A community resilience assessment method based on a multi-task neural network

Abstract

The present application relates to a kind of community resilience evaluation methods based on multi-task neural network, comprising the following steps: obtaining community resilience reference index data;The community resilience reference index data collected is transmitted into the community resilience attribute designed in advance, and community resilience attribute includes five aspects of social resilience, economic resilience, institutional resilience, ecological resilience and infrastructure resilience;Community resilience attribute is vectorized, as the input of community resilience evaluation deep neural network;Data artificial labeling: part of data is labeled by expert consultation mode, and the comprehensive resilience of given community and the classification of social resilience, economic resilience, institutional resilience, ecological resilience and infrastructure resilience, that is, which one of excellent, good and general, is needed to be labeled;Community resilience index data is embedded into multi-task neural network space, a high-dimensional feature vector is learned for each community, and classification is carried out using the feature vector.

Description

Technical Field

[0001] This invention belongs to the field of big data mining and application in social governance, and relates to a community resilience classification and evaluation method based on deep learning. Background Technology

[0002] Faced with technological innovation and the increasing intelligence of human life, the construction of smart communities is booming, and the form of communities has undergone fundamental changes. From traditional tribal communities that passively responded to disaster risks, they have evolved into basic urban units that meet the needs of survival and development, and are now transforming into future communities that achieve differentiated resident happiness and refined urban governance. New-generation information technologies, represented by big data and artificial intelligence, are increasingly being seen as effective means to solve urban grassroots governance problems.

[0003] Accurate assessment of community resilience is fundamental to building community resilience capabilities. Traditional resilience assessments often combine resilience-related theories and expert opinions to select indicators, establish a resilient community evaluation model based on hierarchical analysis, and then use methods such as entropy value method and expert scoring to determine the weight of each indicator. Finally, fuzzy comprehensive analysis is used to evaluate the various dimensions of resilient communities and classify community resilience levels.

[0004] Unlike traditional processes such as indicator system construction, hierarchical analysis, and expert scoring, the use of big data has opened up new directions for community resilience assessment. Summary of the Invention

[0005] This invention transforms the traditional community resilience evaluation model based on indicator weighting into a big data-based evaluation model, providing a community resilience assessment method based on a multi-task neural network. This invention's big data-based community resilience assessment utilizes deep learning to quantify various levels of community resilience indicators and inputs them into a neural network to extract community resilience features. It processes and analyzes various data required for community resilience calculation, achieving accurate and efficient assessment of community resilience. The technical solution is as follows:

[0006] A community resilience assessment method based on multi-task neural networks includes the following steps:

[0007] Step 1: Collect data and obtain community resilience reference index data;

[0008] Step 2 involves inputting the collected community resilience reference index data into the pre-designed community resilience attributes, which include five aspects: social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience.

[0009] (1) The community resilience attributes related to social resilience are as follows:

[0010]

[0011]

[0012] (2) The community resilience attributes related to economic resilience are as follows:

[0013]

[0014] (3) The community resilience attributes related to institutional resilience are as follows:

[0015]

[0016] (4) The community resilience attributes related to ecological resilience are as follows:

[0017]

[0018] (5) The community resilience attributes related to infrastructure resilience are as follows:

[0019]

[0020] The community resilience attributes are vectorized by: one-hot encoding the relevant index values ​​of secondary indicators that do not have data measurement, and normalizing the remaining index values; concatenating the processed reference index values ​​into a vector according to five basic dimensions, which is then used as the input to the deep neural network for community resilience evaluation.

[0021] Step 3, manual data annotation: Annotate some data through expert consultation. It is necessary to annotate the comprehensive resilience of the given community as well as the classification of social resilience, economic resilience, institutional resilience, ecological resilience and infrastructure resilience, that is, whether it is excellent, good or average.

[0022] Step 4: Divide the training set and the test set according to manual annotation;

[0023] Step 5 involves embedding the community resilience index data into a neural network space, learning a high-dimensional feature vector for each community, and using the feature vector for classification. The neural network is as follows:

[0024] The neural network consists of five layers. The first four layers have the same structure, and the fifth layer has an output dimension of 18. This output is divided into six parts: the first three dimensions represent the confidence level of the neural network in classifying the community's overall resilience as excellent, good, and average; the remaining outputs represent the confidence level of the neural network in classifying the community's resilience in each dimension. Specifically, the fourth to sixth dimensions represent the confidence level of the neural network in classifying the community's social resilience as excellent, good, and average; the seventh to ninth dimensions represent the confidence level of the neural network in classifying the community's economic resilience as excellent, good, and average, and so on.

[0025] Using a multi-task learning-based neural network, this study comprehensively assesses community resilience, including its social, economic, institutional, ecological, and infrastructure resilience. For the first four layers of the neural network, each layer includes the following data processing steps:

[0026] (1) For the kth training set sample community, input h to the corresponding layer k Community resilience features z were extracted using linear transformation. k ;

[0027] (2) Community resilience characteristics z k A nonlinear transformation is performed to learn the nonlinear relationship between community resilience index data and community resilience, thus obtaining the output h of this layer of the network. k,out ;

[0028] The fifth-layer neural network outputs various resilience assessment confidence scores for the community, including the following data processing steps:

[0029] (1) For the kth training set sample community, input h to the fifth layer. k Dimensionality reduction is performed using linear transformations to extract the community resilience feature z. k ;

[0030] (2) Community resilience characteristics z k Each three dimensions are grouped together, and the softmax activation function is used for classification;

[0031] Step 6, iterative optimization, training the neural network model, includes the following steps:

[0032] (1) Initialize neural network parameters W t ;

[0033] (2) Input the training set into the neural network and obtain the output result h. k,i ;

[0034] (3) Calculate the cross-entropy classification loss for multiple tasks.

[0035] (4) Calculate the overall loss of the model

[0036]

[0037] (5) Based on the overall loss of the model Execute the gradient descent algorithm;

[0038] (6) When the overall loss of the model Stop training when convergence occurs;

[0039] Step 7, Unlabeled Community Resilience Assessment: Input the unlabeled data into the neural network and obtain the network output. If the first dimension of the output has the largest value, the community is assessed as having excellent resilience. If the second dimension has the largest value, the community is assessed as having good resilience. If the third dimension has the largest value, the community is assessed as having average resilience.

[0040] The output dimensions 4-18 provide a more granular classification result, providing an assessment of the social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience of the evaluated community.

[0041] This invention is applicable to community resilience assessment tasks, providing a comprehensive evaluation of community resilience and separately assessing its social, economic, institutional, ecological, and infrastructure resilience. The method of this invention utilizes deep learning to better extract the nonlinear relationship between community resilience indicators and community resilience assessments, thereby improving the effectiveness of community resilience assessment.

[0042] This invention can be applied to:

[0043] (1) Growth analysis: Through resilience assessment analysis at different time points, the development and changes of a community after entering the assessment system can be clearly shown.

[0044] (2) Comparative analysis: By analyzing the community resilience scores of adjacent geographical locations at a certain time point, the average level of resilience and relative advantages and disadvantages can be obtained.

[0045] (3) Evaluation of resilience enhancement measures: For resilience enhancement measures implemented in the community, the effectiveness of the measures can be evaluated by analyzing the resilience score and evaluation level output by the multi-task neural network over a period of time. Attached Figure Description

[0046] Figure 1 This is a diagram of the method model.

[0047] Figure 2 Flowchart of implementation steps. Detailed Implementation

[0048] This invention provides a method for assessing community resilience. It utilizes a neural network to extract features from resilience indicators obtained from multiple sources, integrates various community resilience dimensions for model training, and finally categorizes community resilience into three classes: excellent, good, and average. The invention is described below.

[0049] 1) Data acquisition and processing

[0050] The first step is data collection. Five basic dimensions for evaluating community resilience were established through expert interviews and relevant literature review: social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience. Reference indicator data for community resilience were obtained through field surveys and interviews with community staff.

[0051] The second step is to input the collected community resilience reference index data into the pre-designed community resilience attributes, which mainly include five aspects: social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience.

[0052] (6) The community resilience attributes related to social resilience are as follows:

[0053]

[0054] (7) The community resilience attributes related to economic resilience are as follows:

[0055]

[0056] (8) The community resilience attributes related to institutional resilience are as follows:

[0057]

[0058] (9) The community resilience attributes related to ecological resilience are as follows:

[0059]

[0060]

[0061] (10) The community resilience attributes related to infrastructure resilience are as follows:

[0062]

[0063] The community resilience attributes are vectorized. Secondary indicators without data measurement, such as the presence of public enterprises and institutions, the presence of hazardous sources, whether gas pipelines have been updated in the past year, and whether water supply pipelines have been updated in recent years, are one-hot encoded. The remaining indicator values ​​are normalized. These processed reference indicator values ​​can be concatenated into a vector according to five basic dimensions, serving as input to a deep neural network for community resilience assessment.

[0064] The third step involves labeling 50% of the data through expert consultation, determining whether the overall resilience of the labeled communities is excellent, good, or average, and indicating whether social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience are excellent, good, or average.

[0065] The fourth step is to divide the training set and the test set according to manual annotation, with a ratio of 7:3.

[0066] 2) Construction of a multi-index evaluation neural network

[0067] The neural network consists of five layers, with output dimensions of 256, 1024, 256, 128, and 18 for each layer. The last layer has an output dimension of 18, which can be divided into six parts: dimensions 1-3 represent the confidence level of the neural network in classifying the community's overall resilience as excellent, good, and average; the remaining outputs represent the confidence levels of the neural network's assessments of the community's resilience in each dimension. Specifically, dimensions 4-6 represent the confidence levels of the neural network in classifying the community's social resilience as excellent, good, and average; dimensions 7-9 represent the confidence levels of the neural network in classifying the community's economic resilience as excellent, good, and average, and so on.

[0068] This method uses a multi-task learning-based neural network to comprehensively assess community resilience, including its social, economic, institutional, ecological, and infrastructure resilience. The first four layers of the neural network each include the following process:

[0069] The first step is to, for the k-th training set sample community, input h to this layer. k Community resilience features are extracted using a linear transformation, as shown in the following formula:

[0070] z k =W t h k

[0071] in, Let d be the weight matrix of the fully connected layer. in Let d be the input dimension for this layer. out This is the output dimension for this layer.

[0072] The second step involves performing a nonlinear transformation on the extracted features to learn the nonlinear relationship between community resilience indicators and community resilience, ultimately obtaining the output h of this layer of the network. k,out The specific formula is as follows:

[0073] h k,out =ReLU(z) k )

[0074] ReLU(z k ) = max(0, z k )

[0075] Among them, max() is the operation to retrieve the maximum value.

[0076] The fifth layer of the neural network needs to output the confidence scores of various resilience assessments of the community. The specific process is as follows:

[0077] The first step is to, for the k-th training set sample community, input h to this layer. k Dimensionality reduction is performed using linear transformations to extract the community resilience feature z. k The formula is the same as the first four layers.

[0078] The second step is to analyze the community resilience characteristics z. k Each dimension is grouped together, and the softmax activation function is used for classification, as shown in the following formula:

[0079] h k,i =softmax({z k,3*i-2 ,z k,3*i-1 ,z k,3*i})

[0080] Where i∈{1,2,3,4,5,6}, h k,i For the i-th output, z k,x For z k The xth dimension.

[0081] 3) Iterative optimization

[0082] The first step is to initialize the neural network parameters W. t .

[0083] The second step is to input the entire training set into the neural network and obtain the output result h. k,i .

[0084] The third step is to calculate the classification loss for multi-task tasks using the following formula.

[0085]

[0086] Among them, y k,i The label represents the label of the k-th data point in the i-th dimension of the training set, and N is the set of all data in the training set.

[0087] The fourth step is to calculate the overall loss of the model.

[0088]

[0089] Fifth step, based on the loss Execute the gradient descent algorithm.

[0090] Step 6, when losses Stop training when convergence occurs.

[0091] 4) Unlabeled community resilience assessment

[0092] The first step is to input unlabeled data into the neural network and obtain the network output.

[0093] The second step is to assess the community as having excellent resilience if the first dimension of the output is the largest, good resilience if the second dimension is the largest, and average resilience if the third dimension is the largest.

Claims

1. A community resilience assessment method based on multi-task neural networks, characterized in that, Includes the following steps: Step 1: Collect data and obtain community resilience reference indicators; Step 2 involves inputting the collected community resilience reference index data into the pre-designed community resilience attributes, which include five aspects: social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience. The community resilience attributes are vectorized by: one-hot encoding the relevant index values ​​of secondary indicators that do not have data measurement, and normalizing the remaining index values; concatenating the processed reference index values ​​into a vector according to five basic dimensions, which is then used as the input to the deep neural network for community resilience evaluation. Step 3, manual data annotation: Annotate some data through expert consultation. It is necessary to annotate the comprehensive resilience of the given community as well as the classification of social resilience, economic resilience, institutional resilience, ecological resilience and infrastructure resilience, that is, whether it is excellent, good or average. Step 4: Divide the training set and the test set according to manual annotation; Step 5 involves embedding community resilience reference index data into a neural network space to learn a high-dimensional feature vector for each community, and then using the feature vector for classification. The neural network is as follows: The neural network consists of five layers. The first four layers have the same structure, and the output dimension of the fifth layer is 18. The output is divided into 6 parts: the first to third dimensions represent the confidence level of the neural network in judging the overall resilience of the community as excellent, good, and average. The remaining outputs represent the confidence levels of the neural network's assessment of the community's resilience across various dimensions. Dimensions 4-6 represent the confidence levels of the neural network's assessment of the community's social resilience as excellent, good, and average; dimensions 7-9 represent the confidence levels of the neural network's assessment of the community's economic resilience as excellent, good, and average, and so on. Using a multi-task learning-based neural network, this study comprehensively assesses community resilience, including its social, economic, institutional, ecological, and infrastructure resilience. For the first four layers of the neural network, each layer includes the following data processing steps: For the A training set sample community, and its input at the corresponding layer. Extracting community resilience features using linear transformation : in, This is the weight matrix of the fully connected layer. Input dimension for this layer, This is the output dimension for this layer; Community resilience characteristics A nonlinear transformation is performed to learn the nonlinear relationship between community resilience reference index data and community resilience, thus obtaining the output of this layer of the network. : in, This is for retrieving the maximum value. The fifth-layer neural network outputs various resilience assessment confidence scores for the community, including the following data processing steps: (1) For the first A training set sample community, and its input in the fifth layer. Dimensionality reduction is performed using linear transformations to extract community resilience features. ; (2) Community resilience characteristics Each three dimensions are grouped together, using Classification of activation functions: in , For the first i Group output, for The dimension; Step 6: Iterative optimization and training of the neural network model, as follows: (1) Initialize neural network parameters ; (2) Input the training set into the neural network and obtain the output results. ; (3) Calculate the cross-entropy classification loss for multiple tasks. : in, Representative training set number The data in the first Labels in each dimension N It is the collection of the entire training set; (4) Calculate the overall loss of the model : (5) Based on the overall loss of the model Execute the gradient descent algorithm; (6) When the overall model loss Stop training when convergence occurs; Step 7, Unlabeled Community Resilience Assessment: Input the unlabeled data into the neural network and obtain the network output. If the first dimension of the output has the largest value, the community is assessed as having excellent resilience. If the second dimension has the largest value, the community is assessed as having good resilience. If the third dimension has the largest value, the community is assessed as having average resilience. The output dimensions 4-18 provide a more granular classification result, providing an assessment of the social resilience, economic resilience, institutional resilience, ecological resilience, and infrastructure resilience of the evaluated community.

2. The community resilience assessment method based on multi-task neural networks according to claim 1, characterized in that, In step 2, the community resilience attributes related to social resilience are as follows: 。 3. The community resilience assessment method based on multi-task neural networks according to claim 1, characterized in that, In step 2, the community resilience attributes related to economic resilience are as follows: 。 4. The community resilience assessment method based on multi-task neural networks according to claim 1, characterized in that, In step 2, the community resilience attributes related to institutional resilience are as follows: 。 5. The community resilience assessment method based on multi-task neural networks according to claim 1, characterized in that, In step 2, the community resilience attributes related to ecological resilience are as follows: 。 6. The community resilience assessment method based on multi-task neural networks according to claim 1, characterized in that, In step 2, the community resilience attributes related to infrastructure resilience are as follows: 。

Images