CN105403245A - Sunlight greenhouse wireless sensor multi-data fusion method - Google Patents

Abstract

The invention discloses a sunlight greenhouse wireless sensor multi-data fusion method. Greenhouse data are acquired in real time through a wireless sensor; a Grubbs judging criterion is used for data pretreatment; an adaptive weighted average fusion algorithm is then used, a weighted factor is derived for fusion calculation on the data, real-time north sunlight greenhouse data fusion is realized, and the real-time data precision is improved. An experimental result shows that the Grubbs judging criterion can effectively eliminate gross errors, and the data precision is improved by 8%; and the adaptive weighted average data fusion can obviously improve the data precision, the data precision is improved by 6% after fusion, the fusion result can describe the real-time greenhouse environment accurately, and more accurate basic data can be provided for control on the greenhouse environment.

Description

Heliogreenhouse wireless senser multi-data fusion method

Technical field

The present invention relates to sensing data process field, be specifically related to a kind of heliogreenhouse wireless senser multi-data fusion method.

Background technology

The northern area of China is very long for winter, the time that can meet crop growth needs is very short, but utilize heliogreenhouse can create good environment for crop growth, solve long-standing problem north of china in winter vegetable supply problem, for the cheap crops of peasant's High-efficient Production bring facility.Along with the development of Greenhouse in North Polyhouse technology, greenhouse monitor and forecast has been become inevitable, in greenhouse, arrange that various kinds of sensors implementing monitoring is to the precision and the management level that improve environmental Kuznets Curves, the acquisition high-quality greenhouse product also final greenhouse economic benefit that improves plays a part very important.But due to the natural characteristic of greenhouse environment parameter, its main environment factor (temperature, humidity, illumination etc.) distribution is uneven, by the impact of many factors, therefore must carry out data fusion to the greenhouse information of multipoint acquisition.Data fusion technique be utilize computer technology the observation information from multiple sensor or multi-source to be carried out analyzing, overall treatment, thus draw decision-making and estimate the information process of required by task.Should in this way to the problem such as ambiguity, shortage of data, Comprehensive Evaluation solved in greenhouse multisensor sampled data, to the various useful information of greenhouse acquisition promptly and accurately, in good time complete comprehensive description is carried out to greenhouse situation, adjustment is implemented to greenhouse and controls to be all very important.

The problem of data fusion having a lot of researchist how to improve data precision for heliogreenhouse is in recent years studied, as the method merged data based on Huffman tree thought, needing to carry out two-pass scan to data divides two states to store data again, and its processing procedure is very complicated; Use the function for support of gray system theory and carry out in conjunction with third index flatness the fusion results that greenhouse data fusion can be calculated degree of precision, but taking resource space larger existence when this method requires system flexibly and processes data optimizes space problem; The method of loci in batches counting average is adopted to carry out fusion treatment to greenhouse data, system is only needed to provide the measurement data of sensor collection just can carry out data fusion calculating, but the method has very high requirement to sensor group, if improper meeting of dividing into groups causes final fusion results to occur deviation.

Summary of the invention

For solving the problem, the invention provides a kind of heliogreenhouse wireless senser multi-data fusion method, by wireless senser Real-time Collection greenhouse data, Grubbs decision criteria is utilized to carry out data prediction, recycling self-adaptive weighted average blending algorithm, derivation weighting factor carries out fusion calculation to data, achieves Greenhouse in North real time data and merges, improve real time data precision.

For achieving the above object, the technical scheme that the present invention takes is:

Heliogreenhouse wireless senser multi-data fusion method, comprises the steps:

S1, gather greenhouse indoor temperature, the soil moisture, indoor relative humidity, soil moisture, greenhouse indoor illumination intensity, greenhouse CO by multiple wireless environment Temperature Humidity Sensor, wireless optical sensor, wireless soil temperature-moisture sensor and wireless environment carbon dioxide sensor ₂concentration, dewpoint temperature and outdoor temperature humidity parameter;

S2, employing Grubbs decision criteria carry out effective gross error in real time to Monitoring Data and reject:

S21, establish through measure obtain variable x _j1, x _j2..., x _jn, calculate its mean value x by following formula _jand standard deviation sigma _j:

$x_j=\frac{1}{n}\×\underset{i=1}{\overset{n}{\Σ}}{x}_{ji},i=1,2,...,n---\left(1\right)$

${\mathrm{\σ}}_j=\sqrt{\frac{1}{n}\×\underset{i=1}{\overset{n}{\Σ}}{(x_{ji}-\stackrel{\‾}{x_j})}^2}---\left(2\right)$

S22, according to above-mentioned formulae discovery Grubbs statistic T _j, then have

$T_j=\frac{\left|\right|x_{ii}-\stackrel{\‾}{x_j}\left|\right|}{{\mathrm{\σ}}_j}---\left(3\right);$

If meet T _j(n, a), then thinks that these data contain the exceptional value of gross error value, should give up the measured value that T is corresponding >=T; Wherein T (n, a) be Grubbs decision criteria critical value, n is pendulous frequency, and a is the level of signifiance, and a gets 0.01 or 0.05 (referring to table 1); After rejecting, repeat (1)-(3) formula, until all data all meet Grubbs decision criteria there is not new abnormal data;

(n a) shows table 1 Grubbs test method critical value T

S3, employing self-adaptive weighted average algorithm carry out fusion calculation to step S2 the data obtained:

Be provided with n sensor to measure same envirment factor, the variance of n sensor is respectively sensing data is respectively x ₁, x ₂..., x _n; Weights are respectively w ₁, w ₂..., w _n; X value then after adaptive weighted fusion and weights meet

$\stackrel{\‾}{x}=\underset{p=1}{\overset{n}{\Σ}}{w}_p\×x_p,$ Wherein $\underset{p=1}{\overset{n}{\Σ}}{w}_p=1---\left(4\right)$

In formula: for n sensor is at the arithmetic mean of synchronization collection envirment factor, w _pfor the weighted value of a certain sensor, x _pfor the data that this moment sensor gathers

Wherein, square error σ ²for

${\mathrm{\σ}}^2=E\[x-\stackrel{\‾}{x}\]=E\[{\mathrm{\Σ}}_{p=1}^n{w}_p^2\×{(x-x_p)}^2+2{\mathrm{\Σw}}_p\×w_q(x-x_p)\×(x-x_q)\]---\left(5\right)$

In formula: w _qwith w _pbe respectively the weighted value of synchronization different sensors, x _qwith x _pfor the data that the different sensors of synchronization gathers

Each sensor image data is the unbiased esti-mator of x, therefore thinks that sensing data is independent of one another, so have

E[(x-x _p)×(x-x _q)]＝0，(p≠q；p＝1，2，…，n；q＝1，2，…，n)

Then

${\mathrm{\σ}}^2=E\[\underset{p=1}{\overset{n}{\Σ}}{w}_p^2\×{(x-x_p)}^2\]=\underset{p=1}{\overset{n}{\Σ}}{w}_p^2\×{\mathrm{\σ}}_p^2---\left(6\right)$

In formula: for a certain sensor square error

By above formula known when square error is minimum weighted value w _pfor

$w_p=1/{\mathrm{\σ}}_p^2\×\underset{p=1}{\overset{n}{\mathrm{\Σ}}}\frac{1}{{\mathrm{\σ}}_p^2}(p=\mathrm{1,2},...n)---\left(7\right)$

Now corresponding least mean-square error for

${\mathrm{\σ}}_{min}^2=1/{\Σ}_{p=1}^n\frac{1}{{\mathrm{\σ}}_p^2}(p=1,2,...,n)---\left(8\right).$

The present invention has following beneficial effect:

By wireless senser Real-time Collection greenhouse data, utilize Grubbs decision criteria to carry out data prediction, recycling self-adaptive weighted average blending algorithm, derivation weighting factor carries out fusion calculation to data, achieve Greenhouse in North real time data to merge, improve real time data precision.

Accompanying drawing explanation

Fig. 1 is adaptive weighted algorithm model in the embodiment of the present invention.

Fig. 2 (a) is No. 1 sensor collecting temperature data in the embodiment of the present invention.

Fig. 2 (b) is No. 2 sensor collecting temperature data in the embodiment of the present invention.

Fig. 2 (c) is No. 3 sensor collecting temperature data in the embodiment of the present invention.

Fig. 2 (d) is No. 4 sensor collecting temperature data in the embodiment of the present invention.

Fig. 2 (e) is No. 5 sensor collecting temperature data in the embodiment of the present invention

Fig. 2 (f) is No. 6 sensor collecting temperature data in the embodiment of the present invention.

Fig. 3 is sensor image data process comparison diagram in the embodiment of the present invention.

The result that Fig. 4 (a) merges for the temperature data and its average algorithm utilizing adaptive weighted algorithm fusion in the embodiment of the present invention is at the contrast sectional drawing of 5:00-6:00.

Fig. 4 (b) contrasts sectional drawing at 12:00-13:00 for the result utilizing the temperature data of adaptive weighted algorithm fusion and its average algorithm in the embodiment of the present invention and merge.

Fig. 4 (c) contrasts sectional drawing at 15:00-16:00 for the result utilizing the temperature data of adaptive weighted algorithm fusion and its average algorithm in the embodiment of the present invention and merge.

Fig. 4 (d) contrasts sectional drawing at 19:00-20:00 for the result utilizing the temperature data of adaptive weighted algorithm fusion and its average algorithm in the embodiment of the present invention and merge

Embodiment

In order to make objects and advantages of the present invention clearly understand, below in conjunction with embodiment, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

In this specific embodiment, the model of self-adaptive weighted average algorithm is as x in Fig. 1, figure ₁, x ₂..., x _nfor multisensor is in the data acquisition amount of synchronization to same envirment factor; w ₁, w ₂..., w _nfor each sensor weights of this moment; X is worth after adaptive weighted fusion.

Embodiment

The heliogreenhouse choosing Agricultural University Of Shenyang's Beishan Mountain Experimental Base is research object, this greenhouse ridge height 2.8m, and span is 8m, long 50m, indoor without heating and forced draft equipment, belongs to typical Greenhouse in North.

Image data selected equipment comprises: wireless environment Temperature Humidity Sensor, wireless optical sensor, wireless soil temperature-moisture sensor, wireless environment carbon dioxide sensor.Wireless senser experimental facilities precision is as shown in table 2.Data acquisition time is 9 to 11 November in 2014, and frequency acquisition is 1 beat/min.The greenhouse environment parameter gathered is respectively: greenhouse indoor temperature, the soil moisture, indoor relative humidity, soil moisture, greenhouse indoor illumination intensity, greenhouse CO ₂concentration, dewpoint temperature and outdoor temperature humidity.This research is based on northern day-light greenhouse, according to the principle that the day-light greenhouse overwhelming majority time is in insulation, the less and indoor relative humidity of ventilating is mainly derived from soil evaporation and photosynthesis of plant, choose greenhouse indoor temperature, indoor humidity, indoor illumination intensity, soil moisture, the soil moisture as main enter factor.

Table 2 wireless sensor devices accuracy table

This research is described for the temperature signal of 6 sensor collections in winter, experience shows that typical Greenhouse in North Interior Temperature Environment changes 1 degree Celsius, the time interval is about 20 minutes, often the data of group sampling employing sampling interval collection in 10 minutes can meet production requirement, and in one day 24 hours, collection is counted is 144 groups.As shown in Figure 2, transverse axis is the time to the sensor collecting temperature signal on November 9, and unit is hour, and the longitudinal axis is temperature, and unit is degree Celsius.

Carry out Grubbs decision criteria data prediction for temperature sensor image data, get wherein one group of real time data and compare, as shown in table 3.

Table 3 sensor collecting temperature data prediction table

Grubbs statistic as seen from the above table, according to T (n, a) Grubbs critical value n=6, a=0.05, tabling look-up, (n, a)=1.822 < 2.215 then can judge that No. 6 sensor image data are rejected as exceptional value to 1 known T.5 groups of detected values are remained after rejecting, continue the temperature data judging that sensor gathers, at this moment the Grubbs statistic of No. 4 sensor collecting temperature data is up to 1.21, and Grubbs critical value T (n now, a)=1.672 > 1.21, then can judge No. 4 sensor image data not containing gross error, do not rejected.Before and after data prediction, precision improves about 2%-9%, as shown in Figure 3, gets and select 6 groups of samplings in figure, and often organizing data is 10 minutes sample sets, and bold portion is raw data relative deviation, and dotted portion is the relative deviation of data after data prediction.

Research and application data are known, greenhouse indoor temperature when 11-12 about reach the highest, average maximum is 31.51 DEG C; Indoor temperature arrives minimum when 16-17, and average minimum temperature is 15.96 DEG C.

The present embodiment gathers 6 groups of temperature signals for sensor and carries out fusion results analysis. and after adaptive weighted algorithm fusion, each sensor weights distribution assignment is as shown in table 4.

Table 4 temperature sensor best initial weights apportioning cost

Day-light greenhouse plantation cucumber be main, rule of thumb the optimum temperature range of cucumber growth is: daytime 25-32 degree, night 14-18 degree, optimum humidity scope is 60%-85%, and optimum illumination strength range is 400-600w/ square metre.In day-light greenhouse, settle 6 cc2531 nodes respectively, each node gathers the data such as humiture, intensity of illumination respectively.Wherein the temperature measuring data in a certain moment and the corresponding optimal weighting value of each sensing data are in table 3.Its mean value is calculated to this moment measurement data and self-adaptive weighted average fusion value can obtain: self-adaptive weighted average fusion results is 18.4813, arithmetic mean fusion results is 17.0933, and known self-adaptive weighted average algorithm fusion result is more accurate than average algorithm fusion results.The result that the temperature data of adaptive weighted algorithm fusion and its average algorithm merge is utilized to carry out such as Fig. 4.In Fig. 4, (a) figure is depicted as 5:00-6:00 self-adaptive weighted average blending algorithm and average blending algorithm Comparative result; B () figure is 12:00-13:00 self-adaptive weighted average blending algorithm and average blending algorithm Comparative result; C () figure is 15:00-16:00 self-adaptive weighted average blending algorithm and average blending algorithm Comparative result; D () figure is 19:00-20:00 self-adaptive weighted average blending algorithm and average blending algorithm Comparative result.

This tests the average blending algorithm relative deviation of image data of each sensor and adaptive weight fusion estimated algorithm relative deviation comparative result in table 5.

Table 5 temperature sensor merges relative deviation table

The present embodiment carries out data prediction for Greenhouse in North real-time data collection application Grubbs decision criteria, then utilizes self-adaptive weighted average algorithm to carry out data fusion.Data precision can be obtained from the comparative analysis of the adaptive weighted fusion results of preprocessed data and raw data tradition average algorithm fusion results and be enhanced about 5%-15%.In sum, self-adaptive weighted average blending algorithm is adopted herein for Greenhouse in North Real-time Monitoring Data, effectively can improve data accuracy, meet the data fusion requirement of heliogreenhouse, simultaneously for the relation analyzed between parameters provides basic data support accurately, stability and the reliability of control system can be strengthened.

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (1)

1. sight greenhouse wireless sensor device multi-data fusion method, is characterized in that, comprise the steps:

S1, gather greenhouse indoor temperature, the soil moisture, indoor relative humidity, soil moisture, greenhouse indoor illumination intensity, greenhouse CO by multiple wireless environment Temperature Humidity Sensor, wireless optical sensor, wireless soil temperature-moisture sensor and wireless environment carbon dioxide sensor ₂concentration, dewpoint temperature and outdoor temperature humidity parameter;

S2, employing Grubbs decision criteria carry out effective gross error in real time to Monitoring Data and reject:

S21, establish through measure obtain variable x _j1, x _j2..., x _jn, calculate its mean value x by following formula _jand standard deviation sigma _j:

$x_j=\frac{1}{n}\&times;\underset{i=1}{\overset{n}{\&Sigma;}}{x}_{ji},i=1.2,...,n---\left(1\right)$

${\mathrm{\&sigma;}}_j=\sqrt{\frac{1}{n}\&times;\underset{i=1}{\overset{n}{\&Sigma;}}{(x_{ji}-\stackrel{\&OverBar;}{x_j})}^2}---\left(2\right)$

S22, according to above-mentioned formulae discovery Grubbs statistic T _j, then have

$T_j=\frac{\left|\right|x_{ji}-\stackrel{\&OverBar;}{x_j}\left|\right|}{{\mathrm{\&sigma;}}_j}---\left(3\right);$

If meet T _j(n, a), then thinks that these data contain the exceptional value of gross error value, should give up the measured value that T is corresponding >=T; Wherein T (n, a) be Grubbs decision criteria critical value, n is pendulous frequency, and a is the level of signifiance, and a gets 0.01 or 0.05; After rejecting, repeat (1)-(3) formula, until all data all meet Grubbs decision criteria there is not new abnormal data;

S3, employing self-adaptive weighted average algorithm carry out fusion calculation to step S2 the data obtained:

Be provided with n sensor to measure same envirment factor, the variance of n sensor is respectively sensing data is respectively x ₁, x ₂..., x _n; Weights are respectively w ₁, w ₂..., w _n; X value then after adaptive weighted fusion and weights meet

$\stackrel{\&OverBar;}{x}=\underset{p=1}{\overset{n}{\&Sigma;}}{w}_p\&times;x_p,$ Wherein $\underset{p=1}{\overset{n}{\&Sigma;}}{w}_p=1---\left(4\right)$

In formula: for n sensor is at the arithmetic mean of synchronization collection envirment factor, w _pfor the weighted value of a certain sensor, x _pfor the data that this moment sensor gathers

Wherein, square error σ ²for

${\mathrm{\&sigma;}}^2=E\&lsqb;x-\stackrel{\&OverBar;}{x}\&rsqb;=E\&lsqb;{\mathrm{\&Sigma;}}_{p=1}^n{w}_p^2\&times;{(x-x_p)}^2+2{\mathrm{\&Sigma;w}}_p\&times;w_q(x-x_p)\&times;(x-x_q)\&rsqb;---\left(5\right)$

In formula: w _qwith w _pbe respectively the weighted value of synchronization different sensors, x _qwith x _pfor the data that the different sensors of synchronization gathers

Each sensor image data is the unbiased esti-mator of x, therefore thinks that sensing data is independent of one another, so have

E[(x-x _p)×(x-x _q)]＝0，(p≠q；p＝1，2，…，n；q＝1，2，…，n)

Then

${\mathrm{\&sigma;}}^2=E\&lsqb;\underset{p=1}{\overset{n}{\&Sigma;}}{w}_p^2\&times;{(x-x_p)}^2\&rsqb;=\underset{p=1}{\overset{n}{\&Sigma;}}{w}_p^2\&times;{\mathrm{\&sigma;}}_p^2---\left(6\right)$

In formula: for a certain sensor square error

By above formula known when square error is minimum weighted value w _pfor

$w_p=1/{\mathrm{\&sigma;}}_p^2\&times;\underset{p=1}{\overset{n}{\&Sigma;}}\frac{1}{{\mathrm{\&sigma;}}_p^2}(p=1,2,...,n)---\left(7\right)$

Now corresponding least mean-square error for

${\mathrm{\&sigma;}}_{min}^2=1/{\mathrm{\&Sigma;}}_{p=1}^n\frac{1}{{\mathrm{\&sigma;}}_p^2}(p=1,2,...,n)---\left(8\right).$