CN117609898A - Information fusion water quality grade judging method based on sequential multi-classification three-branch decision - Google Patents

Abstract

An information fusion water quality grade judging method based on three decisions of sequential multi-classification comprises the following steps: 1) Measuring a plurality of water quality parameter values of a water sample through a sensing network of a plurality of sensor nodes to obtain a water quality measurement information system and a decision table; 2) Dividing IS into a multi-layer granularity structure according to the number of water quality parameters; 3) Classifying the samples according to a one-vs-rest multi-classification algorithm; 4) Dividing the object with decision conflict in each layer of multiple classification results into conflict domains; 5) Performing multi-layer classification based on a plurality of water quality parameters to obtain a final water quality grade judgment result; 6) The decision quality of the model is evaluated with a decision accuracy CCA and a decision speed UCD. The invention can reduce the influence of the conditions of sensor measurement errors, data missing, data abnormality and the like on water quality judgment, and has the advantages of high judgment accuracy and high robustness.

Description

Information fusion water quality grade judging method based on sequential multi-classification three-branch decision

Technical Field

The invention relates to a water quality evaluation method, in particular to a method for constructing three decision models with sequential multi-classification and combining measurement data of various sensors to judge water quality grade.

Background

Water quality monitoring is an important link in water pollution control. The traditional monitoring method mainly comprises manual sample collection and laboratory instrument analysis, and has the problems of low monitoring frequency, scattered data, large sampling error, untimely analysis of water pollution conditions and the like, and is difficult to meet the requirements of related departments on water environment management. The method utilizes modern technological means such as computers, network communication and information processing to establish a perfect water quality detection monitoring and information management system, and is one of the technical problems to be solved in the current urgent need.

At present, automatic water quality detection is an effective means for continuously acquiring water quality data. The most common approach is to collect water quality data by deploying multiple sensors, i.e. to build a wireless sensor network (WSN, wireless sensors network). However, in actual work, the water quality judgment is carried out according to the data measured by the sensor network, namely, firstly, each sensor of the sensor network is responsible for measuring the water quality of a certain area, and because the data measured by the sensors are numerous, how to process the numerous data measured by the sensors is a difficult point in the water quality judgment process. Secondly, the data measured by the sensor has a certain error. Certain water quality parameters, such as dissolved oxygen, can change along with temperature, illumination and other factors at any time, so that measured data are disturbed within a certain range. This can cause uncertainty in the data measured by the sensor network each time. Thirdly, the sensor may face the problems of aging, faults, influence of abnormal changes of surrounding environment, interference of data transmission and the like, and the phenomena of missing, conflict, redundancy and the like of monitoring data are caused, so that the difficulty and the calculated amount of accurately and reliably analyzing and predicting the water quality trend and rule are increased.

For uncertainty and imperfection existing in water quality data measured by a sensor network, the invention expresses the water quality data with uncertainty measured by the sensor by interval number, and calculates a loss function by adopting an aggregation method based on a reasonable granularity principle. On the basis, the invention constructs three decision models of sequential multi-classification, and judges the water quality grade by fusing the measured values of a plurality of water quality parameters.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a water quality grade judging method based on three decisions of sequential multi-classification. The method can reduce the influence of the conditions of sensor measurement errors, data missing, data abnormality and the like on water quality judgment, and has the advantages of high judgment accuracy and high robustness.

In order to solve the technical problems, the invention provides the following technical scheme:

an information fusion water quality grade judging method based on three decisions of sequential multi-classification, the method comprises the following steps:

1) Measuring a plurality of water quality parameter values of a water sample through a sensing network of a plurality of sensor nodes to obtain a water quality measurement information system IS= (U, AT, V, f), and a decision table D= { D ¹ ,D ² ,…,D ^s Consists of water quality levels;

2) Performing multi-layer granularity division on IS according to water quality parameters to obtain an ith layer information structure GS _i ；

3) Performing multi-classification on the sample by adopting a multi-classification algorithm one-vs-res;

4) Dividing the object with decision conflict in each layer of multiple classification results into conflict domains CON (D _i )；

5) Performing multi-layer classification based on a plurality of water quality parameters to obtain a final water quality grade judgment result;

6) The decision quality of the model is evaluated with a decision accuracy CCA and a decision speed UCD.

Further, in the step 1), the water quality measurement value with uncertainty is represented by the number of intervals.

Still further, in the step 2), the multi-layer granularity structure IS divided based on water quality parameters, i.e., is= { GS ₁ ,GS ₂ ,…,GS _m }。

In the step 3), the multi-classification algorithm one-vs-rest is to divide a plurality of classes into two classes: belonging to this class and to other classes, i.eAnd judging each classification, and integrating the results of each classification.

In said step 3), the grain structure GS for each layer _i ＝(U _i ,AT _i ,V _i ,f _i ) The classification process for each two categories is as follows:

31 Any two sample objects x, y E U are calculated in a certain water quality parameter a _j The similarity is Sim _(ai) (x,y)：

32 Given a threshold L, 0.ltoreq.L.ltoreq.1, with the object x _i The combination of the objects with the similarity greater than or equal to L is the similarity class of the objects

33 Calculating conditional probabilities of similarity classes for respective objects

34 Calculating aggregate interval loss functions for similarity classes of respective objectsAnd converts the interval loss function into an exact value +.>

35 Calculating the threshold of three decisionsAnd->

Classification is performed according to three decision rules:

(P) ifThen->

(B) If it isThen->

(N) ifThen->

In said step 4), object x _i May belong to multiple positive domains, resulting in a decision conflict, a conflicting domain being defined as a set of objects partitioned into multiple positive domains, the conflict domain is marked out, and the result of multi-classification of each layer is positive domain ++> Boundary field->Negative domain NEG (D) _i )＝U-POS(D _i )-BND(D _i )。

In the step 5), BND (D) in the classification result of each layer is obtained _i ) And NEG (D) _i ) Forms a new object set for the next level of classification, i.e. U _i+1 ＝BND(D _i )∪NEG(D _i )。

In the step 6), the indexes for evaluating the performance of the model are decision accuracy CCA and decision speed UCD, wherein CCA represents an average value of the determined correct acceptance rate of the decision class, and UCD represents an uncertain decision rate;

in 34), the method for calculating the aggregation interval loss function of the similar class comprises the following steps:

341 Calculating the mean of the left and right end points of the function, respectivelyAnd->

342 Calculation of (c)And->Is>And->

Wherein ε is a positive parameter;

343 Maximum composite multiplicative index)As the lower boundary of the interval, the maximum composite multiplicative index is +.>As the upper bound of the interval, finally, the similarity class +.>Interval loss function->

344 Interval loss functionConversion to an exact real number +.>

Where θ represents the attitudes of the decision maker to the risk, i.e., the utility.

The beneficial effects of the invention are as follows: 1. the invention expresses the data measured by the sensor by interval number, and adopts an aggregation method based on reasonable granularity principle to obtain a reasonable interval loss function for subsequent decision; 2. the invention constructs a sequential multi-classification three-branch decision model for judging the water quality grade, and the model combines a multi-classification algorithm and a conflict area division method, integrates various water quality parameter measurement values and realizes accurate judgment of the water quality grade.

Drawings

FIG. 1 is a water quality level judging process of the present invention;

FIG. 2 is a three-branch decision process based on the principle of reasonable granularity;

FIG. 3 is a sequential multi-classification three-branch decision process based on one-vs-rest.

Detailed Description

The following examples are presented to provide those skilled in the art with a further detailed description of the embodiments of the invention and to enable them to practice in the art with reference to the specification.

Referring to fig. 1 to 3, a water quality level judgment model construction method based on three decisions of sequential multi-classification includes the following steps:

1) Measuring a plurality of water quality parameter values of a water sample through a sensing network of a plurality of sensor nodes to obtain a water quality measurement information system IS= (U, AT, V, f), and a decision table D= { D ¹ ,D ² ,…,D ^s The water quality grade consists of water quality grades, and the water quality measured value with uncertainty is represented by interval numbers;

definition of a water quality measurement information system is= (U, AT, V, f), referring to table 1, u= { x ₁ ,x ₂ ,…,x _j ,…,x _n The sample object set is represented by n, which represents the number of samples, at= { a ₁ ,a ₂ ,…,a _i ,…,a _m The water quality parameter set is represented by m, wherein the m represents the number of the water quality parameters, and V= U _a∈AT V _a Is a measured value of water quality parameters, the water quality data measured by the sensors has uncertainty, and the number of intervals is usedRepresenting the mapping fUxAT→V, f (x, a) ∈V _a ，D＝{D ¹ ,D ² ,…,D ^t ,…,D ^s The water quality level is composed of s decision classes, namely s water quality levels.

2) Performing multi-layer granularity division on IS according to water quality parameters to obtain an ith layer information structure GS _i The multi-layer granularity structure IS divided based on water quality parameters, namely IS= { GS ₁ ,GS ₂ ,…,GS _m }；

Dividing IS into a multi-granularity structure is= { GS ₁ ,GS ₂ ,…,GS _m The total grain number m is equal to the number of water quality parameters, and the ith grain structure GS _i ＝(U _i ,AT _i ,V _i ,f _i ) I=1, 2, …, m, see table 2, decision class

3) The method comprises the steps of performing multi-classification on samples by adopting a multi-classification algorithm one-vs-rest, wherein the multi-classification algorithm one-vs-rest is used for classifying a plurality of classes into two classes: belonging to this class and to other classes, i.eJudging each classification, and integrating the results of each classification;

performing multi-classification three-branch decision, converting s classification into s two classifications, and regarding decision class D of the ith layer _i D is to _i Represented asWherein->Behavior set a= { a _P ,a _B ,a _N Comprises three actions, a _P Representing accepted actions, classifying x.epsilon.U as positive region, i.e., determining +.>a _B Representing actions not promised, i.e. deciding that x.epsilon.U is classified as boundary domaina _N Representing the action of rejection, classifying x.epsilon.U as negative region, i.e. deciding +.>Definitions->Is a matrix->

The ith layer grain structure is classified according to the measured value of the ith water quality parameter, taking the ith second classification of the ith layer grain structure as an example, the specific classification process is as shown in fig. 2, and the ith layer information system GS is given _i ＝(U _i ,AT _i ,V _i ,f _i ) And D _i The classification process for each two categories is as follows:

31 For any two sample objects x, y E U in a certain water quality parameter a) _j The similarity is Sim _(ai) (x, y), the similarity between x and y is Sim (x, y), the formula is as follows:

32 To U) _i Each object x of (a) _i Is of the similar class of (1)Given a threshold L, L is more than or equal to 0 and less than or equal to 1, and all meet Sim _(ai) (x _i Y) composition x of y of L _i Is->

33 Computing object x _i Is of the similar class of (1)Conditional probability of->

34 Computing similarity classesInterval loss function->x _i Interval loss function-> (. Cndot.=p, B, N), similar class ++>The mean value of the left and right end points of the interval loss function of all objects in (a) is +.>And->The formula is as follows:

definition of the definitionAnd->Respectively->And->Is a complex multiplication index of (a):

where ε is a positive parameter.

First, all are calculatedAnd->Is a complex multiplication exponent of (c). Then, the composite multiplicative index is maximized +.>As the lower boundary of the interval, the maximum composite multiplicative index is +.>As the upper bound of the interval.

Finally, obtaining the similar classInterval loss function->

Will break the intervalLoss functionConversion to an exact real number +.>

Where θ represents the attitudes of the decision maker to the risk, i.e., the utility.

The following relationship exists:

f _θ (λ _PP )≤f _θ (λ _BP )≤f _θ (λ _NP )；f _θ (λ _NN )≤f _θ (λ _BN )≤f _θ (λ _PN )；

35 Three decision rules based on the above model are as follows:

(P) ifThen->

(B) If it isThen->

(N) ifThen->

Wherein, three decision thresholdsAnd->The following are provided:

4) Dividing the object with decision conflict in each layer of multiple classification results into conflict domains CON (D _i )；

In the above classification method, the object x _i May belong to a plurality of positive areas, i.e. may belong to a plurality of decision classes, resulting in a decision conflict, for which a conflict area CON (D is defined _i ) For objects containing conflicting decisions, the result of the classification at layer i is therefore as follows: conflict domain Positive domainBoundary field-> Negative domain NEG (D) _i )＝U-POS(D _i )-BND(D _i )。

5) Performing multi-layer classification based on a plurality of water quality parameters to obtain a final water quality grade judgment result; BND (D) in the classification result of each layer _i ) And NEG (D) _i ) Forms a new object set for the next level of classification, i.e. U _i+1 ＝BND(D _i )∪NEG(D _i )；

BND (D) obtained from the ith layer _i ) And NEG (D) _i ) Constructing a new object set U _i+1 As input to the i+1-th layer classification, i.e. U _i+1 ＝BND(D _i )∪NEG(D _i )，Repeating 3) and 4) to carry out multi-layer classification to obtain a final decision result, as shown in figure 3.

6) And evaluating the performance of the model. Decision accuracy CCA and decision speed UCD are used to evaluate the decision quality of the model.

Wherein |POS (D) ^t )∩D ^t |/|POS(D ^t ) I represents the determined correct acceptance rate for each decision class, then CCA represents the average of the determined correct acceptance rates for s decision classes, and UCD represents the uncertain decision rate, including boundary, negative and collision domains.

The water quality grade judging method for three decisions of sequential multi-classification in the embodiment comprises the following steps:

step 1: the four water quality parameters are selected for Dissolving Oxygen (DO) and ammonia Nitrogen (NH) ₃ -N), total Phosphorus (TP), total Nitrogen (TN). With four sensor nodes (S ₁ 、S ₂ 、S ₃ 、S ₄ ) The sensor network is constructed to measure the above parameters of the sample. In the experiment, n samples are used, and the measured water quality information system is= (U, AT, V and f) IS shown in table 1. Let the precision of the sensor be p _DO ＝0.5,p _NH3-N ＝0.05,p _TP ＝0.005,p _TN =0.5. The water quality is divided into 5 grades (I, II, III, IV, V) according to the national surface water environment quality standard (GB 3838-2002), and the characteristic range of the water quality parameters of each grade is shown in Table 2.

TABLE 1

TABLE 2

Step 2: dividing IS into a four-layer multi-granularity structure is= { GS ₁ ,GS ₂ ,…,GS ₄ The ith grain structure GS _i ＝(U _i ,AT _i ,V _i ,f _i )，i＝1,2,3,4。

Step 3: multiple classification three-branch decision is carried out by adopting a one-vs-rest algorithm, five classifications are converted into 5 two classifications, for example, for the first two classifications, the decision classes are classified intoI.e. belonging to class i and to the other four classes.

Step 4: classifying the first layer grain structure according to the measured value of the parameter DO, firstly performing first classification and second classification, and judging whether the sample belongs to the first level or not: calculating the similarity between every two samples, and taking the samples with the similarity greater than the threshold L as the similarity class of the samplesCalculating conditional probability of similarity class of the n samples +.>And aggregate interval loss function->And converts the interval loss function into an exact value +.>Representing that two decision thresholds are calculated +.>And->Then dividing the sample grade according to three decision rules, and thenAnd continuing the second classification, the third classification, the fourth classification and the fifth classification, and respectively judging whether the sample belongs to the II th stage, the III th stage, the IV th stage and the V th stage.

Step 5: in the first layer of grain structure, namely, the result of 5 times of secondary classification according to the measured value of the water quality parameter DO is fried, part of the object x _i May be divided into POS domains multiple times, decision conflicts occur, and therefore, these samples are divided into conflict domains CON (D ₁ ). The five results are then integrated to obtain a first layer of classification results POS (D ₁ )，BND(D ₁ )，NEG(D ₁ )。

Step 6: constructing a new object set U according to the classification result of the first layer ₂ ＝BND(D _i )∪NEG(D _i ) NH based on these samples ₃ -the N measurements are subjected to a second layer classification, followed by a third and fourth layer classification according to the TP and TN measurements as described above, the fourth layer classification result being the final sample water quality class classification result POS (D), BND (D), NEG (D).

The embodiments described in the present specification are merely examples of implementation forms of the inventive concept, and the scope of protection of the present invention should not be construed as being limited to the specific forms set forth in the embodiments, and the scope of protection of the present invention and equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims (9)

1. The method for judging the water quality level of the information fusion based on three decisions of sequential multi-classification is characterized by comprising the following steps:

1) Measuring a plurality of water quality parameter values of a water sample through a sensing network of a plurality of sensor nodes to obtain a water quality measurement information system IS= (U, AT, V, f), and a decision table D= { D ¹ ,D ² ,…,D ^s Consists of water quality levels;

2) Performing multi-layer granularity division on IS according to water quality parameters to obtain an ith layer information structure GS _i ；

3) Performing multi-classification on the sample by adopting a multi-classification algorithm one-vs-res;

4) Pairs with decision conflict in multi-time classification results of each layerImage division into conflict domains CON (D _i )；

5) Performing multi-layer classification based on a plurality of water quality parameters to obtain a final water quality grade judgment result;

6) The decision quality of the model is evaluated with a decision accuracy CCA and a decision speed UCD.

2. The method for determining the level of the water quality based on the information fusion of three decisions of sequential multi-classification according to claim 1, wherein in the step 1), the number of intervals is used to represent the water quality measurement value with uncertainty.

3. The method for determining the water quality level of information fusion based on three decisions of sequential multi-classification as claimed in claim 1 or 2, wherein in said step 2), the multi-layer granularity structure IS based on the water quality parameter division, i.e., is= { GS ₁ ,GS ₂ ,…,GS _m }。

4. The method for determining the level of the information fusion water quality based on three decisions of sequential multi-classification as claimed in claim 1 or 2, wherein in the step 3), the multi-classification algorithm one-vs-rest is to divide a plurality of classes into two classes: belonging to this class and to other classes, i.eAnd judging each classification, and integrating the results of each classification.

5. The method for determining the grade of the information fusion water quality based on three decisions of sequential multi-classification as claimed in claim 1 or 2, wherein in said step 3), the grain structure GS for each layer _i ＝(U _i ,AT _i ,V _i ,f _i ) The classification process for each two categories is as follows:

31 Any two sample objects x, y E U are calculated in a certain water quality parameter a _j The similarity is Sim _(ai) (x,y)：

32 Given a threshold L, 0.ltoreq.L.ltoreq.1, with the object x _i The combination of the objects with the similarity greater than or equal to L is the similarity class of the objects

33 Calculating conditional probabilities of similarity classes for respective objects

34 Calculating aggregate interval loss functions for similarity classes of respective objectsAnd converts the interval loss function into an exact value +.>

35 Calculating the threshold of three decisionsAnd->

Classification is performed according to three decision rules:

(P) ifThen->

(B) If it isThen->

(N) ifThen->

6. The method for determining the level of information fusion water quality based on three decisions of sequential multi-classification as claimed in claim 1 or 2, wherein in said step 4), the object x is _i May belong to multiple positive domains, resulting in a decision conflict, a conflicting domain being defined as a set of objects partitioned into multiple positive domains, the conflict domain is marked out, and the result of multi-classification of each layer is positive domain ++>Boundary fieldNegative domain NEG (D) _i )＝U-POS(D _i )-BND(D _i )。

7. The method for determining the level of water quality based on three decisions of sequential multi-classification based on the information fusion of claim 1 or 2, wherein in the step 5), the BND (D _i ) And NEG (D) _i ) Forms a new object set for the next level of classification, i.e. U _i+1 ＝BND(D _i )∪NEG(D _i )。

8. The method for determining the level of the information fusion water quality based on three decisions of sequential multi-classification according to claim 1 or 2, wherein in the step 6), the indexes for evaluating the performance of the model are decision accuracy CCA and decision speed UCD, wherein CCA represents an average value of the determined correct acceptance rate of the decision class, and UCD represents an uncertain decision rate;

9. the method for determining the water quality level of information fusion based on three decisions of sequential multi-classification according to claim 1 or 2, wherein in the step 34), the method for calculating the aggregation interval loss function of the similar classes comprises the following steps:

341 Calculating the mean of the left and right end points of the function, respectivelyAnd->

342 Calculation of (c)And->Is>And->

Wherein ε is a positive parameter;

343 Maximum composite multiplicative index)As the lower boundary of the interval, the maximum composite multiplicative index is +.>As the upper bound of the interval, finally, the similarity class +.>Interval loss function->

344 Interval loss functionConversion to an exact real number +.>

Where θ represents the attitudes of the decision maker to the risk, i.e., the utility.