CN105184479A - Urban resident water-consumption behavior classification method based on intelligent water meter - Google Patents

Abstract

The invention discloses an urban resident water-consumption behavior classification method based on an intelligent water meter. The method, based on urban resident domestic water consumption detail data collected by the high-precision intelligent water meter, carries out clustering analysis on resident daily water-consumption behaviors by adopting a fuzzy clustering analysis method and with the day being time scale, and carries out classification on family structures and job natures and the like according to water consumption characteristics by utilizing an empirical value cut matrix method, and thus a better effect is achieved. The method can carry out analysis and excavation on the urban resident domestic water consumption detail data, and carry out fuzzy clustering without prior samples on the water consumption behaviors, provides decision support for management of residential quarters and planning, supplying and research and application of urban resident domestic water consumption, and lays a foundation for analyzing the urban resident domestic water consumption behaviors in different temporal and spatial scales and simulating water consumption amount.

Description

Based on city dweller's use water behaviour classification method of intellectual water meter

Technical field

The present invention relates to town water technical field, be specifically related to a kind of city dweller's use water behaviour classification method based on intellectual water meter.

Background technology

Water is the necessity that people live, and city dweller's is closely related by the factor such as water behavior and its family structure, job specification, habits and customs.City dweller's management of the use of water department needs to carry out with water planning and supply, and prerequisite grasps the water habits of all types of user, thus carry out fine-grained management.But, the collection of traditional resident living water data is mainly obtained by the mode of manual metering, general monthly the efficiency of data acquisition and accuracy rate are all on the low side, and do not have suitable method to use water behavioural characteristic according to a large amount of water number according to what determine all types of user.

Summary of the invention

For the deficiencies in the prior art, the object of the present invention is to provide a kind of city dweller's use water behaviour classification method based on intellectual water meter, to provide Data support for water supply department.

To achieve these goals, the technical scheme that the present invention takes is:

Based on city dweller's civil water behaviour classification method of intellectual water meter, comprise step:

What gather user more than two days in region to be sorted by intellectual water meter uses water number certificate;

Carry out pre-service to by water number certificate, pretreated process comprises: the data rejecting full small incidental expenses family; By being averaging according to addition with water number of same user's every day same period; The average consumption calculating same each period of user accounts for the ratio of average daily water consumption, take user as row, each period be row by water ratio, build user civil water eigenmatrix;

According to the user's civil water eigenmatrix obtained after pre-service, to the classification carrying out based on fuzzy clustering with water behavior of user, process is as follows:

Calculate the correlation matrix of user's civil water eigenmatrix;

By correlation matrix on the occasion of change, form fuzzy matrix;

Fuzzy matrix is transformed into fuzzy equivalent matrix;

Calculate the intercept battle array of fuzzy equivalent matrix, using the value of the empirical value of λ during optimal classification effect as intercept battle array, obtain the classification results of user's use water behavior.

Compared with prior art, beneficial effect of the present invention is:

Based on the water for city's residential use detail data that the present invention gathers by high precision intelligent water meter, take day as time scale, research, based on the water for city's residential use feature unsupervised classification algorithm by water detail data, is classified according to the habits and customs, family structure, job specification etc. of water feature to community resident.By the classification that becomes more meticulous of user, for infrastructure management company and water undertaking provide decision-making foundation, for water supply administrative authority planning of science activities, supply water as required, energy-saving and cost-reducing and fine-grained management lays the foundation.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the city dweller's water behaviour classification method that the present invention is based on intellectual water meter;

Fig. 2 is that day part daily uses water scale map;

1st class user water characteristic pattern when Fig. 3 is λ=0.9;

2nd class user water characteristic pattern when Fig. 4 is λ=0.9;

Other 7 classes user water characteristic pattern when Fig. 5 is λ=0.9;

Fig. 6 the 3rd and the 7th user use water characteristic pattern;

1st class user water characteristic pattern when Fig. 7 is λ=0.8;

3rd class user water characteristic pattern when Fig. 8 is λ=0.8;

4th class user water characteristic pattern when Fig. 9 is λ=0.8;

5th class user water characteristic pattern when Figure 10 is λ=0.8.

Embodiment

Along with the development of technology of Internet of things, intellectual water meter is progressively applied, and makes us can obtain resident living water data more more accurate than the past.What traditional meter reading method obtained is generally 0.1m with water data precision ³, collection period is generally 30-60 days.And the data precision gathered by intellectual water meter is up to 0.001m ³, frequency acquisition is 15 minutes/time.High-precision water number certificate like this, for studying the water habits of resident, providing the support of basic data by the relation between water feature and the behavior of research water and family structure, job specification, habits and customs, thus the planning of the management of residential quarter and city dweller's water, supply and research are applied etc. bring deep impact.Therefore, the present invention is using intellectual water meter as Data Source, and below in conjunction with embodiment, the present invention is further illustrated.

The present invention is based on city dweller's civil water behaviour classification method of intellectual water meter, as shown in Figure 1, comprise the following steps:

Step s101, to be gathered user more than two days in region to be sorted by intellectual water meter use water number certificate.

Step s102, to water number according to carrying out pre-service, pretreated process comprises: the data rejecting full small incidental expenses family; By being averaging according to addition with water number of same user's every day same period; The average consumption calculating same each period of user accounts for the ratio of average daily water consumption, take user as row, each period be row by water ratio, build user civil water eigenmatrix.

Step s103, according to the user's civil water eigenmatrix obtained after pre-service, to the classification carrying out based on fuzzy clustering with water behavior of user, process is as follows:

Calculate the correlation matrix of user's civil water eigenmatrix;

By correlation matrix on the occasion of change, form fuzzy matrix;

Fuzzy matrix is transformed into fuzzy equivalent matrix;

Calculate the intercept battle array of fuzzy equivalent matrix, using the value of the empirical value of λ during optimal classification effect as intercept battle array, obtain the classification results of user's use water behavior.

Explain the detailed process of above-mentioned pre-service and fuzzy clustering below in detail.

(1) invalid data is rejected

Because the user (day part water number is according to the user being all zero) of useless water number certificate can make model subsequent arithmetic occur invalid value, therefore, first eliminating is the user data of null value entirely, i.e. the vacant user in house.Then, the water consumption average of each user within 96 time periods of every day (every 15 minutes image data once, gather 96 every day and use water number certificate) in a period of time is got, obtain raw data matrix, wherein line number is number of users, and columns is 96, represents 96 periods to use horizontal average.

(2) data merge

For ensureing that data are representational while, reduce data volume.The water consumption of time adjacent segments merges by this model, and totally 8 time periods are analyzed will to be divided into 23:00-2:00,2:00-5:00,5:00-8:00,8:00-11:00,11:00-14:00,14:00-17:00,17:00-20:00,20:00-23:00 with the water time.Table 1 is the time type of representative of each time period.

Table 1 time period type list

Time period	Type
		23:00-2:00	The late into the night
2:00-5:00	Morning
		5:00-8:00	Morning
8:00-11:00	Work
		11:00-14:00	Lunch
14:00-17:00	Work
		17:00-20:00	Dinner
20:00-23:00	Night

(3) data normalization

For get rid of original water number according between not homometric(al) on classification impact, reply raw data carries out standardization, by original water number according to the civil water ratio data be transformed to shared by day part, realize the standardization (all Data distribution8 are between 0 and 1) of data, obtain, with water ratio data matrix, making it participate in classification with identical magnitude by standardized transformation.Account form is as follows:

$x_{ij}=\frac{{x^{\′}}_{ij}}{\underset{j=1}{\overset{8}{\Σ}}{x^{\′}}_{ij}}---\left(1\right)$

X _ijrepresent the civil water ratio of i-th user in the j moment.X' _ijrepresent the true water consumption of i-th user in the j moment.The user's use water eigenmatrix x after standardization is obtained with this _ij.Row representative of consumer in matrix, row represent uses water ratio.

(4) correlation matrix is calculated

Correlation matrix is made up of the related coefficient between matrix respectively arranges, and the element of correlation matrix i-th row jth row is related coefficients that original matrix i-th arranges and jth arranges.Correlation coefficient r _ijcomputing formula is as follows:

$r_{ij}=\frac{\mathrm{cov}(x_i,x_j)}{\sqrt{{\mathrm{Dx}}_i}\sqrt{{\mathrm{Dx}}_j}}---\left(2\right)$

cov(x _i,x _j)＝E((x _i-E(x _i))·(x _j-E(x _j)))(3)

Wherein x _ijfor the original matrix after standardization, Dx _ifor variance, cov (x _i, x _j) be covariance.Correlation coefficient r between observation data matrix computations variable _ij(1≤i, j≤s), m, n are original matrix ranks numbers, s=max (m, n), form correlation matrix r'=(r _ij) _{s × s}.

(5) fuzzy matrix is calculated

Similarity coefficient matrix r ' in element be compressed between 0 and 1, form fuzzy matrix R.

$\begin{array}{cc}{R}_{ij}=\frac{1}{2}(1+{r^{\′}}_{ij})& (1\≤i,j\≤s)\end{array}---\left(4\right)$

(6) fuzzy matrix R is transformed into fuzzy equivalent matrix

Above-mentioned fuzzy matrix does not have equivalence, therefore needs to be translated into fuzzy equivalence battle array.Concrete steps are: R → R ²→ R ⁴→...→ R ^2k→ R ^k, by R involution RR=R ², then involution R ²r ²=R ⁴... until R ^2k=R ^ktill, R ^kbe fuzzy equivalent matrix.

(7) λ Level Matrix completes fuzzy classification

R ^λbe the matrix of a s × s, work as R ^k _ijduring>=λ, R ^λ _ij=1, work as R ^k _ijduring < λ, R ^λ _ij=0, obtain R ^kcorresponding λ cuts a gust R ^λ, R ^λthe matrix of to be a value be 0 and 1, the identical row of its intermediate value is classified as a class.The determination of threshold value λ can produce material impact to result, and often according to the actual needs, its optimum value depends on experience to threshold value λ.Through testing and comparing, obtain optimal classification effect when λ=0.8.

The best value of Level Matrix depends on experience.The fuzzy equivalent matrix numeric distribution of this experiment is between [0.51,1], and therefore this experiment in this interval, will carry out 3 Level Matrix analysis design mothod by λ=0.7/0.8/0.9 altogether, determine that optimal classification cuts value based on experience value.In following analysis, use u ₁, u ₂... u ₁₈, represent 18 validated user water number certificates, the user data of the other 2 useless water in family is invalid data (see table 2).

Table 2 period uses water tables of data after merging

When λ=0.9, be divided into 10 classes, (u ₁, u ₅, u ₈, u ₉, u ₁₃, u ₁₄, u ₁₆); (u ₂); (u ₃) (u ₄) (u ₆, u ₁₁, u ₁₂) (u ₇) (u ₁₀) (u ₁₅) (u ₁₇) (u ₁₈) as seen in figures 3-6.Fig. 3 is first kind user, and its quantity is maximum, and has and significantly use water feature.The peak water time of such 7 family user concentrates on night-time hours, also has a small amount of water in the morning, and compared with rush hour section, other period use water are less.Fig. 4 is Equations of The Second Kind user, comprises 3 family families, and its water use peak concentrates on this period of 17:00-20:00, also has use water on a small quantity at 5:00-8:00 and 20:00-23:00.Fig. 5 is other 8 class water features, respectively comprises a family family.By analyzing, partial category has identical uses water feature, such as u ₃with u ₇, see Fig. 6.Therefore, during λ=0.9 with water tagsort not will part have the user of similar features to be summarized as a class.

During λ=0.7, be divided into two classes.(u ₃, u ₇) be a class, other are another kind of.Fig. 6 is the first kind (u ₃, u ₇), unallocated classification out when this classification is λ=0.9, has and significantly uses water feature, and its water use peak, in 11:00-14:00, the 14:00-17:00 time period, also has in the 8:00-11:00 time period and uses water on a small quantity.But Equations of The Second Kind water feature comprises 16 family family, wherein u ₁₇with u ₁₈obviously different with water with other by water feature, therefore do not reach the requirement with water feature exhaustive division.

When λ=0.8, be divided into following five classes: (u ₁, u ₂, u ₄, u ₅, u ₆, u ₈, u ₉, u ₁₁, u ₁₂, u ₁₃, u ₁₄, u ₅, u ₁₆) (u ₃, u ₇) (u ₁₀) (u ₁₇) (u ₁₈), include (the u in above-mentioned classification ₃, u ₇), remaining segmented by water feature, be best Clustering Effect simultaneously.Its all kinds of water feature as is seen in figs 7-10.

Through on-site verification, result is as follows: Fig. 7 is the first kind, comprises 13 family users, and be characterized as water use peak and be positioned at night hours section, secondary peak is positioned at morning hours section.Such number of users is maximum, and user is common working clan family substantially, so peak water concentrates on morning and night.In family's configuration aspects, the family of two mouthfuls by day working hour almost useless water.Working hour also has and uses water on a small quantity by day for two generations or family of three generations under one roof.Through sample survey checking, consistent with tested result of determination.Equations of The Second Kind comprises 2 family users, sees Fig. 6, and its peak water period concentrates on 11:00-14:00, the 14:00-17:00 time period, also has a small amount of water in the 8:00-11:00 time period, and night use water less.Such user is that therefore peak water concentrates on work by day period and noon the commercial user of community office.Fig. 8 is the 3rd class, comprises 1 family user, and the peak water period of this user is 23:00-2:00, and at 20:00-23:00 and 5:00-8:00 also useful water.This user job period is that the quitting time is more late, and therefore peak water concentrates on morning.Fig. 9 is the 4th class, and comprise 1 family user, peak water concentrates on 8:00-11:00,17:00-20:00 and 23:00-2:00, this category feature and first kind feature similarity, but its water consumption is comparatively large at dead of night, there are habits and customs of staying up late in user.Figure 10 is the 5th class, and comprise 1 family user, peak water concentrates on the late into the night and morning, and daytime, use water was less, this user's daily schedule for working at night, daytime have a rest.

The present invention is based on resident living water detail data, take day as time scale, Fuzzy Cluster Analysis method is adopted to carry out cluster analysis to the behavior of resident's civil water, adopt the method for empirical value Level Matrix, classify according to by the family structure, job specification etc. of water feature to 20 family families (valid data are 18 families), classification results can be applicable to the aspects such as water undertaking's aid decision making and neghborhood services.Such as, for the long-term vacant user in house, because tap water prolonged stay will cause water age long and part water-quality guideline to exceed standard in pipeline, therefore can be this user to provide and use water reminding service, when intellectual water meter perception user's water, water undertaking's note machine system automatically sends after water tap unlatching discharges water 5 minutes by inform by short message user and re-uses, and ensures user's water safety; For the user handled official business at Residential Area, because urbanite water consumption is different from the toll price of commercial water, the Water consumption type verification work that therefore can be water undertaking provides Data support, retrieves the economic loss of water undertaking; To stay up late user be accustomed to existing for night shift user, cell management department can be it provides maintenance to keep healthy relevant information, for user provides intimate service.

To sum up, this method is carried out analyzing to water for city's residential use detail data and is excavated, by to the fuzzy clustering carried out with water behavior without priori sample, thus provide decision support for the planning of the management of residential quarter and city dweller's water, supply and research application, and for analyzing city dweller's water behavior in different time and space scales, simulation water consumption lays the foundation.

Above-listed detailed description is illustrating for possible embodiments of the present invention, and this embodiment is also not used to limit the scope of the claims of the present invention, and the equivalence that all the present invention of disengaging do is implemented or changed, and all should be contained in the scope of the claims of this case.

Claims (5)

1., based on city dweller's civil water behaviour classification method of intellectual water meter, it is characterized in that, comprise step:

What gather user more than two days in region to be sorted by intellectual water meter uses water number certificate;

Carry out pre-service to by water number certificate, pretreated process comprises: the data rejecting full small incidental expenses family; By being averaging according to addition with water number of same user's every day same period; The average consumption calculating same each period of user accounts for the ratio of average daily water consumption, take user as row, each period be row by water ratio, build user civil water eigenmatrix;

According to the user's civil water eigenmatrix obtained after pre-service, to the classification carrying out based on fuzzy clustering with water behavior of user, process is as follows:

Calculate the correlation matrix of user's civil water eigenmatrix;

By correlation matrix on the occasion of change, form fuzzy matrix;

Fuzzy matrix is transformed into fuzzy equivalent matrix;

Calculate the intercept battle array of fuzzy equivalent matrix, using the value of the empirical value of λ during optimal classification effect as intercept battle array, obtain the classification results of user's use water behavior.

2. the city dweller's civil water behaviour classification method based on intellectual water meter according to claim 1, is characterized in that,

The frequency acquisition of collection water number certificate be every 15 minutes once, and the every workday in one month gather.

3. the city dweller's civil water behaviour classification method based on intellectual water meter according to claim 1 and 2, is characterized in that,

Be divided into 8 periods by one day 24 hours, be followed successively by: 23:00-2:00,2:00-5:00,5:00-8:00,8:00-11:00,11:00-14:00,14:00-17:00,17:00-20:00,20:00-23:00.

4. the city dweller's civil water behaviour classification method based on intellectual water meter according to claim 1 and 2, is characterized in that,

The empirical value of λ during optimal classification effect is 0.8.

5. the city dweller's civil water behaviour classification method based on intellectual water meter according to claim 1 and 2, is characterized in that,

Element representation in described user's civil water eigenmatrix is x _ij, and x _ijrepresent the civil water ratio of i-th user in the j period, x' _ijrepresent the true water consumption of i-th user in the j period;

Correlation coefficient r in correlation matrix _ijcomputing formula is as follows:

$r_{ij}=\frac{\mathrm{cov}(x_i,x_j)}{\sqrt{{\mathrm{Dx}}_i}\sqrt{{\mathrm{Dx}}_j}}$

cov(x _i,x _j)＝E((x _i-E(x _i))i(x _j-E(x _j)))

Wherein, x _i, x _jrepresent the element of described user's civil water eigenmatrix i-th row and the element of jth row respectively, Dx _ifor variance, cov (x _i, x _j) be covariance, correlation matrix r'=(r _ij) _{s × s}, s represents the maximal value of described user's civil water eigenmatrix ranks number;

Element in correlation matrix r' is compressed between 0 and 1, forms fuzzy matrix R:

$\begin{array}{cc}{R}_{ij}=\frac{1}{2}(1+{r^{\&prime;}}_{ij})& (1\&le;i,j\&le;s)\end{array}$

Fuzzy matrix R is transformed into the process of fuzzy equivalent matrix: R → R ²→ R ⁴→...→ R ^2k→ R ^k, by R involution RR=R ², then involution R ²r ²=R ⁴... until R ^2k=R ^ktill, R ^kbe fuzzy equivalent matrix;

Work as R ^k _ijduring>=λ, R ^λ _ij=1, work as R ^k _ijduring < λ, R ^λ _ij=0, obtain R ^kcorresponding Level Matrix R ^λ, R ^λthe matrix of to be a value be 0 and 1, the identical row of its intermediate value is classified as a class.