CN102176117B - Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system - Google Patents
Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system Download PDFInfo
- Publication number
- CN102176117B CN102176117B CN 201110022875 CN201110022875A CN102176117B CN 102176117 B CN102176117 B CN 102176117B CN 201110022875 CN201110022875 CN 201110022875 CN 201110022875 A CN201110022875 A CN 201110022875A CN 102176117 B CN102176117 B CN 102176117B
- Authority
- CN
- China
- Prior art keywords
- model
- value
- fruit
- data
- signal
- 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.)
- Expired - Fee Related
Links
- 235000013399 edible fruits Nutrition 0.000 title claims abstract description 119
- 238000005303 weighing Methods 0.000 title claims abstract description 61
- 238000003672 processing method Methods 0.000 title claims abstract description 12
- 238000005070 sampling Methods 0.000 claims abstract description 59
- 238000001914 filtration Methods 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 16
- 230000001976 improved effect Effects 0.000 claims abstract description 10
- 230000006641 stabilisation Effects 0.000 claims description 16
- 238000011105 stabilization Methods 0.000 claims description 16
- 230000003068 static effect Effects 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000000630 rising effect Effects 0.000 claims description 8
- 230000001360 synchronised effect Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 6
- 238000000926 separation method Methods 0.000 claims description 6
- 241001269238 Data Species 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 5
- 230000001960 triggered effect Effects 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 4
- 230000006872 improvement Effects 0.000 claims description 4
- 230000008093 supporting effect Effects 0.000 claims description 4
- 230000009897 systematic effect Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000014509 gene expression Effects 0.000 claims description 3
- 238000012886 linear function Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 8
- 238000004458 analytical method Methods 0.000 abstract description 2
- 230000002349 favourable effect Effects 0.000 abstract description 2
- 230000007423 decrease Effects 0.000 description 5
- 101100190621 Arabidopsis thaliana PLC4 gene Proteins 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 244000144730 Amygdalus persica Species 0.000 description 1
- 241001123248 Arma Species 0.000 description 1
- 244000276331 Citrus maxima Species 0.000 description 1
- 235000001759 Citrus maxima Nutrition 0.000 description 1
- 244000131316 Panax pseudoginseng Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 235000006040 Prunus persica var persica Nutrition 0.000 description 1
- 241000220324 Pyrus Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003306 harvesting Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 235000021017 pears Nutrition 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention discloses an intelligent processing method for a dynamic weighing signal of a fruit high-speed sorting system. By using the method, the high-precision acquisition and prediction of fruit weight can be realized on the basis of an improved AR (Autoregressive) model. The method comprises the following steps of: acquiring the grading speed in real time; sampling and storing a group of data of standard fruits; filtering and eliminating low-frequency noise in a signal sampling value; reading initial data, modeling by using the improved AR model and calculating orders and parameters of the model by an operation station system; storing the parameters and the orders of the model, which are obtained by training at different sorting speeds, in a file; predicting fruit weight information by using the trained model parameters according to the sorting speed through a control station system; and carrying out precision analysis on the predication result by the operation station system. According to the method, at the grading speed of 20 fruits/second, the weighing precision can reach within 1.5+/-0.5 percent and the optimal orders and parameters of the model can be acquired again after the grading speed is changed. The intelligent processing method has simple structure, low requirement on hardware and favorable extensibility and can be conveniently transplanted for processing high-speed weighing signals in other fields.
Description
Technical field
The present invention relates to a kind of intelligent processing method of fruit high-speed separation system dynamic weighing signal.
Background technology
According to statistics, the apple of China, pears, peach, the yield of shaddock rank first in the world.To improve the economic worth of fruit, it is necessary to be cleaned to the fruit after harvest, be classified, the postharvest treatment such as pack.Dynamic Weighing Technology is one of key technology of postharvest treatment, and the precision of dynamic weighing directly influences the accuracy of the processes such as classification, vanning.The accuracy class of dynamic weighing depends primarily on two aspects:One is the Design of Mechanical Structure of fruit cup and the Design of Mechanical Structure of weighing area;Two be the design of weighing-up wave processing system.Weighing-up wave processing system generally comprises active station system and control station system.The operation interface of active station system with user interaction, is made up of server and client side;Control station system acquisition field data, receives the operational order of active station system, calculates control instruction of the output to field apparatus, and important state of a control is transmitted into active station system.At high speeds, the delay due to system, output signal in itself is begun to decline before not up to stationary value, and how signal processing system obtains high-precision weight information from existing signal, the bottleneck as system for restricting hierarchical speed.
Existing high-speed weighing signal processing method document has:
[1]J.Calpe,E.Soria,M.Martinez,J.V.Frances,A.Rosado,L.Gomez-Chova,J.Vila High-speed weighing system based on DSP[J] IEEE.2002. 0-7803-7474-6/02:High-speed weighing systems of the 1579-1583. based on DSP, adaptive filter algorithm and arma modeling are realized using DSP, it is desirable to which fruit, at least above twice of the rise time, has higher precision by the time for piece of weighing when hierarchical speed was less than for 20 fruit/second.
[2] Lv Xinming, Wang Weiming improve a kind of new filtering method of weighing precision, it is proposed that a kind of method that weighing precision is improved with adaptive-filtering, this method precision is higher, but can not meet the requirement of high speed.
[3] Zhang Rui, Lv Wenhong, vehicle dynamic weighing system researchs of the Zhang Ruixi based on neutral net adaptive-filtering, propose a kind of processing method of vehicle dynamic load signal, this method filters noise of the weighing-up wave in each frequency range using the adaptive-filtering New variable step-size LMS based on neutral net, speed is fast, but algorithm is complicated, and precision can not meet the requirement of fruit weighing system.
It can be seen that, the processing scheme research both at home and abroad for high speed fruit weighing-up wave is less at present, improves the scheme of weighing precision mostly just for the relatively low situation of hierarchical speed.
The content of the invention
The purpose of the present invention is solves existing fruit weighing system at high speeds, and measurement accuracy is low, and there is provided a kind of intelligent processing method of fruit high-speed separation system dynamic weighing signal the problem of can not meet requirement of real-time.
The step of intelligent processing method of fruit high-speed separation system dynamic weighing signal, is as follows:
1)When fruit conveying device is static, sampling obtains the stationary value of one group of standard fruit, is then transported on device and starts rotation, according to hierarchical speed, determines the sampling period, and sampling obtains each standard fruit by one group of data during weighing sensor;
2)Filtered using first-order lag, eliminate the low-frequency noise in sampled signal, obtain pretreated signal;
3)The characteristics of active station system reads preprocessed data, analyze data, according to requirement of real-time, obtains improved AR models;
4)It is determined that improving the parameter and exponent number of AR models, obtained model parameter and exponent number will be trained to be stored in file under different hierarchical speeds;
5)Control station system utilizes the model parameter prediction fruit weight information trained;
6)To the carry out precision evaluation that predicts the outcome.
The step 1)For:Need the n standards varied in weight really, in a static condition, the value after each standard fruit multiple repairing weld is averaged, as the Static Sampling value that fruit is final, is designated as respectively:b1, b2..., bn;At high speeds, according to hierarchical speed, the sampling period is determined:If the supporting foot length of weighing sensor is L, hierarchical speed is V, time interval of the signal from beginning to ramp up to beginning to decline, sampling can only beCompleted in time span, if the sampling number of each fruit is m, sampling period;Finally by the sampling period sampling of setting, the sampled data y of each standard fruit is stored1(1), y1(2) ..., y1(m) ;y2(1), y2(2) ..., y2(m) ;…… ;yn(1), yn(2) ..., yn(m)。
The step 2)For:The low-frequency noise eliminated in sampled data is filtered using first-order lag, y is defined(K) it is kth time sampled value, x (k-1) is the filtering output value of kth -1 time, and α is filter factor, and its value is less than 1, substitutes into first-order lag Filtering Formula, obtain the corresponding filtering output x (k) of kth time sampled data;After every group of sampled data filtering process, one group of data x is obtained1(1), x1(2) ..., x1(m) ;x2(1), x2(2) ..., x2(m) ;…… ;xn(1), xn(2) ..., xn(m), as the data of model training.
The step 3)For:The output signal of weighing sensor is approximately steady random time series, and the AR models of a time series are described as:The currency x of sequence(n)The linear function of past value is represented as plus an error term e(n), use following equation(1)Represent:
Equation(1)Middle a1, a2..., apFor model coefficient;x(n-1), x(n-2)..., x(n-p)For p past value of signal, p rank AR models are constituted;It is now assumed that known p initial samples value x (1), x(2)..., x (p), signal reaches stabilization after N number of moment, prediction signal stationary value x(p+N), utilize equation(1)Recursion obtains following equation(2)For:
Equation(2)Middle coefficient A1, A2..., ApFor constant, signal stabilization value x(p+N)By initial samples value x (1), x(2)..., x (p) linear expressions;For the fruit of Different Weight, by same weighing sensor, signal stabilization value is consistent with the linear relationship of initial samples value.
The step 4)For:AR model orders and ginseng are improved using active station system-computed
Number, first active station system reading model training data x1(1), x1(2) ..., x1(m) ;x2(1), x2(2) ...,
x2(m) ;…… ;xn(1), xn(2) ..., xn, and steady stability value b (m)1, b2..., bn,
It is stored in matrix variables, is designated as:
Wherein m represents the sampling number of each fruit;N represents the standard fruit quantity predicted for model parameter, and A matrixes represent the primary data of group model training per a line, and the corresponding signal stabilization value of the row of A matrixes i-th is the i-th row of matrix B;The exponent number of first hypothesized model is p, defines another set variable, is respectively:
Wherein,Training data for determining model parameter,For training dataCorresponding stationary value,For the true timing estimation model error of model order,ForThe corresponding stationary value of data,Model parameter when for exponent number being P,For storingObtained prediction data is calculated,The standard deviation of predicted value and practical stability value, is used as model error during for storing order for i;
By equation(2)It is as follows that obtained improvement AR models are used for model parameter calculation:
Obtain one group of predicted value, then compareWithDifference, calculate standard deviation during p rank models, computing formula is:
Calculating obtains the corresponding model error of a group model exponent number, takesCorresponding exponent number i when minimum, as the exponent number of model, while calculating model parameter when exponent number is i;Finally by the exponent number and parameter of model, as the optimal coefficient under present sample speed, control station system is transmitted to, the prediction for stationary value of being sampled for fruit weight signal;In system initialization, repeat the above steps and repeatedly trained, obtain corresponding model parameter and exponent number under different hierarchical speeds, stored with file;When needing to change hierarchical speed according to field condition, the model parameter and exponent number under corresponding hierarchical speed in file are directly transmitted to control station system, or recalculate the parameter and exponent number of model by active station Systematic selection, are transmitted to control station system;If field condition changes, the value stored in the value and file that recalculate is inconsistent, may choose whether more new file.
The step 5)For:Weighing-up wave stationary value is predicted, hierarchical speed is obtained first, determines that the corresponding model parameter of the hierarchical speed and exponent number are had stored in control station system;When fruit passes through weighing sensor, the interruption of control station system is triggered by outer synchronous signal rising edge, sampling number is determined by model order;When sampling obtain one group of data after filtering after, control station system is predicted with the AR models trained, and weight information is converted into after obtaining signal stabilization value, demarcation.
The step 6)For:K fruit is taken at random, the sampling period is determined, sampling obtains K group high speed dynamic datas, stable value forecast is carried out with the model trained, obtains being converted into K pre- measured weight after K prediction data, demarcation, calculate the deviation for obtaining predicting weight and actual weight, relative error, precision of prediction.
The present invention is as a result of AR model prediction fruit weight informations are improved, in the high speed of 20 fruits per second
Effectiveness of classification can reach under classification status(1.5±0.5)Within %, it is allowed to needed to change hierarchical speed, the optimal model order of reacquisition and parameter according to scene, so as to meet the requirement of weighing precision, simple in construction, favorable expandability low to hardware requirement is portable good.
Brief description of the drawings
Fig. 1 is using the high speed fruit weighting grading system structural representation for improving AR models, wherein fruit conveying device 1, variable-frequency motor 2, PC 3, PLC4, fruit port 5, electromagnet 6, fruit cup 7, signal pre-processing module 8, load-bearing piece 9, weighing sensor 10, weighing area 11, synchronous generator 12;
Fig. 2 is the signal processing module general flow chart of the present invention;
Fig. 3 is the improvement AR model training algorithm flow charts of the present invention;
Fig. 4 is the high-speed weighing flow chart of data processing figure of the present invention;
Fig. 5 is the data obtained under the fruit/second of hierarchical speed 20, carries out model training, obtained exponent number and maximum
The corresponding table of relative error, it is known that when exponent number takes 9, maximum relative error is 0.81%, i.e. the corresponding model parameter of this exponent number is optimal;
The model parameter that Fig. 6 is when with exponent number being 9 is predicted, and 10 groups of data are predicted altogether, maximum
Relative error is less than 1%, meets required precision.
Embodiment
The step of intelligent processing method of fruit high-speed separation system dynamic weighing signal, is as follows:
1)When fruit conveying device is static, sampling obtains the stationary value of one group of standard fruit, is then transported on device and starts rotation, according to hierarchical speed, determines the sampling period, and sampling obtains each standard fruit by one group of data during weighing sensor;
2)Filtered using first-order lag, eliminate the low-frequency noise in sampled signal, obtain pretreated signal;
3)The characteristics of active station system reads preprocessed data, analyze data, according to requirement of real-time, obtains improved AR models;
4)It is determined that improving the parameter and exponent number of AR models, obtained model parameter and exponent number will be trained to be stored in file under different hierarchical speeds;
5)Control station system utilizes the model parameter prediction fruit weight information trained;
6)To the carry out precision evaluation that predicts the outcome.
The step 1)For:Need the n standards varied in weight really, in a static condition, the value after each standard fruit multiple repairing weld is averaged, as the Static Sampling value that fruit is final, is designated as respectively:b1, b2..., bn;At high speeds, according to hierarchical speed, the sampling period is determined:If the supporting foot length of weighing sensor is L, hierarchical speed is V, time interval of the signal from beginning to ramp up to beginning to decline, sampling can only beCompleted in time span, if the sampling number of each fruit is m, sampling period;Weighing sensor can be equivalent to inertial element, and output signal when fruit passes through is equivalent to the step response of inertial element, rise timeDepending on the time constant T of inertial element, meet at high speeds, so can not directly sample obtains the sampled value of steady component, can only be predicted with riser portions divided data;When standard fruit is by weighing sensor, the interruption of control station system is triggered by outer synchronous signal rising edge, after delay a period of time, is sampled by the sampling period of setting, stores the sampled data y of each standard fruit1(1), y1(2) ..., y1(m) ;y2(1), y2(2) ..., y2(m) ;…… ;yn(1), yn(2) ..., yn(m)。
The step 2)For:The low-frequency noise eliminated in sampled data is filtered using first-order lag, y is defined(K) it is kth time sampled value, x (k-1) is the filtering output value of kth -1 time, and α is filter factor, and its value is less than 1, substitutes into first-order lag Filtering Formula, obtain the corresponding filtering output x (k) of kth time sampled data;After every group of sampled data filtering process, one group of data x is obtained1(1), x1(2) ..., x1(m) ;x2(1), x2(2) ..., x2(m) ;…… ;xn(1), xn(2) ..., xn(m), as the data of model training.
The step 3)For:The output signal of weighing sensor is approximately steady random time series, the AR of a time series(Autoregression)Model is described as:The currency x of sequence(n)The linear function of past value is represented as plus an error term e(n), use following equation(1)Represent:
Equation(1)Middle a1, a2..., apFor model coefficient;x(n-1), x(n-2)..., x(n-p)For p past value of signal, p rank AR models are constituted;Due to the requirement of real-time, it is impossible to which sampling is required for carrying out complicated calculating after terminating every time, and the sampled signal to each fruit determines model order and parameter, stationary value is obtained by successive ignition respectively, can so expend for a long time;It is now assumed that known p initial samples value x (1), x(2)..., x (p), signal reaches stabilization after N number of moment, prediction signal stationary value x(p+N), utilize equation(1)Recursion obtains following equation(2)For:
……
Equation(2)Middle coefficient A1, A2..., ApFor constant, signal stabilization value x(p+N)By initial samples value x (1), x(2)..., x (p) linear expressions;It is demonstrated experimentally thatAs long as amplitude reach 50% of stationary value or so, with regard to the precision of prediction within 2% can be reached;Assuming that pin length of weighing is L, it is signal amplitude from 0 to the time used when stablizing amplitude half to define T1, now only requires that extension set speed V is less than L/T1, and directly reads the Weighing method contrast of stationary value, and system level speed can be doubled.Again by the feature for output signal of being weighed under the architectural characteristic and low speed of analysis weighing sensor, understand system be equivalent to inertial element, output signal when fruit passes through equivalent to inertial element step response, for the fruit of Different Weight, by same weighing sensor, inertial elementT values it is identical, signal stabilization value is consistent with the linear relationship of initial samples value.
The step 4)For:AR model orders and parameter, first active station system reading model training data x are improved using active station system-computed1(1), x1(2) ..., x1(m) ;x2(1), x2(2) ..., x2(m) ;…… ;xn(1), xn(2) ..., xn, and steady stability value b (m)1, b2..., bn, it is stored in matrix variables, is designated as:
Wherein m represents the sampling number of each fruit;N represents the standard fruit quantity predicted for model parameter, and A matrixes represent the primary data of group model training per a line, and the corresponding signal stabilization value of the row of A matrixes i-th is the i-th row of matrix B;The exponent number of first hypothesized model is p, defines another set variable, is respectively:
Wherein,Training data for determining model parameter,For training dataCorresponding stationary value,For the true timing estimation model error of model order,ForThe corresponding stationary value of data,Model parameter when for exponent number being P,For storingObtained prediction data is calculated,The standard deviation of predicted value and practical stability value, is used as model error during for storing order for i;
By equation(2)It is as follows that obtained improvement AR models are used for model parameter calculation:
(4)
Obtain one group of predicted value, then compareWithDifference, calculate standard deviation during p rank models, computing formula is:
Calculating obtains the corresponding model error of a group model exponent number, takesCorresponding exponent number i when minimum, as the exponent number of model, while calculating model parameter when exponent number is i;Finally by the exponent number and parameter of model, as the optimal coefficient under present sample speed, control station system is transmitted to, the prediction for stationary value of being sampled for fruit weight signal;In system initialization, repeat the above steps and repeatedly trained, obtain corresponding model parameter and exponent number under different hierarchical speeds, stored with file;When needing to change hierarchical speed according to field condition, the model parameter and exponent number under corresponding hierarchical speed in file are directly transmitted to control station system, or recalculate the parameter and exponent number of model by active station Systematic selection, are transmitted to control station system;If field condition changes, the value stored in the value and file that recalculate is inconsistent, may choose whether more new file.
The step 5)For:Weighing-up wave stationary value is predicted, hierarchical speed is obtained first, determines that the corresponding model parameter of the hierarchical speed and exponent number are had stored in control station system;When fruit passes through weighing sensor, the interruption of control station system is triggered by outer synchronous signal rising edge, starts sampling after delay a period of time, sampling number is determined by model order;When sampling obtain one group of data after filtering after, control station system is predicted with the AR models trained, and weight information is converted into after obtaining signal stabilization value, demarcation.
The step 6)For:K fruit is taken at random, the sampling period is determined, sampling obtains K group high speed dynamic datas, stable value forecast is carried out with the model trained, obtains being converted into K pre- measured weight after K prediction data, demarcation, calculate the deviation for obtaining predicting weight and actual weight, relative error, precision of prediction.
Embodiment:
As shown in figure 1, fruit weighting grading system includes fruit conveying device 1, variable-frequency motor 2, PC 3, PLC4, fruit port 5, electromagnet 6, fruit cup 7, signal pre-processing module 8, load-bearing piece 9, weighing sensor 10, weighing area 11, synchronous generator 12;Fruit conveying device 1 is connected with variable-frequency motor 2, synchronous generator 13 is arranged on the driven pulley of fruit conveying device 1, the conveyer belt of fruit conveying device 1 is provided with fruit cup 7, fruit is placed with fruit cup 7, it is weighing area 11 below one end of the conveyer belt of fruit conveying device 1, load-bearing piece 9 and weighing sensor 10 are provided with weighing area 11, electromagnet 6 is provided with below the other end of the conveyer belt of fruit conveying device 1, weighing sensor 10 and signal pre-processing module 8, PLC4, electromagnet 6 is sequentially connected, the lower section of electromagnet 6 is provided with fruit port 5, PLC4 respectively with variable-frequency motor 2, PC 3 is connected.Signal pre-processing module 8 is responsible in real time being amplified the output signal of weighing sensor 11 pretreatment such as filtering.When there is fruit cup 7 by weighing area 11, PLC processing modules 8 calculate the accurate weight information that this fruit cup is carried in real time, judge fruit grade, real-time tracking is carried out to it, control electromagnet 6 unloads the hierarchical speed that fruit acts and controls variable-frequency motor 2.PC 2 obtains accurate algorithm model according to different hierarchical speeds, and whole fruit grading process is monitored in real time.Control station system of the present invention has selected the S7-300 series of PLC of Siemens Company, and active station system uses VB monitoring softwares, realizes the intelligent processing method of high-speed weighing.
Assuming that now the speed of fruit conveying device was 20 fruit/seconds, i.e., each fruit is 50 milliseconds by the time interval of weighing sensor.When fruit passes through weighing sensor, because residence time is very short, transmitter output waveform has not arrived stationary value, and fruit just have left sensor, and output signal is begun to decline, so can not directly obtain the true samples value of fruit from the sampled value of transmitter.
As shown in Fig. 2 the main-process stream of system is:After the completion of initialization, if carrying out model training, then the data of standard fruit are obtained;If model training is completed, sampled data is the sampled data of fruit to be fractionated, is then predicted with corresponding model.The Data Preparation Process of model training is:The standard fruit 15 of Different Weight is taken, the weight demands of standard fruit cover larger scope, for example, weighed for apple, and weight is desirable from 50g to 500g.In a static condition, 15 standards are first put successively on fruit cup really, to being averaged after each standard fruit multiple repairing weld, as the steady stability value of fruit weight, be designated as respectively:B1, b2 ..., b15.Then fruit conveying device starts high-speed rotation, when synchronizing signal rising edge arrives, and produces interruption, then obtains hierarchical speed by the time interval of rising edge, calculates the sampling period:If the supporting foot length of weighing sensor is L, hierarchical speed is V, then time interval of the signal from beginning to ramp up to beginning to decline, sampling can only beCompleted in time, if the sampling number of each fruit is m, sample rate, according to the restrictive condition in sampling interval, while it is also contemplated that the limitation of PLC interrupt interval minimum times, takes a suitable sampling time.When outer synchronous signal rising edge arrives, then postpone a sampling period, PLC starts the sampling period by setting, n times of being sampled respectively to each standard fruit, sampled data is stored in PLC DB blocks with Dint types, takes the memory space of 60*m byte.
As shown in figure 3, improving the training of AR model orders and parameter.The low-frequency noise eliminated in sampled data is filtered first with first-order lag, y is defined(K) it is kth time sampled value, x (k-1) is the filtering output value of kth -1 time, and α is filter factor, and its value is less than 1, substitutes into first-order lag Filtering Formula, obtain the corresponding filtering output x (k) of kth time sampled data;After every group of sampled data filtering process, one group of data x is obtained1(1), x1(2) ..., x1(m) ;x2(1), x2(2) ..., x2(m) ;…… ;xn(1), xn(2) ..., xn(m), as the data of model training.From experiment, a value has preferable filter effect between 0.4-0.6, and filtered data, continuation is stored in PLC DB blocks with Dint types.After the completion of filtering, active station system interface " reading primary data " button is changed into visible, clicks on after button, the filtering output data and steady stability value of 15 fruits are stored in matrix variables, are designated as:
Wherein m=10, it is 10 to represent each fruit sampling number;N=15, represents the standard fruit quantity predicted for model parameter;A matrixes represent one group of sampled value, the corresponding stationary value homography B of the i-th row the i-th row per a line.The exponent number of first hypothesized model is, another set matrix variables are defined, are respectively
Wherein,Training data for determining model parameter,For the corresponding stationary value of training data,With
Model error when model order is determined,ForThe corresponding stationary value of data,Model parameter when for exponent number being P,For storingObtained prediction data is calculated,Model error during for storing order for i.By step 3)Obtain improving the formula of AR models, calculate linear coefficient, then with the coefficient pairMiddle data are predicted, order=*, obtain one group of predicted value, then compareWithDifference, calculate p rank models when standard deviation, finally obtain the corresponding standard deviation of a group model exponent number, takeExponent number i when minimum, as the exponent number of final mask, while calculating model parameter when exponent number is i, obtain the exponent number and parameter of model.Change hierarchical speed, obtain hierarchical speed from 5 fruit/seconds to 20 fruit/seconds, corresponding model order and parameter are stored in file.When hierarchical speed changes, active station Systematic selection can be passed through, if clicking on " re -training model ", then to repeat the above steps, obtain new model parameter and be transmitted to PLC, if clicking on button " write-in exponent number and parameter ", the model order and parameter value under correspondence hierarchical speed in file are directly transmitted to PLC.Under the hierarchical speed of 20 fruit per second, model criteria difference and the relation of exponent number, as shown in Figure 5, it is known that model order takes model parameter when 9 optimal.
As shown in figure 4, under high-speed case weighing-up wave prediction, hierarchical speed is obtained first, determines that the corresponding model parameter of the hierarchical speed and exponent number are had stored in control station system.When fruit passes through weighing sensor, the interruption of control station system is triggered by synchronizing signal rising edge, with step 1)The sample rate of middle determination is sampled, and sampling number is determined by model order;When sampling obtain one group of data after filtering after, control station system is predicted with the AR models trained, obtains the stationary value of signal.Demarcated again, the linear relationship between the size and fruit actual weight of signal stabilization value is, wherein,For Monomial coefficient,For constant term coefficient,WithDetermined by calibration experiment, calculate the actual weight for obtaining fruit.
As a result the method for precision evaluation is:Random to select 10 fruits, sampling obtains 10 groups of high speed dynamic datas, carries out stable value forecast with the model trained, obtain 10 prediction data.VB reads prediction data, and demarcation is output in text box after being converted into fruit weight, then corresponding fruit actual weight entered from the keyboard, calculates the deviation for obtaining predicted value and actual value, relative error, precision of prediction.During experiment, under the hierarchical speed of 20 fruit/seconds, the 10 groups of sampled datas randomly selected with obtained model prediction, obtained relative error is as shown in Figure 6.
Claims (1)
1. a kind of intelligent processing method of fruit high-speed separation system dynamic weighing signal, it is characterised in that it the step of it is as follows:
1)When fruit conveying device is static, sampling obtains the stationary value of one group of standard fruit, is then transported on device and starts rotation, according to hierarchical speed, determines the sampling period, and sampling obtains each standard fruit by one group of data during weighing sensor;
2)Filtered using first-order lag, eliminate the low-frequency noise in sampled signal, obtain pretreated signal;
3)The characteristics of active station system reads preprocessed data, analyze data, according to requirement of real-time, obtains improved AR models;
4)It is determined that improving the parameter and exponent number of AR models, obtained model parameter and exponent number will be trained to be stored in file under different hierarchical speeds;
5)Control station system utilizes the model parameter prediction fruit weight information trained;
6)To the carry out precision evaluation that predicts the outcome;
The step 1)For:Need the n standards varied in weight really, in a static condition, the value after each standard fruit multiple repairing weld is averaged, as the Static Sampling value that fruit is final, is designated as respectively:b1, b2..., bn;At high speeds, according to hierarchical speed, the sampling period is determined:If the supporting foot length of weighing sensor is L, hierarchical speed is V, time interval t of the signal from beginning to ramp up to beginning to declineAC=L/V, sampling can only be in tACCompleted in time span, if the sampling number of each fruit is m, sampling period ts< L/ (V*m);Finally by the sampling period sampling of setting, the sampled data y of each standard fruit is stored1(1), y1(2) ..., y1(m);y2(1), y2(2) ..., y2(m);……;yn(1), yn(2) ..., yn(m);The step 2)For:The low-frequency noise eliminated in sampled data is filtered using first-order lag, y is defined(K) it is kth time sampled value, x (k-1) is the filtering output value of kth -1 time, α is filter factor, its value is less than 1, first-order lag Filtering Formula x (k)=a*y (k)+(1-a) * y (k-1) are substituted into, the corresponding filtering output x (k) of kth time sampled data is obtained;After every group of sampled data filtering process, one group of data x is obtained1(1), x1(2) ..., x1(m);x2(1), x2(2) ..., x2(m);……;xn(1), xn(2) ..., xn(m), as the data of model training;
The step 3)For:The output signal of weighing sensor is approximately steady random time series, and the AR models of a time series are described as:The currency x of sequence(n)The linear function of past value is represented as plus an error term e(n), use following equation(1)Represent:
X (n)=a1x(n-1)+a2x(n-2)+...+apx(n-p)+e(n) (1)
Equation(1)Middle a1, a2..., apFor model coefficient;x(n-1), x(n-2)..., x(n-p)For p past value of signal, p rank AR models are constituted;It is now assumed that known p initial samples value x (1), x(2)..., x (p), signal reaches stabilization after N number of moment, prediction signal stationary value x(p+N), utilize equation(1)Recursion obtains following equation(2)For:
X (p+N)=a1x(N)+a2x(N+1)+...+apx(N+p-1)
=A1x(1)+A1x(2)+...+Apx(p) (2)
Equation(2)Middle coefficient A1, A2..., ApFor constant, signal stabilization value x(p+N)By initial samples value x (1), x(2)..., x (p) linear expressions;For the fruit of Different Weight, by same weighing sensor, signal stabilization value is consistent with the linear relationship of initial samples value;
The step 4)For:AR model orders and parameter, first active station system reading model training data x are improved using active station system-computed1(1), x1(2) ..., x1(m);x2(1), x2(2) ..., x2(m);……;xn(1), xn(2) ..., xn, and steady stability value b (m)1, b2..., bn, it is stored in matrix variables, is designated as:
Wherein m represents the sampling number of each fruit;N represents the standard fruit quantity predicted for model parameter, and A matrixes represent the primary data of group model training per a line, and the corresponding signal stabilization value of the row of A matrixes i-th is the i-th row of matrix B;The exponent number of first hypothesized model is p, defines another set variable, is respectively:
Wherein, Ap1Training data for determining model parameter, Bp1For training data Ap1Corresponding stationary value, Ap2For the true timing estimation model error of model order, Bp2For Ap2The corresponding stationary value of data, XpModel parameter when for exponent number being P, YpFor storing Ap2Calculate obtained prediction data, eiThe standard deviation of predicted value and practical stability value, is used as model error during for storing order for i;
By equation(2)It is as follows that obtained improvement AR models are used for model parameter calculation:
Ap1*Xp=Bp1 (3)
Calculate linear coefficient Xp, then with the coefficient to Ap2Middle data are predicted, order
Yp=Ap2*Xp (4)
Obtain one group of predicted value Yp, then compare and YpAnd Bp2Difference, calculate standard deviation during p rank models, computing formula is:
Calculating obtains the corresponding model error of a group model exponent number, takes eiCorresponding exponent number i when minimum, as the exponent number of model, while calculating model parameter X when exponent number is ii;Finally by the exponent number and parameter of model, as the optimal coefficient under present sample speed, control station system is transmitted to, the prediction for stationary value of being sampled for fruit weight signal;In system initialization, repeat the above steps and repeatedly trained, obtain corresponding model parameter and exponent number under different hierarchical speeds, stored with file;When needing to change hierarchical speed according to field condition, the model parameter and exponent number under corresponding hierarchical speed in file are directly transmitted to control station system, or recalculate the parameter and exponent number of model by active station Systematic selection, are transmitted to control station system;If field condition changes, the value stored in the value and file that recalculate is inconsistent, may choose whether more new file;
The step 5)For:Weighing-up wave stationary value is predicted, hierarchical speed is obtained first, determines that the corresponding model parameter of the hierarchical speed and exponent number are had stored in control station system;When fruit passes through weighing sensor, the interruption of control station system is triggered by outer synchronous signal rising edge, sampling number is determined by model order;When sampling obtain one group of data after filtering after, control station system is predicted with the AR models trained, and weight information is converted into after obtaining signal stabilization value, demarcation;
The step 6)For:K fruit is taken at random, the sampling period is determined, sampling obtains K group high speed dynamic datas, stable value forecast is carried out with the model trained, obtains being converted into K pre- measured weight after K prediction data, demarcation, calculate the deviation for obtaining predicting weight and actual weight, relative error, precision of prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110022875 CN102176117B (en) | 2011-01-20 | 2011-01-20 | Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110022875 CN102176117B (en) | 2011-01-20 | 2011-01-20 | Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102176117A CN102176117A (en) | 2011-09-07 |
CN102176117B true CN102176117B (en) | 2012-12-05 |
Family
ID=44519308
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201110022875 Expired - Fee Related CN102176117B (en) | 2011-01-20 | 2011-01-20 | Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102176117B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103840970B (en) * | 2014-01-24 | 2017-09-15 | 珠海多玩信息技术有限公司 | A kind of method and device for obtaining service operation state |
CN104386419B (en) * | 2014-11-17 | 2016-08-24 | 浙江大学 | Elliposoidal fruit conveying whirligig |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101085442A (en) * | 2006-06-06 | 2007-12-12 | 中国农业大学 | Method for treating and classifying orange image based on RGB composite model |
CN101898191A (en) * | 2010-07-16 | 2010-12-01 | 浙江大学 | Weighting method and fruit weighting grading system adopting DSP (Digital Signal Processor) processing module and intelligent algorithm |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104154983B (en) * | 2008-12-09 | 2017-01-11 | 数据逻辑Adc公司 | Systems and methods for reducing weighing errors associated with partially off-scale items |
-
2011
- 2011-01-20 CN CN 201110022875 patent/CN102176117B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101085442A (en) * | 2006-06-06 | 2007-12-12 | 中国农业大学 | Method for treating and classifying orange image based on RGB composite model |
CN101898191A (en) * | 2010-07-16 | 2010-12-01 | 浙江大学 | Weighting method and fruit weighting grading system adopting DSP (Digital Signal Processor) processing module and intelligent algorithm |
Non-Patent Citations (1)
Title |
---|
毛华先.水果品质智能化实时检测和分级系统研究.《农机化研究》.2010,(第10期),95-97. * |
Also Published As
Publication number | Publication date |
---|---|
CN102176117A (en) | 2011-09-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP2103915B1 (en) | Apparatus and method for determining a resonant frequency of a wind turbine tower | |
CN106054617B (en) | The on-line prediction method of network control system forward path random delay | |
CN102930155B (en) | Obtain the method and device of the early-warning parameters of electricity needs | |
CN104009734A (en) | Gradient variable-step LMS self-adaptation filtering method | |
CN110246481A (en) | A kind of automobile active denoising method of estimated engine speed | |
CN104048676B (en) | MEMS (Micro Electro Mechanical System) gyroscope random error compensating method based on improved particle filter | |
CN109460613A (en) | Model method of cutting out and device | |
CN102176117B (en) | Intelligent processing method for dynamic weighing signal of fruit high-speed sorting system | |
CN1925310B (en) | Motor speed control system and load inertia estimating method using inverse operating model | |
CN105760213A (en) | Early warning system and method of resource utilization rate of virtual machine in cloud environment | |
CN101299052A (en) | Electric generator one time frequency modulation real time quantitative analysis method based on phasor measuring technique | |
CN113156200B (en) | Power grid low-frequency oscillation real-time monitoring device | |
CN106786675A (en) | A kind of power system stabilizer, PSS and its implementation | |
CN113225047A (en) | TVLP-MF-based dynamic checkweigher rapid filtering method and system | |
Kelemençe et al. | Dynamic weighing using a time-variant low pass filter | |
CN106500811B (en) | Dynamic weighing method and device | |
Meller et al. | Adaptive filtering approach to dynamic weighing: A checkweigher case study | |
CN111561929B (en) | Time delay and noise reduction method for vehicle-mounted MEMS inertial sensor | |
Mahata et al. | Direct identification of continuous-time errors-in-variables models | |
JP2000121424A (en) | Decision device for metering condition of dynamic metering apparatus | |
CN118095106B (en) | Method, system and equipment for predicting residual fuel value of airplane | |
Çımen et al. | Classification of dynamic egg weight using artificial neural network | |
CN111091191A (en) | Data processing method and device and data deep learning system | |
He et al. | Delay Estimation of Dynamic System Based on Correlation Coefficient | |
CN107147373B (en) | Composite filtering method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20121205 Termination date: 20130120 |
|
CF01 | Termination of patent right due to non-payment of annual fee |