Real estate market popularity index prediction method based on multi-source data space-time association
Through the method of space-time correlation of multi-source data, a multi-source index system is formed and a time graph convolution network is used to solve the problem of insufficient prediction of single indicators in the existing technology, and a more accurate prediction of the real estate market popularity index is achieved, and the city is considered Dependencies between.
Patent Information
- Application Number
- CN202411260505.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-10
AI Technical Summary
The existing real estate market popularity index prediction method is based on a single indicator, and cannot comprehensively and accurately predict the real estate market popularity and fail to effectively characterize the characteristics of market laws.
Using a method based on space-time correlation of multi-source data, a multi-source indicator system is formed, including new house transaction prices, new house sales transaction volume, second-hand house sales transaction volume, second-hand house sales transaction volume, land transaction price, land transfer area, house rental transaction price and Market sentiment value is predicted through data preprocessing, normalization, entropy value calculation, cross-correlation matrix construction and time graph convolution network processing.
A more comprehensive and accurate prediction of the real estate market popularity index is achieved, taking into account the spatial and temporal dependence between different cities, and improving the accuracy of the prediction.
Smart Images

Figure CN119067714A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of real estate, and in particular to a method for predicting a real estate market heat index based on spatiotemporal association of multi-source data. Background Art
[0002] The real estate market is a pillar industry for economic growth, and its development has driven the prosperity of basic industries of the national economy, such as construction, industry, commerce and transportation. However, once the real estate market is in a state of disorder and structural imbalance, it will also threaten the healthy development of the macro-economy.
[0003] In order to grasp the real estate situation in a timely manner, the real estate market heat is usually used as a description of the growth rate of various indicators in the current real estate market. Once the real estate market heat situation can be accurately predicted, it will be helpful to grasp the real estate market situation.
[0004] However, the existing real estate market heat index prediction methods have shortcomings: since the heat of the real estate market will be affected by multiple factors and the mutual influence of these multiple factors, the existing real estate market heat index prediction methods are usually based on a single indicator, and cannot form a systematic multi-source indicator system that characterizes the regular characteristics of the real estate market, resulting in difficulty in predicting the real estate market heat index more comprehensively and accurately. Summary of the invention
[0005] The technical problem to be solved by the present invention is to provide a real estate market heat index prediction method based on spatiotemporal association of multi-source data in view of the above-mentioned prior art.
[0006] The technical solution adopted by the present invention to solve the above technical problems is: a real estate market heat index prediction method based on spatiotemporal association of multi-source data, characterized in that it includes the following steps:
[0007] Step 1, pre-form a multi-source indicator system that characterizes the laws of the urban real estate market; wherein the multi-source indicator system includes new house transaction prices, new house sales transaction volumes, second-hand house transaction prices, second-hand house sales transaction volumes, land transaction prices, land transfer areas, house rental transaction prices, and market sentiment values within the city, and the market sentiment values are obtained by processing news text data related to the real estate market within the city; the city is a municipality directly under the Central Government or a prefecture-level city within the national administrative region;
[0008] Step 2: obtaining a multi-source indicator data set of all cities in the national administrative region according to the multi-source indicator system, and preprocessing the multi-source indicator data set to delete abnormal data to obtain a preprocessed multi-source indicator data set;
[0009] Among them, the indicator types in the multi-source indicator data set correspond to the indicator types in the multi-source indicator system one by one, and the multi-source indicator data set includes the indicator data corresponding to different cities at different times; the multi-source indicator data set after preprocessing is marked as X, X = {X i,p (t)};X i,p (t) is the i-th indicator data X of the p-th city in the preprocessed multi-source indicator data set X i The preprocessed value at time t, 1≤p≤M, 1≤i≤N; M is the total number of cities in the national administrative region, and N is the total number of indicator data types in the multi-source indicator system;
[0010] Step 3, normalize all preprocessed data in the preprocessed multi-source indicator data set to obtain a normalized multi-source indicator data set, and calculate the indicator weight of each indicator data in the normalized multi-source indicator data set at different times and the entropy value of each indicator data;
[0011] Step 4: Calculate the difference coefficient and the corresponding weight value of each indicator data according to the entropy value of each indicator data, and calculate the real estate heat index of each city based on the obtained weight values;
[0012] Step 5, calculating the real estate heat index of all cities in the national administrative region at different times, and obtaining the logarithmic change rate of the real estate heat of each city in a preset time period based on the obtained real estate heat index;
[0013] Step 6, based on the obtained logarithmic change rate of real estate heat of each city, calculate the cross-correlation matrix representing the real estate heat index of all cities in the national administrative region at different times;
[0014] Step 7, performing random decomposition processing on the cross-correlation matrix to obtain decomposed matrix elements, and performing denoising processing on the decomposed matrix elements to obtain a denoised cross-correlation matrix;
[0015] Step 8: Based on the obtained denoised cross-correlation matrix, a plane maximum filter map of the denoised cross-correlation matrix is constructed, and the plane maximum filter map is used as a real estate heat correlation network representing the mutual correlation between all cities in the national administrative region;
[0016] Step 9: The obtained real estate heat association network and the time series of the real estate heat index of each city in the national administrative region are used as input and input into the time graph convolution network for processing, and the output result after processing by the time graph convolution network is used as the predicted value of the real estate market heat index of each city in the national administrative region.
[0017] Improved, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, the abnormal data includes duplicate data, missing data and deleted abnormal data.
[0018] Furthermore, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 3, the normalization process is performed to obtain the normalized multi-source indicator data set as follows:
[0019]
[0020] The normalized multi-source indicator data set is labeled as x, where x = {x i,p (t)}; x i,p (t) represents the i-th indicator data X of the p-th city i,p The normalized data value at time t, max(X i,p ) represents the i-th indicator data X i,p The maximum preprocessed data value in the preprocessed indicator data set X, min(X i,p ) represents the i-th indicator data X of the p-th city i,p The minimum preprocessed data value in the preprocessed indicator data set X.
[0021] Improved, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 3, the index weights of each index data at different times and the entropy value calculation methods of each index data are as follows:
[0022]
[0023] Among them, U i,p (t) is the i-th indicator data X of the p-th city i,p The weight of the index it occupies at time t; is the normalized value x i,p (t) is the value of the index data after the self-increment processing, △ i is a positive number infinitely close to zero;
[0024]
[0025] Among them, e i,p is the entropy value of the i-th indicator data of the p-th city; Represents the i-th indicator data X of the p-th city i,p The self in the time period [t1,t m ] is the total number of corresponding indicator weights.
[0026] Further improvement, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 4, the difference coefficient of each indicator data and the real estate heat index of each city are calculated as follows:
[0027] g i,p =1-e i,p ;
[0028]
[0029] Among them, g i,p is the i-th indicator data X of the p-th city i,p The coefficient of variation, W i,p is the i-th indicator data X of the p-th city i,p The weight value of S p (t) is the real estate heat index of the p-th city at time t.
[0030] Furthermore, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 5, the logarithmic change rate of real estate heat in the city within a preset time period is calculated as follows:
[0031]
[0032] H p (t) = lnS p (t+△t)-lnS p (t); 1≤p≤M;
[0033] Among them, h p (t) represents the logarithmic change rate of real estate popularity in the p-th city within the preset time period △t, <H p (t) represents the average value of the sum of the real estate heat index of the p-th city within the preset time period △t, σ p It is the standard deviation of the real estate heat index of the p-th city within the preset time period △t; the preset time period is a multiple of one day.
[0034] Further improvement, in the real estate market heat index prediction method based on multi-source data spatiotemporal association, in step 6, the cross-correlation matrix is calculated as follows:
[0035] C pq = <h P (t)·h q (t)>1≤q≤M;
[0036] The cross-correlation matrix is marked as C, and the cross-correlation matrix C is a real symmetric matrix, each matrix element on the diagonal of the real symmetric matrix is 1, and the matrix element on the non-diagonal of the real symmetric matrix has a value range of [-1, 1]; C pq is a matrix element of the cross-correlation matrix C. pq represents the strength of the correlation effect between city p and city q on real estate popularity, h q (t) shows the logarithmic change rate of real estate heat in the qth city within the preset time period △t.
[0037] Furthermore, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 7, the manner of processing the matrix element after the decomposition process is as follows:
[0038]
[0039] Among them, λ α Represents the matrix element C pq The αth eigenvalue of , K is the total number of elements in the eigenvector corresponding to the αth eigenvalue, represents the pth element in the eigenvector corresponding to the αth eigenvalue; represents the qth element in the eigenvector corresponding to the αth eigenvalue; Represents the cross-correlation value between the p-th element and the q-th element under the eigenvector corresponding to the α-th eigenvalue;
[0040] In step 7, the method of obtaining the denoised cross-correlation matrix is as follows:
[0041]
[0042] Among them, P r (λ) is the probability distribution function of the eigenvalue λ, is the maximum value of the eigenvalue λ, is the minimum value of the eigenvalue λ, T is the length of the random time series, and N is the total number of random time series with a length of T.
[0043] Furthermore, in the real estate market heat index prediction method based on spatiotemporal association of multi-source data, in step 9, the real estate market heat index prediction value of each city in the national administrative region is obtained as follows:
[0044] f(X,A)=σ(RELU(A * XW0)W1); and f(X,A)∈R N×T ';
[0045] W0∈R P×H ; W1∈RH×T ';
[0046] Where f(X,A) represents the output of the temporal graph convolutional network GCN when the input is X and A, X is the feature matrix, A is the real estate heat association network as an adjacency matrix, A * represents the preprocessed matrix of the adjacency matrix A; W0 represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix X, and H is the number of hidden units in the hidden layer; W1 represents the weight matrix from the hidden layer to the output layer, T' represents the prediction length, σ represents the activation function, and RELU represents the nonlinear activation function in the temporal graph convolutional network GCN.
[0047] Compared with the prior art, the advantages of the present invention are as follows: the real estate market heat index prediction method based on the spatiotemporal correlation of multi-source data of the present invention forms a multi-source indicator system to comprehensively characterize the investment intentions of real estate market entities and their expectations of market trends, and obtains a multi-source indicator data set of all cities in the national administrative region and a pre-processed multi-source indicator data set based on the multi-source indicator system, and then normalizes the pre-processed multi-source indicator data set, calculates the weights of different indicators, calculates the entropy values of different indicator data, and calculates the indicator data difference coefficient and the indicator data weight value, and then calculates the real estate heat index and the logarithmic change rate of real estate heat of each city, and then calculates the real estate heat index of all cities in the national administrative region based on the logarithmic change rate of real estate heat of each city. The cross-correlation matrix of the real estate heat index at different times is obtained, and the denoised cross-correlation matrix is obtained for the cross-correlation matrix. The planar maximum filtered graph obtained by the denoised cross-correlation matrix is then used as the real estate heat association network. The real estate heat association network and the time series of the real estate heat index of cities in the national administrative region are input into the temporal graph convolutional network for processing, and the output results of the temporal graph convolutional network are used as the predicted values of the real estate market heat index of cities in the national administrative region. The temporal graph convolutional network is used to capture the spatial and temporal dependence of the real estate markets of different cities, thereby taking into account the impact of surrounding adjacent cities on the real estate market of the target city, making the prediction of the real estate market heat index of different cities more comprehensive and accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a flow chart of a method for predicting real estate market heat index based on spatiotemporal association of multi-source data in an embodiment of the present invention. DETAILED DESCRIPTION
[0049] The present invention is further described in detail below with reference to the accompanying drawings.
[0050] This embodiment provides a real estate market heat index prediction method based on the spatiotemporal association of multi-source data. Figure 1 As shown, the real estate market heat index prediction method based on spatiotemporal association of multi-source data includes the following steps:
[0051] Step 1, pre-form a multi-source indicator system that characterizes the laws of the urban real estate market; wherein the multi-source indicator system includes new house transaction prices, new house sales transaction volumes, second-hand house transaction prices, second-hand house sales transaction volumes, land transaction prices, land transfer areas, house rental transaction prices, and market sentiment values within the city, and the market sentiment values are obtained by processing news text data related to the real estate market within the city; the city is a municipality directly under the Central Government or a prefecture-level city within the national administrative region;
[0052] Specifically, in this embodiment, the specific calculation method of the "market sentiment value" is: using the "Jieba" tool ( https: / / github.com / fxsjy / jieba) , this method can reliably extract positive and negative emotions from short texts, thereby quantifying the emotions in the text. Specifically, an artificially generated Chinese sentiment vocabulary dictionary is used to match sentiment words based on the dictionary method to determine the sentiment tendency of the text. The sentiment of each text is represented by two values, s + and - Representing positive and negative emotions respectively, the sentiment values of all texts every day are added together to obtain the time series of market sentiment values;
[0053] Step 2: obtaining a multi-source indicator data set of all cities in the national administrative region according to the multi-source indicator system, and preprocessing the multi-source indicator data set to delete abnormal data to obtain a preprocessed multi-source indicator data set;
[0054] Among them, the indicator types in the multi-source indicator data set correspond to the indicator types in the multi-source indicator system one by one, and the multi-source indicator data set includes the indicator data corresponding to different cities at different times; the multi-source indicator data set after preprocessing is marked as X, X = {X i,p (t)};X i,p (t) is the i-th indicator data X of the p-th city in the preprocessed multi-source indicator data set X i The preprocessed value at time t, 1≤p≤M, 1≤i≤N; M is the total number of all cities in the national administrative region, and N is the total number of indicator data types in the multi-source indicator system; the abnormal data here include duplicate data, missing data, and deleted abnormal data;
[0055] Step 3, normalize all preprocessed data in the preprocessed multi-source indicator data set to obtain a normalized multi-source indicator data set, and calculate the indicator weight of each indicator data in the normalized multi-source indicator data set at different times and the entropy value of each indicator data;
[0056] Specific to this embodiment:
[0057] The normalization process to obtain the normalized multi-source indicator data set is as follows:
[0058]
[0059] Among them, the normalized multi-source indicator data set is marked as x, x = {x i,p (t)}; x i,p (t) represents the i-th indicator data X of the p-th city i,p The normalized data value at time t, max(X i,p ) represents the i-th indicator data X i,p The maximum preprocessed data value in the preprocessed indicator data set X, min(X i,p ) represents the i-th indicator data X of the p-th city i,p The minimum preprocessed data value in the preprocessed indicator data set X;
[0060] The indicator weights of the indicator data at different times and the entropy value calculation methods of the indicator data are as follows:
[0061]
[0062] Among them, U i,p (t) is the i-th indicator data X of the p-th city i,p The weight of the index it occupies at time t; is the normalized value x i,p (t) is the value of the index data after the self-increment processing, △ i is a positive number infinitely close to zero;
[0063]
[0064] Among them, e i,p is the entropy value of the i-th indicator data of the p-th city; Represents the i-th indicator data X of the p-th city i,p The self in the time period [t1,t m ] The total number of corresponding indicator weights;
[0065] Step 4: Calculate the difference coefficient and the corresponding weight value of each indicator data according to the entropy value of each indicator data, and calculate the real estate heat index of each city based on the obtained weight values;
[0066] Specifically in this embodiment, the difference coefficient of each indicator data and the calculation method of the real estate heat index of each city are as follows:
[0067] g i,p =1-e i,p ;
[0068]
[0069] Among them, g i,p is the i-th indicator data X of the p-th city i,p The coefficient of variation, W i,p is the i-th indicator data X of the p-th city i,p The weight value of S p (t) is the real estate heat index of the p-th city at time t;
[0070] Step 5, calculating the real estate heat index of all cities in the national administrative region at different times, and obtaining the logarithmic change rate of the real estate heat of each city in a preset time period based on the obtained real estate heat index; wherein, in this embodiment, the calculation method of the logarithmic change rate of the real estate heat of a city in a preset time period is as follows:
[0071]
[0072] H p (t) = lnS p (t+△t)-lnS p (t); 1≤p≤M;
[0073] Among them, h p (t) represents the logarithmic change rate of real estate popularity in the p-th city within the preset time period △t, <H p (t) represents the average value of the sum of the real estate heat index of the p-th city within the preset time period △t, σ p is the standard deviation of the real estate heat index of the p-th city in the preset time period △t; the preset time period is a multiple of one day, for example, the preset time period is one month, one quarter or one year;
[0074] Step 6, based on the obtained logarithmic change rate of real estate heat of each city, calculate the cross-correlation matrix representing the real estate heat index of all cities in the national administrative region at different times; wherein, in step 6, the cross-correlation matrix is calculated as follows:
[0075] C pq = <h P (t)·h q (t)>1≤q≤M;
[0076] The cross-correlation matrix is marked as C. The cross-correlation matrix C is a real symmetric matrix. Each matrix element on the diagonal of the real symmetric matrix is 1, and the matrix element on the non-diagonal of the real symmetric matrix has a value range of [-1, 1]; C pq is a matrix element of the cross-correlation matrix C. pq represents the strength of the correlation effect between city p and city q on real estate popularity, h q (t) shows the logarithmic change rate of real estate popularity in the qth city within the preset time period △t;
[0077] Step 7, performing random decomposition processing on the cross-correlation matrix to obtain decomposed matrix elements, and performing denoising processing on the decomposed matrix elements to obtain a denoised cross-correlation matrix;
[0078] Specifically, in this embodiment, the manner of obtaining the matrix element after the decomposition process is as follows:
[0079]
[0080] Among them, λ α Represents the matrix element C pq The αth eigenvalue of represents the pth element in the αth eigenvector; represents the qth element in the αth eigenvector; Characterize the cross-correlation value between the p-th element and the q-th element in the α-th characteristic mode;
[0081] In step 7, the method of obtaining the denoised cross-correlation matrix is as follows:
[0082]
[0083]
[0084] and
[0085] Among them, P r (λ) is the probability distribution function of the eigenvalue λ, is the maximum value of the eigenvalue λ, is the minimum value of the eigenvalue λ, T is the length of the random time series, and N is the total number of random time series with a length of T;
[0086] The above-mentioned random time series refers to a sequence composed of random variables, which are arranged in chronological order. In this sequence, each random variable represents an observation or measurement at a specific point in time. It meets the following two conditions: one is a constant mean, that is, the expected value (i.e., mean) of the random variable is constant for all time points; the other is a constant variance, that is, the variance (i.e., variability or degree of dispersion) of the random variable is constant for all time points and does not change over time;
[0087] Step 8: Based on the obtained denoised cross-correlation matrix, construct a plane maximum filter map of the denoised cross-correlation matrix, and use the plane maximum filter map as a real estate heat correlation network that represents the mutual correlation between all cities in the national administrative region; wherein, the method of constructing a plane maximum filter map based on a matrix is a mature technical means and will not be repeated here;
[0088] Step 9: The obtained real estate heat association network and the time series of real estate heat index of each city in the national administrative region are used as inputs to the time graph convolution network for processing, and the output result after the processing of the time graph convolution network is used as the predicted value of the real estate market heat index of each city in the national administrative region. The method for obtaining the predicted value of the real estate market heat index of each city in the national administrative region is as follows:
[0089] f(X,A)=σ(RELU(A * XW0)W1); and f(X,A)∈R N×T ';
[0090] W0∈R P×H ; W1∈R H×T ';
[0091] Where f(X,A) represents the output of the temporal graph convolutional network GCN when the input is X and A, X is the feature matrix, A is the real estate heat association network as an adjacency matrix, A * represents the preprocessed matrix of the adjacency matrix A; W0 represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix X, and H is the number of hidden units in the hidden layer; W1 represents the weight matrix from the hidden layer to the output layer, T' represents the prediction length, σ represents the activation function, and RELU represents the nonlinear activation function in the temporal graph convolutional network GCN.
[0092] In the real estate market heat index prediction method based on spatiotemporal association of multi-source data in this embodiment, a multi-source indicator system that characterizes the laws of the urban real estate market is formed in advance to comprehensively characterize the investment intentions of real estate market entities and expectations of market trends, and a multi-source indicator data set of all cities in the national administrative region is obtained based on the multi-source indicator system, and a pre-processed multi-source indicator data set is further obtained, and then the pre-processed multi-source indicator data set is normalized, different indicator weights are calculated, entropy values of different indicator data are calculated, and indicator data difference coefficients and indicator data weight values are calculated, and then the real estate heat index of each city is calculated based on the obtained weight values, and the logarithmic change rate of the real estate heat of each city in a preset time period is obtained based on the obtained real estate heat index, and then the real estate heat of each city is calculated based on the obtained real estate heat index. The cross-correlation matrix of the real estate heat index of all cities in the national administrative region at different times is calculated by the logarithmic change rate, and the denoised cross-correlation matrix is obtained for the cross-correlation matrix. The planar maximum filtered graph based on the denoised cross-correlation matrix is used as the real estate heat association network, and the obtained real estate heat association network and the time series of the real estate heat index of each city in the national administrative region are used as input and input into the temporal graph convolutional network for processing. The output result of the temporal graph convolutional network processing is used as the predicted value of the real estate market heat index of each city in the national administrative region, which realizes the use of the temporal graph convolutional network to capture the spatial and temporal dependence of the real estate markets of different cities, thereby taking into account the impact of surrounding adjacent cities on the real estate market of the target city, thereby making the prediction of the real estate market heat index of different cities more accurate.
[0093] Although the preferred embodiments of the present invention are described in detail above, it should be clearly understood that various modifications and variations are possible for those skilled in the art. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A real estate market heat index prediction method based on spatiotemporal correlation of multi-source data, characterized in that: The steps include: Step 1, pre-form a multi-source indicator system that characterizes the laws of the urban real estate market; wherein the multi-source indicator system includes new house transaction prices, new house sales transaction volumes, second-hand house transaction prices, second-hand house sales transaction volumes, land transaction prices, land transfer areas, house rental transaction prices, and market sentiment values within the city, and the market sentiment values are obtained by processing news text data related to the real estate market within the city; the city is a municipality directly under the Central Government or a prefecture-level city within the national administrative region; Step 2: obtaining a multi-source indicator data set of all cities in the national administrative region according to the multi-source indicator system, and preprocessing the multi-source indicator data set to delete abnormal data to obtain a preprocessed multi-source indicator data set; Among them, the indicator types in the multi-source indicator data set correspond to the indicator types in the multi-source indicator system one by one, and the multi-source indicator data set includes the indicator data corresponding to different cities at different times; the multi-source indicator data set after preprocessing is marked as X, X = {X i,p (t)};X i,p (t) is the i-th indicator data X of the p-th city in the preprocessed multi-source indicator data set X i The preprocessed value at time t, 1≤p≤M, 1≤i≤N; M is the total number of cities in the national administrative region, and N is the total number of indicator data types in the multi-source indicator system; Step 3, normalize all preprocessed data in the preprocessed multi-source indicator data set to obtain a normalized multi-source indicator data set, and calculate the indicator weight of each indicator data in the normalized multi-source indicator data set at different times and the entropy value of each indicator data; Step 4: Calculate the difference coefficient and the corresponding weight value of each indicator data according to the entropy value of each indicator data, and calculate the real estate heat index of each city based on the obtained weight values; Step 5, calculating the real estate heat index of all cities in the national administrative region at different times, and obtaining the logarithmic change rate of the real estate heat of each city in a preset time period based on the obtained real estate heat index; Step 6, based on the obtained logarithmic change rate of real estate heat of each city, calculate the cross-correlation matrix representing the real estate heat index of all cities in the national administrative region at different times; Step 7, performing random decomposition processing on the cross-correlation matrix to obtain decomposed matrix elements, and performing denoising processing on the decomposed matrix elements to obtain a denoised cross-correlation matrix; Step 8: Based on the obtained denoised cross-correlation matrix, a plane maximum filter map of the denoised cross-correlation matrix is constructed, and the plane maximum filter map is used as a real estate heat correlation network representing the mutual correlation between all cities in the national administrative region; Step 9: The obtained real estate heat association network and the time series of the real estate heat index of each city in the national administrative region are used as input and input into the time graph convolution network for processing, and the output result after processing by the time graph convolution network is used as the predicted value of the real estate market heat index of each city in the national administrative region.
2. The real estate market heat index prediction method based on spatiotemporal association of multi-source data according to claim 1 is characterized in that: The abnormal data includes duplicate data, missing data and deleted abnormal data.
3. The real estate market heat index prediction method based on spatiotemporal association of multi-source data according to claim 2 is characterized in that: In step 3, the normalization process is performed to obtain the normalized multi-source indicator data set as follows: Among them, the normalized multi-source indicator data set is marked as x, x = {x i,p (t)}; x i,p (t) represents the i-th indicator data X of the p-th city i,p The normalized data value at time t, max(X i,p ) represents the i-th indicator data X i,p The maximum preprocessed data value in the preprocessed indicator data set X, min(X i,p ) represents the i-th indicator data X of the p-th city i,p The minimum preprocessed data value in the preprocessed indicator data set X.
4. The real estate market heat index prediction method based on spatiotemporal association of multi-source data according to claim 3 is characterized in that: In step 3, the indicator weights of the indicator data at different times and the entropy values of the indicator data are calculated as follows: Among them, U i,p (t) is the i-th indicator data X of the p-th city i,p The weight of the index it occupies at time t; is the normalized value x i,p (t) is the value of the index data after the self-increment processing, △ i is a positive number infinitely close to zero; Among them, e i,p is the entropy value of the i-th indicator data of the p-th city; Represents the i-th indicator data X of the p-th city i,p The self in the time period [t1,t m ] is the total number of corresponding indicator weights.
5. The method for predicting real estate market heat index based on spatiotemporal association of multi-source data according to claim 4 is characterized in that: In step 4, the coefficient of variation of each indicator data and the real estate heat index of each city are calculated as follows: g i,p =1-e i,p ; Among them, g i,p is the i-th indicator data X of the p-th city i,p The coefficient of variation, W i,p is the i-th indicator data X of the p-th city i,p The weight value of S p (t) is the real estate heat index of the p-th city at time t.
6. The method for predicting real estate market heat index based on spatiotemporal association of multi-source data according to claim 5 is characterized in that: In step 5, the logarithmic change rate of real estate heat in the city within a preset time period is calculated as follows: H p (t)=lnS p (t+△t)-lnS p (t);1≤p≤M; Among them, h p (t) represents the logarithmic change rate of real estate popularity in the p-th city within the preset time period △t, <H p (t)> represents the average value of the sum of the real estate heat index of the p-th city within the preset time period △t, σ p It is the standard deviation of the real estate heat index of the p-th city within the preset time period △t; the preset time period is a multiple of one day.
7. The method for predicting real estate market heat index based on spatiotemporal association of multi-source data according to claim 6 is characterized in that: In step 6, the cross-correlation matrix is calculated as follows: C pq =<h P (t)·h q (t)>;1≤q≤M; The cross-correlation matrix is marked as C, and the cross-correlation matrix C is a real symmetric matrix, each matrix element on the diagonal of the real symmetric matrix is 1, and the matrix element on the non-diagonal of the real symmetric matrix has a value range of [-1, 1]; C pq is a matrix element of the cross-correlation matrix C. pq represents the strength of the correlation effect between city p and city q on real estate popularity, h q (t) shows the logarithmic change rate of real estate heat in the qth city within the preset time period △t.
8. The method for predicting real estate market heat index based on spatiotemporal association of multi-source data according to claim 7 is characterized in that: In step 7, the matrix element after the decomposition is obtained in the following manner: Among them, λ α Represents the matrix element C pq The αth eigenvalue of represents the pth element in the αth eigenvector; represents the qth element in the αth eigenvector; Characterize the cross-correlation value between the p-th element and the q-th element in the α-th characteristic mode; In step 7, the method of obtaining the denoised cross-correlation matrix is as follows: and Among them, P r (λ) is the probability distribution function of the eigenvalue λ, is the maximum value of the eigenvalue λ, is the minimum value of the eigenvalue λ, T is the length of the random time series, and N is the total number of random time series with a length of T.
9. The method for predicting real estate market heat index based on spatiotemporal association of multi-source data according to claim 8 is characterized in that: In step 9, the predicted value of the real estate market heat index of each city in the national administrative region is obtained as follows: f(X,A) = σ(RELU(A * XW0)W1); and f(X,A) ∈ R N×T '; W0∈R P×H ;W1∈R H×T '; Where f(X,A) represents the output of the temporal graph convolutional network GCN when the input is X and A, X is the feature matrix, A is the real estate heat association network as an adjacency matrix, A * represents the preprocessed matrix of the adjacency matrix A; W0 represents the weight matrix from the input layer to the hidden layer, P is the length of the feature matrix X, and H is the number of hidden units in the hidden layer; W1 represents the weight matrix from the hidden layer to the output layer, T' represents the prediction length, σ represents the activation function, and RELU represents the nonlinear activation function in the temporal graph convolutional network GCN.
Citation Information
Patent Citations
Method for dynamically monitoring urban land price
CN102156913A
Traffic speed prediction method and device based on space-time attention graph convolutional network
CN113705880A
OD market aviation passenger flow volume prediction method based on space-time convolutional network
CN114118508A
Digital asset market risk prediction system based on time sequence analysis
CN116611696A
Method, apparatus, system, and computer-readable medium for traffic pattern prediction
US20210209938A1
Cited By
Intelligent business district management method and system based on Internet of Things
CN119904081A
