative BP neural network connection weight and the threshold are continuously updated until the set error requirement or the maximum iteration times is reached, and the training is completed;
(7) and outputting the trained ant colony optimization BP neural network model.
The invention designs a rice milling unit comprising 16 intelligent rice mills, which is divided into 4 groups of units, each group comprises four rice mills, and a unit control system comprises three parts, mainly comprises four rice milling processing parameter optimization inside the unit, parameter scheduling optimization between the units and self-learning of a database.
The unit control system optimizes parameters in a group, and is mainly used for refining and distributing the parameters to four times of rice milling according to the requirement of processing polished rice, the processing parameters of each time of rice milling are adjusted to be suitable parameters according to distributed processing targets, and an Ant Colony Optimization (ACO) is used for training a BP neural network optimization network model to optimize the processing parameters;
the unit control system optimizes the dispatching between the group and the group, mainly realizes the multi-target processing of the rice milling unit and brings a plurality of targets of low energy consumption, low rice crushing rate, high efficiency, high quality and the like into a processing object. The machine set distributes the same polished rice requirement to four groups of rice mills, adjusts and controls different processing parameters according to four groups and evaluates polished rice after processing, optimizes the optimal processing parameters of each rice mill, optimizes the group by using a scheduling algorithm, adjusts and controls the parameters of each rice mill on line between the groups, and uploads data to a database to provide a foundation for subsequent processing scheme optimization.
The optimization of the database of the unit is mainly divided into processing scheme data optimization and processing parameter optimization of each rice milling machine, an initial value is set according to expert experience, and the varieties and precision requirements of different polished rice are continuously processed in the follow-up process, and the quality evaluation of the processed product is adjusted to the optimal processing scheme through iterative optimization.
The invention has the beneficial effects that:
the invention relates to an ACO-BP-based multi-target rice milling unit dispatching optimization system, which has the following technical effects compared with the prior art:
1) the invention designs a 4 x 4 rice milling unit control system, multi-target processing can be realized among units by scheduling optimization, processing parameters of each rice milling are optimized, the rice milling rate of the rice mill can be reduced, and the processing efficiency of the unit is improved.
2) The unit control system optimizes the utilization algorithm inside the unit, can regulate and control the processing parameters of each rice mill according to the monomer processing target, realizes the intelligent control of the online whitening precision of the unit, and reduces the running cost of the rice mill.
3) A rice mill unit control system is built in a mode of optimizing a BP neural network by using an Ant Colony Optimization (ACO), the ant colony algorithm can accelerate the learning rate of the neural network, converge on the optimal parameters more quickly, and a neural network model for optimally regulating and controlling the rice mill unit is built.
4) The set up rice mill group database can carry out iterative optimization on processing parameters and processing schemes through product evaluation and learning capacity of a neural network, and self-learning of the database is learned.
Drawings
FIG. 1 is a system control block diagram of the present invention;
FIG. 2 is a structural view of the sand roller air-jet rice mill of the present invention;
FIG. 3 is a diagram of the detection and regulation of the rice mill of the present invention;
FIG. 4 is a BP neural network topology of the present invention;
FIG. 5 is a flow chart of algorithm optimization of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings:
first, control system network structure design
1. System communication structure
An intelligent rice mill unit dispatching and controlling system is composed of multiple sets of an intelligent rice mill unit, a data center and an SCADA system. The system adopts a centralized distributed structure, the intelligent rice mill units adopt local controllers, the data center adopts a central controller, the controllers, the data center and the SCADA system adopt an industrial Ethernet communication mode, each rice mill unit can read and write data of the data center in real time, a real-time data cache area of each rice mill unit is established on the data center, the data are analyzed and screened through the SCADA system, information is exchanged between an OPC server and the SCADA system, an intelligent rice mill group database is established in the data center, various technical and economic data of a rice milling process are analyzed through the SCADA system, the optimal rice mill group data operation scheme is found, and operation control parameters of each intelligent rice mill unit are written into the data center for calling and executing by terminal equipment. Each set of rice mill unit is provided with a field control touch screen. The system can form 4 groups of intelligent rice milling machine sets at most, and each group has 4 intelligent rice mills. The control system block diagram is shown in FIG. 1;
single-machine controller input and output and storage allocation (tables 1-4)
In order to improve the reliability, stability, intelligent function of a control system and cost performance of equipment, a single unit control unit of a single intelligent rice mill adopts a Siemens Smart200 series programmable controller and an analog quantity expansion module EM AE08 thereof, all single unit configuration controllers are configured identically, and different units only need to modify corresponding IP addresses.
Input output cell assignment
TABLE 1 digital quantity input table
TABLE 2. digital quantity output table
| Output the output | Description of the function |
| 1 | Y1 lower gate closing operation |
| 2 | Y1 lower gate opening operation |
| 3 | Y2 pressing left pressing |
| 4 | Y2 pressure reduction |
| 5 | Y3 pressing left pressing |
| 6 | Y3 pressure reduction left pressure reduction |
| 7 | Upper gate of Y4: switching off the gate by 0 and switching off the gate by 1; |
| 8 | the temperature channel is gated by a signal that is, |
| 9 | the control box refrigerates the fan heater driving signal, 1 refrigerates; |
| 10 | a weight detection actuator; extended EM223 module |
| 11 | A refrigeration system drive signal; extended EM223 module |
| 12 | A heating system driving signal; extended EM223 module |
| 13 | A whiteness detection mechanism drive signal; extended EM223 module |
TABLE 3 analog output
| Input device | Description of the function |
| AIW0 | A current signal of a rice machine motor; |
| AIW2 | a temperature signal; |
| AIW4 | a weight sensor signal; extended EM231 module |
| AIW6 | The signal of the built-in negative pressure sensor; extended EM231 module |
| AIW8 | A weight sensor signal; extended EM231 module |
| AIW10 | A weight sensor signal; extended EM231 module |
Memory allocation (all control units configured identically)
TABLE 4 storage Allocation Table
Data center
In order to improve the efficiency of the intelligent rice mill, a [ data center ] CPU1215C is a small dynamic database of the system, realizes data exchange between the intelligent rice mills, and records real-time dynamic data; the rice milling machine is also an information transfer station (slave station) of an intelligent rice milling machine and an SCADA system, and provides real-time read-write small dynamic database data for the SCADA system (an OPC client master station) through an OPC server; for a user of the rice milling production line with a PLC control system simulation screen, a Smart200 is adopted to provide a rice milling machine screen display state signal for the rice milling production line PLC control system simulation screen, and control parameters and related information are exchanged with the rice milling production line PLC control system simulation screen. Its storage allocation is as follows.
Memory allocation
A dynamic database of 16 intelligent rice mills is established in a data center CPU (Central processing Unit), and is shown in the following table
Table 5 stand-alone control unit and data center storage distribution table (16 rice machine)
Data center storage allocation method
1. Firstly, establishing a data cache region of 16 stations of intelligent rice mills sharing read-write data centers;
2. writing 16 intelligent rice mill control parameters through a data center, and establishing a data cache region with the same memory number on the intelligent rice mill; it is distributed in the following way
Table 6 meter machine write data buffer work common buffer table
Table 7 data center rice writing machine data area table
SCADA system design
The SCADA system adopts configuration (WINCC or configuration king), establishes data exchange between a PC and a data center, establishes a large-scale database of an intelligent rice milling unit on a PC computer, and realizes the following functions of various real-time state monitoring, data management, data analysis and the like:
communication program
The SCADA system reads and writes the data of the slave station through an OPC server;
2. the control room downloads each PLC program through the SCADA system;
3. monitoring and upgrading programs through the Internet;
establishing a database of the system: a16-platform intelligent rice mill database is established in an SCADA system, the record is formed when the equipment is started to stop for 1 time, and the record fields of the database are shown in a table 8
TABLE 8 database field table and Address assignment table of the System
Human-computer interface function:
(1) working state dynamic interface: the working state, current, flow, whitening pressure and temperature rise … of each intelligent rice mill are displayed in real time;
(2) raw material product parameter interface: the interface inputs parameters of raw materials such as rice variety, water content, roughness, polished rice rate and finished product precision …;
(3) running parameter dynamic interface: displaying technical and economic data of each intelligent rice mill in real time in a tabulated form;
(4) control parameter setting interface: setting key operation control parameters of each rice machine such as current, flow and pressure control points (invariable parameters such as machine type, power, correction coefficient and the like are set on a touch screen of the rice machine);
(5) and (4) a fault alarm function: displaying a fault point and a fault description;
the process operation scheme subprogram:
(1) flow balance of the production line: according to the flow balance principle, the processing yield is ensured, broken rice caused by broken materials is reduced, and the continuous material feeding of each machine at the rice mill workshop section is realized: the rear elevator does not stop (VB559 is 0), the full time of the rear bin exceeds 3 times (VB2205 is more than 3), and the opening control rate of the gate of the elevator is increased by 2% (VB2240 is VB2240+ 2);
(2) controlling the crushing of the rice milling section: on the premise of ensuring the processing yield and the finished product precision, reasonably distributing the whitening pressure (the reduction rate) of each rice mill, and determining the whitening pressure control value and the temperature rise control value of each rice mill;
(3) yield maximization: the finished product precision and the rice milling and crushing increase (the first rice mill does not break) are ensured as the premise, and the flow of each rice mill is increased and decreased in a balanced manner under the condition of allowance of limit current;
setting of reference time: and taking the time of the first rice machine as a reference time (the touch screen of the first rice machine is arranged), and memorizing the reference time (the running current VD2206 is more than 2.1A) by the SCADA system at the moment of starting the first rice machine.
Second, design of unit control system
The rice mill unit control system of this design contains 16 intelligent rice mills, divide into 4 rice mill units of group, and every group is four times rice mills to two emery roll air-blast rice mills and two iron rod air-blast rice mills combination realize four machines as a set of and go out white, and the unit can realize multi-target processing according to polished rice requirement or different varieties, and can set up the processing prescription after optimizing to processing parameter conveying database. For the purpose of clearly explaining the method, a grinding roller air-jet rice mill is used as a mechanical part for detailed description, a group of rice mill units are used as a control system for description, only part of detection parameters are selected at present, and the number of a plurality of machine units or sensors is increased subsequently.
Improvement of rice mill set
The processing technology of the design adopts multi-machine light grinding combined low-temperature-rise rice milling processing. An improved structure of a sand roller air-jet rice mill is shown in figure 2.
The rice milling unit takes the requirement of finishing rice precision as an overall target, the processing requirement is distributed to four times of rice milling, proper processing parameters are matched for the processing target of each rice milling machine to be used as a monomer target, and the ratio of bran powder removed by each rice milling is preset to be 3.5:3.0:2.0:1.5 (the ratio can be adjusted according to the processing requirement and the variety of the finished rice). The unit control system takes the first rice mill as a main control object, detects processing input parameters in the unit on line, judges whether the processing target is reached or not through PLC processing, and outputs parameters to the actuator if the processing target is not reached, so that the regulation and control of corresponding parts are realized. Meanwhile, the regulation and control records, the processing parameters and the quality detection of the processed polished rice are uploaded to a database, so that a data base is provided for the subsequent scheme adjustment.
Referring to fig. 2 and fig. 3, the improved rice mill control system of the invention mainly inputs detection quantities of flow Q1 at the feed inlet (I), temperature T1 and pressure P1 in the whitening chamber (II), roller speed V1, wind pressure P2 of the wind jet device (III), wind pressure P3 of the wind suction device (III), and flow detection Q2 of the rice cooling bucket at the discharge outlet (IV), wherein the polished rice rate and whiteness detection are used as evaluation standards; the output regulating quantity comprises a feed inlet opening D1 for regulating the feed flow, a roller shaft rotating speed regulating V1X, an air jet pressure regulating P2X and an air suction pressure regulating P3X, and the specific detection and regulating parameters are shown in figure 3.
Inter-unit scheduling optimization
When the machine set processes the same target, the quality of polished rice after processing is different due to the difference of the processing parameters of each rice milling in the four subgroups at the initial processing time, the optimal processing parameters of each rice milling can be scheduled according to the quality evaluation of the polished rice, and the processing efficiency and the polished rice quality of the rice milling machine set are improved.
The unit control system adopts a multi-core task scheduling algorithm for regulation and control, the main list scheduling algorithm comprises a task list sorting part and a list scheduling part, and the task priority calculation method is used for judging the priority of tasks, determining the execution sequence of the tasks and then distributing the tasks to proper calculation units according to the sequence to complete scheduling.
TL-scheme (formula 1) is a common task priority calculation mode, is a top-down weighting mode, and calculates the length of the longest path from an entry node to a task nk as the weight of the task nk, wherein the longest path refers to the sum of the weights of all task nodes and communication edges on the path. The list sequence of the tasks is a dynamic list which is updated in real time in the scheduling process, priority calculation is carried out in real time according to quality evaluation of polished rice for arrangement, the original communication edge weight value is 0 as the tasks are distributed to the calculation unit where the father node is located, and the tl value of the follow-up task node is changed.
Design of regulation and control algorithm in unit
The factors influencing the processing efficiency of the rice milling unit are many, complex nonlinear relations exist among the factors, an artificial neural network in intelligent control can approach any nonlinear continuous function with any precision by simulating the habit of thinking of human brains, and the BP neural network is a multilayer feedforward neural network trained according to an error back propagation algorithm. The BP neural network has good fault tolerance, self-adaptability and self-learning performance, but the learning speed is low in the training process, the BP neural network is easy to fall into a local limit, and aiming at the problem, the rice mill unit control method for optimizing the BP neural network by utilizing an Ant Colony Optimization (ACO) is designed. The ant colony algorithm has the characteristics of strong global optimization, high searching speed, high precision and the like, and the convergence speed and precision can be improved by optimizing and adjusting the weight and the threshold of the BP neural network through the ant colony algorithm.
BP neural network design
The error reversal (BackPropagation) neural network is mainly composed of parts: and the input samples are transmitted forward, and the result and the error are output and transmitted backward to update the network weight. The working principle is that when sample data is transmitted forwards, an input sample is transmitted from an input layer, and is transmitted to an output layer after being processed by a hidden layer, and if the actual output of the output layer is not consistent with the expected output, a stage of updating the network weight by error back propagation is entered. The back propagation of the error is to invert the output error layer by layer to the input layer through the hidden layer in a certain form, distribute the error to each nerve unit of each layer of neuron, and carry out the learning process of the network in a continuous weight adjustment cycle until the output error is in the set range or the learning times are finished.
The topological structure of the BP neural network is shown in FIG. 4, 6 nodes of the input layer mainly comprise flow Q1, the temperature T1 of a mill white chamber, pressure P1, the rotation speed V1 of a mill roller shaft, air-jet air pressure P2 and air-suction air pressure P3, 4 nodes of the output layer mainly regulate and control the opening D1 of a feed inlet, the rotation speed of the mill roller is corrected by V1X, the air-jet air pressure is adjusted by P2X, and the air-suction air pressure is adjusted by P3X.
When the neural network is initialized, wherein X (i) represents the input of the neural network, Y (i) represents the expected output of the neural network, the number of nodes of an input layer is set to be n, the number of nodes of a hidden layer is set to be l, the number of output layers is set to be m, and the connection weight omega between the network input layer and the hidden layer ij Connection weight between hidden layer and output layer ω jk Implicit layer threshold a j Output layer threshold b k Learning rate is η and excitation function is
Output of the hidden layer:
output of the output layer:
and (3) calculating an error: take the error formula as
Wherein Y is k -O k ＝e k ，
Weight value updating formula:
ω jk ＝ω jk +ηH j e k (8)
ACO-BP design
The Ant Colony Optimization (ACO) is a heuristic random search algorithm obtained by simulating the routing mode of natural ants, and is a simulated evolution algorithm combining distributed computation, a positive feedback mechanism and greedy search. The principle is that each ant individual can secrete pheromones continuously in the foraging process to identify a path which the ant passes through, other ants can perceive the pheromones in a certain range and influence the behaviors of the ants, when the ant passes through the cave and is short, the pheromones left at the same time are increased, the ants on the path are increased to the intensity of the pheromones and then increased, and finally the influence range is increased to form an autocatalysis behavior, wherein certain pheromones on the shortest path are the highest, namely the optimal path.
The process of optimizing BP neural network by ant colony algorithm includes determining the total number of all weights and thresholds of neural network as m, each weight and threshold has n values which can be selected, and the n values are [0,1 ]]Is randomly generated to form a set C i (1<i<m). Now there are m ants, each according to the transition probability P i And selecting a path, simultaneously recording the weight and a threshold value, simultaneously comparing an output error and an expected error of a training sample to adjust pheromone and path selection on the ant travelling path, and finally converging to the same path to complete iteration through continuous updating and optimization of the pheromone tau to obtain an optimal initial solution. And finally, taking the obtained optimal initial solution as an initial weight and a threshold of the BP neural network to further train the network until the error precision reaches an expected value. The specific algorithm flow is shown in fig. 5.
The specific optimization steps are as follows:
(1) initializing parameters, setting BP neural network parameters, training error epsilon, learning efficiency eta and maximum iteration times N max (ii) a Set of settings C i The initial pheromone value of the middle element j is tau j (C i ) Determining the ant colony algorithm time as t and the iteration times as N in the initial set t 0, initial pheromone value is τ 0 Initial concentration is the same, information volatilization coefficient rho and maximum iteration number is N max 。
(2) Updating selection parameters of ant state, wherein m ants are located at initial positions of ant holes, and ant i is randomly selected from a set C i Selecting a set of weights and thresholds in a set, and then according to the probability P i Selecting a next set by a formula, and randomly selecting a group of weight values and threshold values until all elements selected by each ant are combined to form a group of initial weight values and threshold values of a neural network, wherein tau j (C i ) Is set C i The pheromone value of a certain j group of weight values and threshold value combination;
(3) after the circulation of m ants is finished, substituting the selected m groups of initial weights and threshold values into the network, calculating the error epsilon between the actual output and the expected output of the BP neural network, and recording the minimum error epsilon min And comparing ε min With the magnitude of the expected error E, if
ε min <E, if the corresponding ant parameters are the optimal solution, executing the step 5, otherwise, continuing the next step;
(4) the pheromone is updated, the distribution of the pheromone is more reasonable, the updating rule of the pheromone is improved, and the pheromone is updated according to the following formula after all ants complete one-time circulation of the whole set;
τ j (C i )(t+1)＝(1-ρ)τ j (C i )(t)+ρ△τ j (C i ) (10)
in the formula:
q is a constant, typically taken as 1;
(5) repeating the steps 2 and 3 until the minimum error meets the condition or all ants converge to the same group of paths, and the iteration number meets the maximum iteration number N max Ending the circulation;
(6) the ant parameters obtained by algorithm optimization are used as the initial weight and the threshold of the BP neural network, the iterative BP neural network connection weight and the threshold are continuously updated until the set error requirement or the maximum iteration times is reached, and the training is completed;
(7) and outputting the trained ant colony optimization BP neural network model.
The foregoing shows and describes the general principles and features of the present invention, together with the advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (1)
1. A multi-target rice mill unit dispatching optimization system based on ACO-BP is characterized in that: the system comprises a plurality of groups of rice milling units, a data center and an SCADA (supervisory control and data acquisition) system, wherein each group of rice milling units can read data of the data center, a real-time data cache region of each group of rice milling units is established on the data center, the data is analyzed and screened through the SCADA system, an OPC (object process control) server exchanges information with the SCADA system, an intelligent rice milling unit database is established in the data center, each technical and economic data of a rice milling process is analyzed through the SCADA system, an optimal rice milling unit data operation scheme is obtained, and operation control parameters of each rice milling unit are written into the data center for being called and executed by terminal equipment;
4 single intelligent rice mills form a group of rice mill units, a single rice mill in the group of rice mill units serves as a single intelligent body, each intelligent body can be manually adjusted through a touch screen and adjusted and controlled online by a unit control system, and two adjacent rice mills in the group of rice mill units adjust and control the flow inside the unit by using analog quantity through detection data of a discharge port and a feed port;
the rice mill comprises a feed hopper, a whitening chamber, a main grinding shaft, a discharge port, an air suction device, a flow regulating device, an air spraying device and a frame, wherein the feed hopper is connected with the whitening chamber;
the multi-core task scheduling algorithm is adopted for the multiple groups of rice mill unit control systems to conduct regulation and control, the main list scheduling algorithm comprises two parts of task list ordering and list scheduling, the task priority is judged by utilizing a task priority computing method, the execution sequence of the tasks is determined, and then the tasks are distributed to appropriate computing units according to the sequence to complete scheduling, the rice mill unit control system mainly inputs detection quantities which are respectively detected by flow Q1 at a feed inlet, temperature T1 and pressure P1 in a milling chamber, roller shaft speed V1, air blowing pressure P2 of an air blowing device, air pressure P3 of an air suction device and flow detection Q2 of a cool rice bucket at a discharge outlet, wherein the rice milling rate and whiteness detection serve as evaluation standards; the output regulating quantity comprises a feed inlet opening D1 for regulating the feed flow, a roller shaft rotating speed regulating V1X, an air jet air pressure regulating P2X and an air suction air pressure regulating P3X, the length of the longest path from an inlet node to a task nk is calculated as the weight of the task nk, and the longest path comprises the sum of the weights of all task nodes and communication edges on the path;
6 nodes of a BP neural network input layer of the rice milling unit control system are mainly flow Q1, the temperature T1 of a whitening chamber, pressure P1, the rotating speed V1 of a roller shaft, air injection pressure P2 and air suction pressure P3, 4 nodes of an output layer are mainly regulated and controlled to be a feed inlet opening D1, the rotating speed of the roller is corrected to be V1X, the air injection pressure is adjusted to be P2X, and the air suction pressure is adjusted to be P3X;
when the neural network is initialized, wherein X (i) represents the input of the neural network, Y (i) represents the expected output of the neural network, the number of nodes of an input layer is set to be n, the number of nodes of a hidden layer is set to be l, the number of output layers is set to be m, and the connection weight omega between the network input layer and the hidden layer ij Connection weight ω between hidden layer and output layer jk Hidden layer threshold a j Output layer threshold b k Learning rate is η and excitation function is
Output of the hidden layer:
output of the output layer:
and (3) calculating an error: take the error formula as
Wherein Y is k -O k ＝e k ，
Weight value updating formula:
ω jk ＝ω jk +ηH j e k (8)
the specific optimization steps of the ACO-BP are as follows:
(1) initializing parameters, setting BP neural network parameters, training error epsilon, learning efficiency eta and maximum iteration number N max (ii) a Set of settings C i The initial pheromone value of the middle element j is tau j (C i ) Determining the ant colony algorithm time as t and the iteration times as N in the initial set t 0, initial pheromone value τ 0 Initial concentration is the same, information volatilization coefficient rho and maximum iteration number is N max ；
(2) Updating ant state and selecting parameters, wherein m ants are positioned at the initial positions of the ant holes, and the ants i are randomly selected from the set C i Selecting a set of weights and thresholds in a set, and then according to the probability P i Selecting a next set by a formula and randomly selecting a group of weight values and threshold values until all elements selected by each ant are combined to form a group of initial weight values and threshold values of the neural network, wherein tau j (C i ) As a set C i The pheromone value of a certain j group of weight values and threshold value combination;
(3) after the circulation of m ants is finished, substituting the selected m groups of initial weights and threshold values into the network, calculating the error epsilon of actual output and expected output of the BP neural network, and recording the minimum error epsilon min And comparing ε min With the magnitude of the expected error E, if
ε min <E, if the corresponding ant parameters are the optimal solution, executing the step 5, otherwise, continuing the next step;
(4) the pheromone is updated, so that the allocation of the pheromone is more reasonable, and the updating rule of the pheromone is improved; after all ants complete one cycle of the whole set, updating pheromones according to the following formula;
τ j (C i )(t+1)＝(1-ρ)τ j (C i )(t)+ρ△τ j (C i ) (10)
in the formula:
q is a constant, typically taken as 1;
(5) repeating the steps 2 and 3 until the minimum error meets the condition or all ants converge to the same group of paths, and the iteration number meets the maximum iteration number N max Ending the circulation;
(6) the ant parameters obtained by algorithm optimization are used as the initial weight and the threshold of the BP neural network, the iterative BP neural network connection weight and the threshold are continuously updated until the set error requirement or the maximum iterative times are reached, and training is completed;
(7) and outputting the trained ant colony optimization BP neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110099893.XA CN112808348B (en) | 2021-01-25 | 2021-01-25 | Multi-target rice milling unit dispatching optimization system based on ACO-BP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110099893.XA CN112808348B (en) | 2021-01-25 | 2021-01-25 | Multi-target rice milling unit dispatching optimization system based on ACO-BP |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112808348A CN112808348A (en) | 2021-05-18 |
| CN112808348B true CN112808348B (en) | 2022-08-23 |
Family
ID=75859136
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110099893.XA Active CN112808348B (en) | 2021-01-25 | 2021-01-25 | Multi-target rice milling unit dispatching optimization system based on ACO-BP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112808348B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114160234B (en) * | 2021-11-17 | 2022-11-01 | 长沙荣业软件有限公司 | Rice milling production process control method and rice-pearl production line |
| CN114471798B (en) * | 2022-01-21 | 2023-03-10 | 长沙荣业软件有限公司 | On-line detection method for rice processing reduction rate and rice milling pressure control method |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101634438B1 (en) * | 2014-02-20 | 2016-07-11 | 한국과학기술원 | System and method of bio inspired routing for wireless mesh network |
| CN106902910B (en) * | 2017-04-28 | 2019-07-05 | 国粮武汉科学研究设计院有限公司 | A kind of control method automating husk rice unit |
| CN110162807B (en) * | 2018-02-12 | 2023-09-12 | 沈阳理工大学 | Ammunition storage reliability assessment method based on improved ant colony optimization BP neural network |
| CN108927243B (en) * | 2018-09-20 | 2020-05-05 | 长沙荣业软件有限公司 | Rice mill and rice processing and milling procedure distributed cooperative intelligent control method |
| CN109331906B (en) * | 2018-09-20 | 2020-09-08 | 长沙荣业软件有限公司 | Rice huller and intelligent control method thereof |
| CN211964284U (en) * | 2020-01-23 | 2020-11-20 | 安徽喜洋洋农业科技有限公司 | Automatic fresh rice machine with double bins |
-
2021
- 2021-01-25 CN CN202110099893.XA patent/CN112808348B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112808348A (en) | 2021-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112808348B (en) | Multi-target rice milling unit dispatching optimization system based on ACO-BP | |
| CN112232980B (en) | Regulation and control method for heat pump unit of regional energy heat supply system | |
| CN110726218B (en) | Air conditioner, control method and device thereof, storage medium and processor | |
| Li et al. | Toward intelligent multizone thermal control with multiagent deep reinforcement learning | |
| CN109507961A (en) | A kind of semiconductor production line dynamic load uniform feeding control method | |
| CN106845012A (en) | A kind of blast furnace gas system model membership function based on multiple target Density Clustering determines method | |
| CN110322007B (en) | Building electrical energy-saving monitoring method | |
| CN105051625A (en) | Method for the computerized control and/or regulation of a technical system | |
| Jamous et al. | Modifications of particle swarm optimization techniques and its application on stock market: A survey | |
| CN109708278A (en) | A kind of control method and system of the operation of compressor difference | |
| CN113589693A (en) | Cement industry decomposing furnace temperature model prediction control method based on neighborhood optimization | |
| CN114510098B (en) | Production environment regulation and control method and system | |
| CN105937823A (en) | Control method and system for ground source heat pump | |
| CN115795992A (en) | Park energy Internet online scheduling method based on virtual deduction of operation situation | |
| CN113281993B (en) | Greedy K-mean self-organizing neural network multi-robot path planning method | |
| CN114046593A (en) | Dynamic predictive machine learning type air conditioner energy-saving control method and system | |
| CN112263012B (en) | Moisture content control method of redrying machine based on formula parameter library | |
| CN112926896A (en) | Production scheduling method for cigarette cut tobacco production | |
| Moin et al. | Sliding mode control design using genetic algorithms | |
| CN110490446A (en) | A kind of modular process recombination method based on improved adaptive GA-IAGA | |
| CN115849020A (en) | Tobacco shred pneumatic conveying speed control method based on fuzzy RBF neural network | |
| CN109683558A (en) | Production process system Model for Multi-Objective Optimization control method based on particle group optimizing | |
| Aurasopon et al. | Multiple hybrid of lambda iteration and bee colony optimization method for solving economic dispatch problem | |
| CN110289643B (en) | Rejection depth differential dynamic planning real-time power generation scheduling and control algorithm | |
| Zhou et al. | Imperialist competitive algorithm based on VNSOBL optimization for distributed parallel machine scheduling problem |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221228 Address after: 432600 No.2, Fuhe Avenue, Anlu City, Xiaogan City, Hubei Province Patentee after: HUBEI YONGXIANG FOOD PROCESSING MACHINE Co.,Ltd. Address before: 430023 No. 68, Xuefu Road, Changqing Garden, Dongxihu District, Wuhan, Hubei Patentee before: WUHAN POLYTECHNIC University |