Mathematical modeling - Urban Smart Growth Model Evaluation System
problem
- Select two specific cities to evaluate the effectiveness of their growth plans
- Propose feasible plans to help our selected cities grow wisely
- Evaluate the success of the plan
terms of settlement
- On the basis of collecting a large number of urban smart growth data, 25 indicators are selected through principal component analysis. On this basis, a new three-level index system is constructed, and the weight vector and grouping decision-making method are obtained.
- Establish a measurement system.
- Predict the changes of the city. Then the combined prediction model is used to minimize the prediction error.
- Two different population growth models (PGM) are used to predict population changes.
Main algorithms:
- Fuzzy evaluation: realize the quantification of qualitative indicators.
- Entropy weight method (EWM): obtain the weight vector of the index
Group decision making method (GDM): obtain the weight vector of the second type of index - K-means clustering algorithm: success criteria for obtaining smart growth model
- SVM vector machine and weighted moving average method (WMAM): predict urban change
- Population growth model (PGM): predicting population changes
Fuzzy evaluation
1. Related concepts of fuzzy evaluation
- Factor set (evaluation index set): U U U
- Comment set (result of evaluation): V V V
- Weight set (weight of indicator): A A A
2. Method steps (first level fuzzy evaluation problem)
- The factor set, comment set and weight set are determined respectively. In this paper, A is obtained by analytic hierarchy process (AHP)
- Determining fuzzy judgment synthesis matrix
- Comprehensive evaluation
P
=
A
∘
R
=
(
p
1
,
p
2
,
p
3
.
.
.
p
n
)
P=A\circ R=(p_{1},p_{2},p_{3}...p_{n})
P=A∘R=(p1,p2,p3...pn)
In this paper, Let F represent the score set, then the final formula is Z = P ⋅ F Z=P·F Z=P⋅F
Entropy weight method (EWM)
1. Forward processing: judge whether there are negative numbers in the input matrix. If so, re standardize to non negative interval. In this paper, normalize the data
- Standardized formula
2. Calculate the probability matrix with the standardized matrix:
The entropy value is obtained by using the formula:
3. Calculate the correlation coefficient value (S represents the standard index entropy obtained by clustering a large number of data)
4. Calculate the entropy weight
MATLAB
function [s,w]=shang(x,ind)
%The entropy method is used to calculate each index(Column) and the score of each data row
%x Is the original data matrix, A row represents a sample, Each column corresponds to an indicator
%ind Indicator vector, indicating whether each column is a positive indicator or a negative indicator. 1 represents a positive indicator and 2 represents a negative indicator
%s Return the score of each line (sample), w Returns the weight of each column
[n,m]=size(x); % n Samples, m Indicators
%%Normalization of data
for i=1:m
if ind(i)==1 %Positive index normalization
X(:,i)=guiyi(x(:,i),1,0.002,0.996); %If normalized to[0,1], 0 There will be problems
else %Negative index normalization
X(:,i)=guiyi(x(:,i),2,0.002,0.996);
end
end
%%Calculation section j Under the first indicator, the second i Proportion of samples in the index p(i,j)
for i=1:n
for j=1:m
p(i,j)=X(i,j)/sum(X(:,j));
end
end
%%Calculation section j Entropy of two indexes e(j)
k=1/log(n);
for j=1:m
e(j)=-k*sum(p(:,j).*log(p(:,j)));
end
d=ones(1,m)-e; %Calculate information entropy redundancy
w=d./sum(d); %Weight calculation w
s=100*w*X'; %Seek comprehensive score
K-Means clustering algorithm
In this paper, the data set is divided into three parts (k=3)
The sum of variance of the distance between the sample value and the centroid is obtained.
The goal of K-means clustering algorithm is to minimize the variance. (this is also the purpose of PCA dimensionality reduction previously)
Input sample
- First enter the sample set D = { x 1 , x 2 , . . . x m } D=\{x1,x2,...xm\} D={x1,x2,...xm}, clustering tree k, maximum number of iterations N
- Randomly select k samples from dataset D as the initial K mean vectors: { μ 1 , μ 2 , . . . , μ k } \{μ1,μ2,...,μk\} {μ1,μ2,...,μk}
Start iteration N times
-
Initialize cluster partition C to C t = ∅ t = 1 , 2... k Ct=∅t=1,2...k Ct = ∅ t=1,2...k (k clusters)
-
For i=1,2... m (sample size), calculate the sample xi and each mean vector μ Distance of j(j=1,2,... k) d i j = dij= dij=
-
Mark xi as the category corresponding to Dij λ i. Update at this time C λ i = C λ i ⋃ x i C_{λi}=C_{λi}\bigcup{x_{i}} Cλi=Cλi⋃xi
-
For j=1,2,..., k, recalculate the new centroid for all sample points in Cj μ j = 1 ∣ C j ∣ ∑ x ∈ C j x μ_{j}=\frac{1}{|Cj|}\sum_{x\in C_{j}} x μj=∣Cj∣1∑x∈Cjx
-
If all k mean vectors do not change, it ends
End, output cluster division C = { C 1 , C 2 , . . . C k } C=\{C1,C2,...Ck\} C={C1,C2,...Ck}
code
%%%come from csdn
clc;
clear;
time=0;
k= ; %%%cluster tree k In the paper, it is determined as 3
x=; %%%sample
z=x(1:k,1:2);
z1=zeros(k,2);
while time<=1000
count=zeros(k,1);%%%How many points belong to each cluster center
allsum=zeros(k,2);%%%The sum of the horizontal and vertical distances from each point belonging to the cluster center to the cluster center
num=[];%%%%Record the number of the central point belonging to the cluster
temp=[];%%%%Record the distance from each point to the cluster center and find the smallest one
for i=1:size(x,1)
for j=1:k
temp(j,1)=sqrt((z(j,1)-x(i,1)).^2+(z(j,2)-x(i,2)).^2);%The first i Points to j Distance between cluster centers
temp(j,2)=j;
end
temp=sortrows(temp,1);%about temp Sort in descending order of the first column
c=temp(1,2);%%%Find the number of the cluster center with the minimum distance
count(c)=count(c)+1;%%%The number of points belonging to this center plus 1
num(c,count(c))=i;
allsum(c,1)=allsum(c,1)+x(i,1);%%%Find the sum of abscissa of all points belonging to the cluster center
allsum(c,2)=allsum(c,2)+x(i,2);%%%Find the sum of the ordinates of all points belonging to the cluster center
end
z1(:,1)=allsum(:,1)./count(:);%%%Sum of abscissa of all points in each cluster center/The number of points belonging to the cluster center is the new abscissa
z1(:,2)=allsum(:,2)./count(:);
if (z==z1)%%%If the point no longer changes
break;
else
z=z1;
end
time=time+1;
end
plot(x(:,1),x(:,2),'r*');
hold on;
plot(z1(:,1),z1(:,2),'bo');
num(num==0)=NaN;
for i=1:k
if (count(i)==0)%%%This is to prevent some clustering centers from not matching points
continue;
else
disp(['The first',num2str(i),'Class is:',num2str(num(i,:))]);
end
end
Weighted moving average method (WMAM)
Weighted moving average refers to multiplying individual data by different values when calculating the average. In technical analysis, the latest value of n-day WMA is multiplied by N, the next nearest value is multiplied by n-1, and so on until 0
[external chain picture transfer failed. The source station may have anti-theft chain mechanism. It is recommended to save the picture and upload it directly
- In this thesis,
λ
i
\lambda_{i}
λ i ， is statistical data
x
t
−
i
x_{t-i}
Weight of xt − i +
matlab code is as follows
y= ; %%%% Row vector w= ; %%% Column vector m=length(y); n= ;%%%from w determine for i=1:m-n+1 yhat(i)=y(i:i+n-1)*w;end yhat %%% Multiply row by column vector err=abs(y(n+1:m)-yhat(1:end-1))./y(n+1:m) %%%relative error T_err=1-sum(yhat(1:end-1))/sum(y(n+1:m)) %%%Relative error ratio y=yhat(end)/(1-T_err)
Population growth model (PGM)
The model of population growth in Ningguo City is
1
p
(
t
)
=
1
P
0
−
b
l
n
t
\frac{1}{p(t)}=\frac{1}{P_{0}}-blnt
p(t)1=P01−blnt
Parameter b is based on the population in recent years.