Dynamic evaluation method for extreme rainstorm waterlogging disaster risk for disaster prevention and reduction

Abstract

A dynamic evaluation method for an extreme rainstorm waterlogging disaster risk for disaster prevention and reduction. The method comprises: investigating and surveying urban system data and disaster prevention and reduction data, using GIS technology to divide disaster-bearing objects into refined risk units on the scale of urban buildings and road networks, and determining the spatial distribution of the disaster-bearing objects; on the basis of an extreme rainstorm waterlogging scene, simulating the disaster influence of a dynamic change process of a flood ponding depth on the disaster-bearing objects; developing refined dynamic evaluation on a waterlogging risk by combining the two methods of waterlogging process simulation and an indicator system; using a spatial complex network and a shortest path plan to calculate a traffic capacity and emergency service accessibility of a road network system; and on this basis, taking into comprehensive consideration the rational allocation of disaster prevention emergency drainage and emergency rescue services to a high-risk area, and proposing dynamic evaluation technology for a waterlogging risk that integrates a disaster evolution process and a disaster prevention response process, and ultimately realizing the dynamic evaluation of the waterlogging risk of each disaster-bearing unit during the waterlogging disaster evolution.

Description

Dynamic assessment method of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation Technical Field

[0001] The present invention belongs to the technical field of urban disaster prevention and reduction, and relates to a dynamic assessment method for extreme rainstorm waterlogging disaster risk for disaster prevention and reduction. Background Art

[0002] Due to the combined impacts of global climate change and rapid urbanization, urban flooding caused by extreme rainstorms is becoming an increasingly prominent problem. Urban flooding causes widespread waterlogging, damaging infrastructure and paralyzing transportation systems. Hindered emergency services struggle to meet the rapidly increasing demand for emergency rescue, resulting in significant economic losses and social impacts, even threatening public safety. Risk assessment, as the foundation and technical support for urban flood disaster management and disaster prevention, mitigation, and relief decision-making, is shifting from hazard assessment to assessment of hazard-prone entities in a changing environment. Flooding of buildings and roads, casualties, and property losses have become crucial components of urban flood risk assessment. However, rainstorm flooding processes often exhibit significant spatiotemporal variations. Existing risk assessment methods suffer from unclear hazard classification and a strong reliance on subjective risk assessment. In particular, they overlook the dynamic impact of disaster prevention responses on urban flood risk, creating significant challenges for urban disaster prevention and mitigation efforts. Therefore, new technologies are urgently needed to accurately assess the dynamic risk changes of hazard-prone entities during the evolution of urban flooding.

[0003] Based on the above background, in response to the difficulties faced in urban disaster prevention and mitigation, the present invention can improve the current rainstorm waterlogging risk assessment method and enhance the urban waterlogging disaster management capabilities. The research results can provide technical support for urban disaster management and emergency decision-making. Summary of the Invention

[0004] The purpose of the present invention is to provide a dynamic assessment method for extreme rainstorm waterlogging disaster risks for disaster prevention and mitigation, which solves the problem that existing waterlogging disaster risk assessment methods have strong subjective dependence in risk evaluation in practical applications and cannot dynamically assess the waterlogging risks of each disaster-bearing unit during the evolution of waterlogging disasters.

[0005] The technical solution adopted by the present invention is a dynamic assessment method for extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation. It investigates urban system data and disaster prevention and mitigation data, uses GIS technology to finely divide disaster-prone objects into risk units at the scale of urban buildings and road networks, and determines the spatial distribution of disaster-prone objects; simulates the impact of the dynamic change process of flood water depth on disaster-prone objects based on extreme rainstorm waterlogging scenarios; regards the urban complex system as consisting of three subsystems: regional space, road network system, and disaster prevention and emergency response, and uses spatial complex networks and shortest path planning to calculate the traffic capacity of the road network system and the accessibility of emergency services; on this basis, comprehensively considers the rational allocation of disaster prevention emergency drainage and emergency rescue services to high-risk areas, combines the two methods of waterlogging process simulation and indicator system, and proposes a dynamic waterlogging risk assessment technology that integrates disaster evolution and disaster prevention and response processes. It dynamically assesses the waterlogging risk of each disaster-prone unit during the evolution of waterlogging disasters, constructs a risk database through scenario setting, optimizes the LSTM network by integrating the Harris Eagle optimization algorithm and the particle swarm optimization algorithm, and constructs a real-time waterlogging risk analysis model, ultimately realizing dynamic assessment and calculation of extreme rainstorm waterlogging disaster risk. The specific steps are as follows:

[0006] Step 1: Collect urban system data and urban disaster prevention and mitigation data; the urban system data includes road network system data, population density, land use, regional GDP, and urban lifeline engineering lines; the urban disaster prevention and mitigation data includes the number and location of flood control and drainage projects, and emergency flood control equipment, including the number and location of flood control and drainage vehicles and portable mobile pumps;

[0007] Step 2: Categorize key urban flood prevention targets and vulnerable areas into categories, including: critical infrastructure (main roads, underpasses, urban subways, substations, etc.), public service facilities (government offices, hospitals, schools, fire stations, etc.), residential areas (old urban areas, high-density areas, etc.), economic functional areas (commercial centers, tourist attractions, etc.), and vulnerable industries (bookstores, car dealerships, tobacco shops, furniture stores, etc.). Using building units and road networks as the scales, GIS technology is used to refine the risk unit division of hazard-prone objects and determine their spatial distribution.

[0008] Step 3: Build a high-precision flooding process model. Collect high-resolution terrain elevation data for key areas with high risk (underpasses, urban subways, substations, and old urban areas) to conduct detailed simulations of flood processes and analyze the impact of dynamic changes in flood water depth on disaster-prone structures.

[0009] Step 4: Classify the disaster level into three levels: slight damage, severe damage, and major damage. Calculate the economic losses of the disaster-affected body by comprehensively considering the inundation depth, area, and average economic value.

[0010] Step 5: Calculate the traffic capacity of the road network and evaluate road failure;

[0011] Step 6: Evaluate the accessibility of rescue services in combination with regional waterlogging emergency prevention and control strategies;

[0012] Step 7: Set a rainstorm flooding scenario and dynamically assess the flooding risk;

[0013] Step 8: Building a waterlogging risk database based on the waterlogging risk dynamic change assessment results of step 7;

[0014] Step 9: Use the waterlogging risk database from step 8 to train the LSTM deep learning network. Use the Harris Eagle optimization algorithm combined with the particle swarm optimization algorithm to optimize the number of hidden layer neurons, weights, and learning rate of the LSTM network to improve the prediction accuracy of the LSTM network and build a real-time waterlogging risk analysis model.

[0015] Step 10, based on the real-time, forecasted or set rainfall conditions and disaster prevention levels, adjust the scenario conditions, use the real-time waterlogging risk analysis model engine to perform real-time calculations, and obtain the dynamic process and deduction results of the spatiotemporal changes of waterlogging risks.

[0016] The present invention is also characterized in that:

[0017] Step 4: The method for calculating the economic losses of the disaster-affected body is shown in formula (1):

[0018] Where L is the economic loss of the hazard-bearing body, I and J represent the total number of hazard-bearing body types and the total number of flooding depths, respectively; A ij is the inundation area corresponding to the jth inundation depth of the i-th hazard-bearing body type, D ij A ij Corresponding disaster level; Disaster level D ij Medium: slightly affected is 0.2, severely affected is 0.3, and severely affected is 0.5; i is the average economic value of the i-th disaster-prone unit type.

[0019] Step 5 is as follows:

[0020] The roads with service functions in the urban area are divided into main roads and secondary roads, and the road intersections are abstracted into nodes using spatial complex networks. p ={v1,v2,…v p}, the road segments connected by the intersection are abstracted as a set of edges, E p ={e1,e2,…e p}, forming a road network graph G(V p ,E p );Adjacent intersection v i v jThe required travel time T is used as the weight to construct the weight matrix Finally, the road network topology M(G) is formed. During the evolution of urban waterlogging disasters, if the water depth is greater than 50 cm, the road functionality will fail, resulting in changes in the road network topology. Based on the idea of ​​the shortest path, the regional road network traffic capacity is quantitatively calculated, as shown in formula (2):

[0021] Where BC t (a) represents the traffic capacity of the road network and is the betweenness of node a at time t; n t is the total number of shortest paths connecting nodes b and c at time t, where a, b, and c represent arbitrary nodes; n t (a) is the total number of shortest paths connecting node a at time t, where the shortest path is calculated using the artificial bee colony optimization algorithm;

[0022] If the depth of water on the road is greater than 50cm, the road is considered to be inoperable.

[0023] Step 6 is as follows:

[0024] Taking the facility unit that provides emergency services as the center, the radiation range of emergency facilities is calculated according to the current status of the road network topology structure and the response arrival time threshold in different scenarios of waterlogging. Within the radiation range, the emergency response demand of the disaster-bearing bodies in high-risk areas is statistically analyzed, and the supply-demand ratio of emergency services is calculated, as shown in formula (3):

[0025] Where R j is the supply-demand ratio of emergency services; S j Represents the spatial area (d ij ,d0) service radiation range of emergency service facilities; N i is the number of high-risk units in the spatial area; G(d ij ,d0) is the Gaussian equation considering the effect of distance on the efficiency of emergency services;

[0026] Taking the flood risk unit k as the center, search for rescue points within a given spatial distance threshold d0, and calculate the emergency service supply and demand ratio R of all risk units. j Gaussian equation is used for weighting, and weighted summation is used to obtain the overall emergency service accessibility A k , as shown in formula (4):

[0027] Step 7 is as follows:

[0028] A refined assessment of urban flooding risk is carried out based on regional space, and an assessment index system considering hazard, exposure, vulnerability and disaster prevention and mitigation capabilities is constructed using a combination of the analytic hierarchy process and the entropy weight method. Scenario analysis is conducted in combination with the urban flooding process model, and the dynamic changes in urban flooding risk in disaster-bearing units are calculated by considering emergency drainage for road disaster prevention, emergency dispatch of flood control and drainage projects, and accessibility of emergency rescue services at different stages of the urban flooding process.

[0029] The specific steps for constructing the dynamic assessment of waterlogging risk in step 7 are as follows:

[0030] Step 7.1: Build an evaluation indicator system

[0031] The evaluation index system is decomposed into the following 13 indicators: rainfall intensity, rainfall duration, flooding depth, flooding duration, flooding range, surface elevation, land use, population density, economic losses, road network traffic capacity, monitoring and early warning capabilities, flood control and drainage capabilities, and accessibility of emergency rescue services. The index weights are calculated using the combination of hierarchical analysis method and entropy weight method. The economic losses of buildings involved are calculated using formula (1) described in step 4.

[0032] a. The specific method of calculating the index weights by combining the analytic hierarchy process and the entropy weight method is as follows:

[0033] (1) Constructing a judgment matrix: Based on expert opinions or experience, use the 1-9 scale method to compare each indicator pairwise to generate a judgment matrix;

[0034] (2) Consistency test: Calculate the maximum eigenvalue of the judgment matrix and use the consistency ratio CR to determine whether the matrix meets the consistency requirements. When CR < 0.1, the matrix passes the consistency test.

[0035] (3) Calculate the subjective weight W j AHP : By solving the eigenvector of the judgment matrix and normalizing it, the subjective weight of each indicator is obtained;

[0036] b. Calculate objective weights using entropy weight method

[0037] (1) Standardized data: All evaluation indicators are dimensionlessly standardized, such as minimum-maximum standardization, to ensure that the dimensions of each indicator are consistent.

[0038] (2) Calculate information entropy: Calculate the information entropy value of each indicator based on the standardized value of each indicator. Information entropy reflects the degree of dispersion of the indicator. The greater the dispersion, the smaller the information entropy.

[0039] The information entropy formula is:

[0040] where p ijis the ratio of the standardized value of the i-th sample of the j-th indicator to the total, k′ is a constant, k′=1 / ln(n), n is the total number of samples;

[0041] The objective weight is thus obtained as:

[0042] in, is the objective weight of the jth indicator; s is the total number of evaluation indicators;

[0043] c. Use the weighted average method to combine the subjective weight and the objective weight calculated by the entropy weight method,

[0044] Among them, α is the subjective weight ratio.

[0045] Step 7.2: Dynamic calculation of the evolution process of extreme rainstorm waterlogging disaster:

[0046] Based on the evolution of urban waterlogging, the dynamic evolution of urban waterlogging disaster scenarios is divided into five stages: initial scenario, evolution scenario, peak scenario, recovery scenario, and termination scenario. A waterlogging process model is used to simulate and calculate the inundation depth, inundation duration, and inundation range required by the indicator system in step 7.1. The simulation considers three factors that contribute to reducing urban waterlogging: temporary drainage using portable road pumps, the rational allocation of emergency flood control and drainage vehicles to high-risk areas, and the emergency dispatch model of flood control and drainage projects. The rational allocation of emergency flood control and drainage vehicles is determined through the following three steps:

[0047] (1) Ranking of high-risk road sections during waterlogging disasters: The water depth on the road is greater than 50 cm and the duration is greater than 15 minutes. The need for flood emergency drainage vehicles to carry out drainage operations is confirmed. High-risk road sections are ranked in the following order: key protection area > vulnerable industry > main road > time of occurrence of high-risk road sections.

[0048] The scope of key protection areas, vulnerable industries, main roads, and high-risk sections can be obtained through observation of satellite image maps. The scope of key protection areas includes key protection objects and surrounding roads;

[0049] The occurrence time of high-risk sections is calculated using a two-dimensional hydrodynamic model simulation;

[0050] (2) Flood control and drainage vehicle route planning: Based on the road conditions under the waterlogging scenario, an artificial bee colony optimization algorithm is used to calculate the shortest path for flood control and drainage vehicles to reach the operation site and calculate the travel time. The new path needs to be re-planned and calculated after each drainage operation is completed;

[0051] (3) Calculation of drainage by flood control drainage vehicles: Based on the peak water volume and the drainage volume of the flood control drainage vehicle, the drainage time is calculated. When the water depth in the area is less than 5 cm, the drainage operation is considered completed;

[0052] In step 7.3, the road network traffic capacity and emergency rescue service accessibility are calculated at each stage of the waterlogging disaster evolution process in step 7.2. The road network traffic capacity is calculated using the method in step 5, and the rescue service accessibility is calculated using formulas (3) and (4) in step 6. Taking into account the risk impact of emergency rescue service accessibility, the waterlogging risk of different hazard-bearing bodies in each scenario stage is calculated:

[0053] According to the combined weights, the comprehensive flood risk score of each region or sample is calculated. Let the combined weight value of the i-th sample under the j-th indicator be x ij , then the comprehensive risk score R i for:

[0054] Step 7.4, based on the standard deviation classification method, the mean and standard deviation of the risk score are used as the basis for classification. The mean μ and standard deviation σ of the comprehensive risk score are calculated, and the risk level is divided according to the standard deviation range of the comprehensive risk score above and below the mean:

[0055] Low risk: score is less than μ-σ;

[0056] Medium risk: score between μ-σ and μ,

[0057] Higher risk: scores between μ and μ+σ,

[0058] High risk: score greater than μ+σ;

[0059] On the basis of risk classification, experts are invited to optimize the risk level standards according to actual conditions and finally obtain the dynamic assessment results of urban waterlogging risk.

[0060] The specific steps for optimizing the LSTM network parameters in step 9 are as follows:

[0061] Step 9.1: Taking the LSTM network prediction accuracy as the optimization target, construct the objective function, which is as shown in formula (5):

[0062] Where y i is the i-th real data; is the average value of the real data; is the data prediction value; m is the number of data;

[0063] Step 9.2: In the early stages of network parameter optimization iterations, the Harris Hawk optimization operator is integrated to improve global search capabilities. The improved energy equation of the prey in the transition phase is used to determine whether to enter the local search phase. The particle swarm optimization algorithm is used to accelerate convergence, and the mixed Cauchy-Gaussian mutation operator is introduced to avoid possible premature convergence. The details are as follows:

[0064] (1) In the early stage of the algorithm, the Harris Eagle algorithm is used for global search. The search process is as shown in formula (6):

[0065] Where, X t+1 and X t are the positions of the individuals in the population at the t+1th iteration and the tth iteration respectively; X rand,t is the random position of the population at the tth iteration; X m,t is the average position of the population at the tth iteration; X prey,t is the prey position at the tth iteration; r1, r2, r3, r4 are random numbers between 0 and 1; ub and lb are the upper and lower bounds of the search space, respectively; q is a random number between (0, 1);

[0066] (2) Using the improved energy equation of prey in the transition phase, determine whether the algorithm has entered the local search phase; the formula is as follows:

[0067] Where E is the energy of the prey to escape. When E≥1, the algorithm performs a global search, otherwise it performs a local search. E0 is the initial energy of the prey, E0=2×rand-1, rand is a random number between 0 and 1. max is the maximum number of iterations;

[0068] (3) In the local search phase, the fast search capability of the particle swarm algorithm is used to accelerate convergence. The search process is as shown in formula (8)-formula (9): V t+1 =ω t V t +c1r1(pbest t -X t )+c2r2(gbest t -X t ) (8) X t+1 =X t +V t+1 (9)

[0069] Where, X t+1 and V t+1 are the position and velocity of the particle in generation t+1 respectively; ω t is the weight; c1 and c2 are the individual learning and social learning factors of the particle respectively; r1 and r2 are random numbers between 0 and 1; pbestt is the individual optimal solution of the particle; gbest t is the global optimal solution for all particles in generation t;

[0070] In the later stage of the algorithm, the Cauchy-Gauss mutation strategy is introduced to improve the global search ability of the algorithm. If the optimal solution of the population does not change after 10 iterations, the current best individual is selected for mutation, as shown in formula (10): X newt =gbest t ×(1+ζ1Cauchy(0,1)+ζ2Gauss(0,1)) (10)

[0071] Where, X newt is the individual position after mutation; Cauchy(0,1) and Gauss(0,1) are random factors obeying Cauchy distribution and Gaussian distribution respectively; ζ1 and ζ2 are adaptive parameters of Cauchy distribution and Gaussian distribution respectively.

[0072] The beneficial effects of the present invention are:

[0073] The present invention provides a dynamic assessment method for extreme rainstorm waterlogging risk for disaster prevention and mitigation. This method has demonstrated excellent results in practical urban smart stormwater management systems. The dynamic waterlogging risk assessment results are derived from a high-precision waterlogging process model combined with an indicator system. In particular, the method performs a refined simulation of the risk of hazard-bearing entities, resulting in both accuracy and comprehensiveness. Furthermore, the method considers the dynamic changes in extreme rainstorm waterlogging processes and the city's disaster prevention and mitigation capabilities, resulting in dynamic risk results that serve disaster prevention and rescue efforts as waterlogging disasters evolve. Furthermore, through machine learning technology, risk results are rapidly predicted, effectively ensuring the timeliness of disaster prevention and rescue efforts. This present invention provides a precise, dynamic, and real-time waterlogging risk assessment method. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] FIG1 is an overall flow chart of a method for dynamically assessing the risk of extreme rainstorm waterlogging disasters for disaster prevention and mitigation according to the present invention;

[0075] FIG2 is a diagram showing the dynamic changes in waterlogging risk of an extreme rainstorm waterlogging disaster risk dynamic assessment method for disaster prevention and mitigation according to the present invention;

[0076] FIG3 is a path planning diagram of disaster prevention and emergency rescue services for the method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to the present invention;

[0077] FIG4 is a diagram showing the dynamic evolution of a waterlogging disaster scenario in the method for dynamic risk assessment of extreme rainstorm waterlogging disasters for disaster prevention and mitigation according to the present invention;

[0078] FIG5 is an optimization flow chart of the method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to the present invention, which integrates the Harris Hawk optimization algorithm and the particle swarm optimization algorithm. DETAILED DESCRIPTION

[0079] The present invention is described in detail below in conjunction with the accompanying drawings and embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process. However, the protection scope of the present invention is not limited to the following embodiments.

[0080] Example 1

[0081] The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation of the present invention is shown in FIG1 and is specifically as follows:

[0082] Step 1: Collect urban system data and urban disaster prevention and mitigation data; the urban system data includes road network system data, population density, land use, regional GDP, and urban lifeline engineering lines; the urban disaster prevention and mitigation data includes the number and location of flood control and drainage projects, and emergency flood control equipment, including the number and location of flood control and drainage vehicles and portable mobile pumps;

[0083] Step 2: Categorize key urban flood prevention targets and vulnerable areas as disaster-prone objects. These disaster-prone objects are categorized into: important infrastructure, public service facilities, residential areas, economic functional areas, and vulnerable industries. Using building units and road networks as the scales, GIS technology is used to refine the risk unit division and determine the spatial distribution of the disaster-prone objects.

[0084] Step 3: Build a high-precision flooding process model. Collect high-resolution terrain elevation data in key areas with high risk, conduct detailed simulations of flooding processes, and analyze the impact of dynamic changes in flood water depth on disaster-prone areas.

[0085] Step 4: Classify the disaster level into three levels: slight damage, severe damage, and major damage. Calculate the economic losses of the disaster-affected body by comprehensively considering the inundation depth, area, and average economic value.

[0086] Minor damage includes people's movement being restricted, cars being unable to start, roads being flooded, and houses being flooded;

[0087] Severe damage included people losing their balance, vehicles losing their balance, roads flooded and traffic interrupted, and houses submerged;

[0088] Major damage included people drowning, cars drowning, roads flooding, and houses damaged;

[0089] Step 5: Calculate the traffic capacity of the road network and the road failure rate;

[0090] Step 6: Evaluate the accessibility of rescue services in combination with regional waterlogging emergency prevention and control strategies;

[0091] Step 7: Set a rainstorm flooding scenario and dynamically assess the flooding risk;

[0092] Step 8: Building a waterlogging risk database based on the waterlogging risk dynamic change assessment results of step 7;

[0093] Step 9: Use the waterlogging risk database from step 8 to train the LSTM deep learning network. Use the Harris Eagle optimization algorithm combined with the particle swarm optimization algorithm to optimize the number of hidden layer neurons, weights, and learning rate of the LSTM network to improve the prediction accuracy of the LSTM network and build a real-time waterlogging risk analysis model.

[0094] Step 10, based on the real-time, forecasted or set rainfall conditions and disaster prevention levels, adjust the scenario conditions, use the real-time waterlogging risk analysis model engine to perform real-time calculations, and obtain the dynamic process and deduction results of the spatiotemporal changes of waterlogging risks.

[0095] Example 2

[0096] Step 1: Collect urban system data and urban disaster prevention and mitigation data. Collect urban system data, including road network data, population density, land use, regional GDP, and urban vital engineering lines. Collect urban disaster prevention and mitigation data, including the number and location of flood control and drainage projects, and the number and location of emergency flood control equipment, such as flood control and drainage vehicles and portable mobile pumps.

[0097] Step 2: Determine the key urban flood prevention targets and vulnerable areas, which are considered as hazard-prone areas, according to the following categories: critical infrastructure (main roads, underpasses, urban subways, substations, etc.), public service facilities (government offices, hospitals, schools, fire stations, etc.), residential areas (old urban areas, high-density areas, etc.), economic functional areas (commercial centers, tourist attractions, etc.), and vulnerable industries (bookstores, car dealerships, tobacco shops, furniture stores, etc.). Using GIS technology, the hazard-prone areas were refined and delineated using building units and road networks to clarify their spatial distribution.

[0098] Step 3: Build a high-precision model of the urban flooding process and collect high-resolution terrain elevation data for the key areas with greater risks in Step 1. Key areas with greater risks include underpasses, urban subways, substations, and old urban areas. Conduct detailed simulations of the flood process and analyze the impact of the dynamic changes in the depth of flood water on disaster-prone objects (disaster-prone objects include people, vehicles, roads, houses, etc.).

[0099] Step 4: Classify and discuss the most unfavorable principle of flood disasters on each disaster-bearing body in Steps 2 and 3, and divide the disaster level into three levels: slight disaster, severe disaster, and major disaster. Considering the inundation depth, area, and average economic value, calculate the economic loss of the disaster-bearing body, see formula (1):

[0100] Where I and J represent the total number of hazard-bearing body types and total inundation depths, respectively; A ij is the inundation area corresponding to the jth inundation depth of the i-th hazard-bearing body type, D ij A ij The corresponding disaster level is 0.2 for minor disaster, 0.3 for severe disaster and 0.5 for major disaster. i is the average economic value of the i-th disaster-prone unit type.

[0101] Step 5: Calculate the traffic capacity of the road network and evaluate the road failure. The roads with service functions in the urban area are mainly divided into main roads and secondary roads. The road intersections are abstracted into nodes using the spatial complex network. p ={v1,v2,…v p}, the road segments connected by the intersection are abstracted as a set of edges, E p ={e1,e2,…e p}, forming a road network graph G(V p ,E p ). Take the adjacent intersection v i v j The required travel time T is used as the weight to construct the weight matrix Ultimately, the road network topology M(G) is formed. During the evolution of urban flooding, if the water depth exceeds 50 cm, the road functionality will fail, leading to changes in the road network topology. Based on the shortest path concept, the regional road network traffic capacity is quantitatively calculated, as shown in Formula (2).

[0102] Where BC t (a) represents the traffic capacity of the road network and is the betweenness of node a at time t; n t is the total number of shortest paths connecting nodes b and c at time t, where a, b, and c represent arbitrary nodes; n t (a) is the total number of shortest paths connecting node a at time t, where the shortest path is calculated using the artificial bee colony optimization algorithm;

[0103] BC t (a) represents the ratio of the number of paths passing through node a in all the shortest paths to the total number of shortest paths, and the total number of betweenness centrality of all nodes represents the traffic capacity;

[0104] If the depth of water on the road is greater than 50cm, the road is considered to be inoperable.

[0105] Step 6: Evaluate the accessibility of rescue services in conjunction with regional waterlogging emergency prevention and control strategies. Centered on the facility unit providing emergency services, calculate the radiation range of emergency facilities based on the current status of the road network topology and response arrival time thresholds under different waterlogging scenarios. Statistically analyze the emergency response needs of disaster-stricken areas in high-risk areas within the radiation range, and calculate the emergency service supply-demand ratio. The calculation method for the emergency service supply-demand ratio is as follows:

[0106] Where R j is the supply-demand ratio of emergency services; S j Represents the spatial area (d ij ,d0) service radiation range of emergency service facilities; N i is the number of high-risk units in the spatial area; G(d ij ,d0) is the Gaussian equation considering the effect of distance on the efficiency of emergency services;

[0107] Taking the flood risk unit k as the center, search for rescue points within a given spatial distance threshold d0, and calculate the emergency service supply and demand ratio R of all risk units. j Gaussian equation is used for weighting, and weighted summation is used to obtain the overall emergency service accessibility A k , the emergency service accessibility is calculated as formula (4):

[0108] Step 7: Dynamically assess urban flooding risk. Leveraging the combined strengths of the indicator system approach and model simulation, a refined assessment of urban flooding risk is conducted on a regional scale. A combined analytic hierarchy process (AHP) and entropy weighting method are used to construct an assessment indicator system that considers hazard, exposure, vulnerability, and disaster prevention and mitigation capabilities. Scenario analysis is conducted in conjunction with the urban flooding process model, taking into account newly increasing or decreasing risk factors such as emergency drainage for road disaster prevention, emergency dispatch of flood control and drainage projects, and accessibility of emergency rescue services at different stages of the urban flooding process. Dynamic changes in urban flooding risk within the affected unit are calculated.

[0109] Step 8: Building a waterlogging risk database based on the waterlogging risk dynamic change assessment results of step 7;

[0110] In step 9, the database in step 8 is used to train the LSTM deep learning network. The number of hidden layer neurons, weights, and learning rate of the LSTM network are optimized by integrating the Harris Eagle optimization algorithm and the particle swarm optimization algorithm to improve the prediction accuracy of the LSTM network and build a real-time analysis model for urban flooding risk.

[0111] Step 10, based on the real-time, forecasted or set rainfall conditions and disaster prevention levels, adjust the scenario conditions, use the real-time waterlogging risk analysis model engine to perform real-time calculations, and obtain the dynamic process and deduction results of the spatiotemporal changes of waterlogging risks.

[0112] Example 3

[0113] Based on Example 2,

[0114] The specific steps for constructing the dynamic assessment of waterlogging risk in step 7 are as follows:

[0115] Step 7.1: Build an evaluation indicator system

[0116] The evaluation index system is decomposed into the following 13 indicators: rainfall intensity, rainfall duration, flooding depth, flooding duration, flooding range, surface elevation, land use, population density, economic losses, road network traffic capacity, monitoring and early warning capabilities, flood control and drainage capabilities, and accessibility of emergency rescue services. The index weights are calculated using the combination of hierarchical analysis method and entropy weight method. The economic losses of buildings involved are calculated using formula (1) described in step 4.

[0117] a. The specific method of calculating the index weights by combining the analytic hierarchy process and the entropy weight method is as follows:

[0118] (1) Constructing a judgment matrix: Based on expert opinions or experience, use the 1-9 scale method to compare each indicator pairwise to generate a judgment matrix;

[0119] (2) Consistency test: Calculate the maximum eigenvalue of the judgment matrix and use the consistency ratio CR to determine whether the matrix meets the consistency requirements. When CR < 0.1, the matrix passes the consistency test.

[0120] (3) Calculate the subjective weight W j AHP : By solving the eigenvector of the judgment matrix and normalizing it, the subjective weight of each indicator is obtained;

[0121] b. Calculate objective weights using entropy weight method

[0122] (1) Standardized data: All evaluation indicators are dimensionlessly standardized, such as minimum-maximum standardization, to ensure that the dimensions of each indicator are consistent.

[0123] (2) Calculate information entropy: Calculate the information entropy value of each indicator based on the standardized value of each indicator. Information entropy reflects the degree of dispersion of the indicator. The greater the dispersion, the smaller the information entropy.

[0124] The information entropy formula is:

[0125] Among them, p ij is the ratio of the standardized value of the i-th sample of the j-th indicator to the total, k′ is a constant, k′=1 / ln(n), n is the total number of samples;

[0126] The objective weight is thus obtained as:

[0127] in, is the objective weight of the jth indicator; s is the total number of evaluation indicators;

[0128] c. Use the weighted average method to combine the subjective weight and the objective weight calculated by the entropy weight method,

[0129] Among them, α is the subjective weight ratio.

[0130] Step 7.2, dynamic calculation of the evolution process of urban flooding disasters caused by extreme rainstorms. According to the evolution process of urban flooding, the dynamic evolution of the urban flooding disaster scenario is divided into five stages: initial scenario, evolution scenario, peak scenario, recovery scenario and termination scenario (as shown in Figures 2 and 4). The circle represents the severity of waterlogging, the colorless part means there is no waterlogging, and the color in the circle changes from light to dark, indicating that the waterlogging is getting more and more serious. The urban flooding process model is used for simulation calculation to calculate the flooding depth, flooding duration and flooding range required by the indicator system in step 7.1. During the simulation, three factors that are conducive to reducing urban flooding are considered: temporary drainage by portable mobile pumps on roads, rational resource allocation of emergency flood control and drainage vehicles to high-risk areas, and emergency dispatch mode of flood control and drainage projects. The rational allocation of emergency flood control and drainage vehicles is determined through the following three steps:

[0131] (1) Ranking of high-risk sections during waterlogging disasters. The need for flood emergency drainage vehicles to perform drainage operations is determined based on the criteria of water accumulation depth greater than 50 cm and duration greater than 15 minutes. High-risk sections are ranked in the following order: key protection area > vulnerable industry > main roads > time of occurrence of high-risk sections;

[0132] The scope of key protection areas, vulnerable industries, main roads, and high-risk sections can be obtained through observation of satellite image maps. The scope of key protection areas includes key protection objects and surrounding roads;

[0133] The occurrence time of high-risk sections is calculated using a two-dimensional hydrodynamic model simulation;

[0134] (2) Flood control and drainage vehicle route planning. Based on the road conditions under the flooding scenario, an artificial bee colony optimization algorithm is used to calculate the shortest path for the flood control and drainage vehicle to reach the operation site and calculate the travel time. The new path needs to be re-planned and calculated after each drainage operation is completed.

[0135] (3) Calculation of drainage by flood control drainage vehicles. Calculate drainage time based on the peak water volume and the drainage capacity of the flood control drainage vehicle. Drainage is considered complete when the water depth in the area is less than 5 cm.

[0136] In step 7.3, the road network capacity and accessibility of emergency rescue services are calculated at each stage of the flooding disaster evolution process, as described in step 7.2. Figure 3 shows the route planning diagram for disaster prevention and emergency rescue services. The boxes in the figure represent different risk areas. The road network capacity is calculated using the method in step 5, and the accessibility of rescue services is calculated using the method in step 6. Taking into account the risk impact of emergency rescue service accessibility, the flooding risk of different hazard-prone bodies is calculated at each stage.

[0137] According to the combined weights, the comprehensive flood risk score of each region or sample is calculated. Let the combined weight value of the i-th sample under the j-th indicator be x ij , then the comprehensive risk score R i for:

[0138] Step 7.4, based on the standard deviation classification method, the mean and standard deviation of the risk score are used as the basis for classification. The mean μ and standard deviation σ of the comprehensive risk score are calculated, and the level is divided according to the standard deviation range of the score above and below the mean:

[0139] Low risk: score is less than μ-σ,

[0140] Medium risk: score between μ-σ and μ,

[0141] Higher risk: scores between μ and μ+σ,

[0142] High risk: score greater than μ+σ;

[0143] On the basis of risk classification, experts are invited to optimize the risk level standards according to actual conditions and finally obtain the dynamic assessment results of urban waterlogging risk.

[0144] The specific steps for optimizing the LSTM network parameters in step 9 are as follows:

[0145] Step 9.1: Taking the LSTM network prediction accuracy as the optimization target, construct the objective function, which is formula (5):

[0146] Where y i is the i-th real data; is the average value of the real data; is the predicted value of the data; m is the number of data.

[0147] Step 9.2, build an optimization framework that integrates the Harris Hawk optimization algorithm and the particle swarm optimization algorithm. The particle swarm optimization algorithm is suitable for solving nonlinear optimization problems and has a fast convergence speed, but it is limited by insufficient global search capabilities and is prone to premature maturity during the evolution process. The Harris Hawk optimization algorithm has a strong search capability in the global exploration stage. Therefore, the algorithm integrates the Harris Hawk optimization operator in the early stage of iteration to improve the global search capability, uses the improved energy equation of the prey in the transition stage to determine whether to enter the local search stage, uses the particle swarm algorithm to accelerate convergence, and then introduces the mixed Cauchy-Gauss mutation operator to jump out of possible premature maturity. The specific algorithm is shown in Figure 5;

[0148] (1) In the early stage of the algorithm, the Harris Eagle algorithm is used for global search. The search process is as shown in formula (6):

[0149] Where, X t+1 and X t are the positions of the individuals in the population at the t+1th iteration and the tth iteration respectively; X rand,t is the random position of the population at the tth iteration; X m,t is the average position of the population at the tth iteration; X prey,t is the prey position at the tth iteration; r1, r2, r3, r4 are random numbers between 0 and 1 respectively; ub and lb are the upper and lower bounds of the search space respectively, and q is a random number between (0,1);.

[0150] (2) Using the improved energy equation of the prey in the transition phase, determine whether the algorithm has entered the local search phase. The energy equation is as follows:

[0151] Where E is the energy of the prey to escape. When E≥1, the algorithm performs a global search, otherwise it performs a local search. E0 is the initial energy of the prey, E0=2×rand-1, rand is a random number between 0 and 1. max is the maximum number of iterations.

[0152] (3) In the local search phase, the fast search capability of the particle swarm algorithm is used to accelerate convergence. The search process is as shown in formula (8)-formula (9): V t+1 =ω t V t +c1r1(pbest t -X t )+c2r2(gbest t -X t ) (8) X t+1 =X t +V t+1 (9)

[0153] Where, X t+1 and V t+1 are the position and velocity of the particle in generation t+1 respectively; ω t is the weight; c1 and c2 are the individual learning and social learning factors of the particle respectively; r1 and r2 are random numbers between 0 and 1; pbest t is the individual optimal solution of the particle; gbest t is the global optimal solution for all particles in generation t.

[0154] In the later stage of the algorithm, the Cauchy-Gauss mutation strategy is introduced to improve the global search ability of the algorithm. If the optimal solution of the population does not change after 10 iterations, the current best individual is selected for mutation, as shown in formula (10): X newt =gbest t ×(1+ζ1Cauchy(0,1)+ζ2Gauss(0,1)) (10)

[0155] Where, X newt is the individual position after mutation; Cauchy(0,1) and Gauss(0,1) are random factors obeying Cauchy distribution and Gaussian distribution respectively; ζ1 and ζ2 are adaptive parameters of Cauchy distribution and Gaussian distribution respectively.

Claims

1. A dynamic assessment method for extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation, characterized by: Research urban system data and disaster prevention and mitigation data, using GIS technology to refine the risk unit division of disaster-prone objects at the scale of urban buildings and road networks and determine the spatial distribution of disaster-prone objects; simulate the impact of dynamic changes in flood water depth on disaster-prone objects based on extreme rainstorm waterlogging scenarios; and use spatial complex networks and shortest path planning to calculate the traffic capacity of the road network system and the accessibility of emergency services; Combining the two methods of urban flooding process simulation and indicator system, a dynamic urban flooding risk assessment technology that integrates disaster evolution and disaster prevention response process is proposed. The urban flooding risk of each disaster-bearing unit during the evolution of urban flooding disasters is dynamically assessed. A risk database is constructed through scenario setting. The Harris Eagle optimization algorithm and the particle swarm optimization algorithm are integrated to optimize the LSTM network, and a real-time urban flooding risk analysis model is constructed. Ultimately, dynamic assessment and calculation of urban flooding disaster risks due to extreme rainstorms are realized.

2. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 1, characterized in that: The specific steps are as follows: Step 1: Collect urban system data and urban disaster prevention and mitigation data; the urban system data includes road network system data, population density, land use, regional GDP, and urban lifeline engineering lines; the urban disaster prevention and mitigation data includes the number and location of flood control and drainage projects, and emergency flood control equipment, including the number and location of flood control and drainage vehicles and portable mobile pumps; Step 2: Categorize key urban flood prevention targets and vulnerable areas as disaster-prone objects. These disaster-prone objects are categorized into: important infrastructure, public service facilities, residential areas, economic functional areas, and vulnerable industries. Using building units and road networks as the scales, GIS technology is used to refine the risk unit division and determine the spatial distribution of the disaster-prone objects. Step 3: Build a high-precision flooding process model. Collect high-resolution terrain elevation data for key areas with high risk, conduct detailed flood simulations, and analyze the impact of dynamic changes in flood water depth on disaster-prone structures. Key areas with high risk include underpasses, urban subways, substations, and older urban areas. Step 4: Classify the disaster level into three levels: slight damage, severe damage, and major damage according to different disaster-prone bodies. Calculate the economic losses of the disaster-prone bodies by comprehensively considering the inundation depth, inundation area, and average economic value. Step 5: Calculate the traffic capacity of the road network and evaluate road failure; Step 6: Evaluate the accessibility of rescue services in combination with regional waterlogging emergency prevention and control strategies; Step 7: Set a rainstorm flooding scenario and dynamically assess the flooding risk; Step 8: Building a waterlogging risk database based on the waterlogging risk dynamic change assessment results of step 7; Step 9: Use the waterlogging risk database from step 8 to train the LSTM deep learning network. Use the Harris Eagle optimization algorithm combined with the particle swarm optimization algorithm to optimize the number of hidden layer neurons, weights, and learning rate of the LSTM network to improve the prediction accuracy of the LSTM network and build a real-time waterlogging risk analysis model. Step 10, based on the real-time, forecasted or set rainfall conditions and disaster prevention levels, adjust the scenario conditions, use the real-time waterlogging risk analysis model engine to perform real-time calculations, and obtain the dynamic process and deduction results of the spatiotemporal changes of waterlogging risks.

3. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 2, characterized in that: Step 4: Classify the disaster level according to different disaster-bearing bodies: Minor damage: people's movement is restricted, cars cannot start, roads are flooded, and houses are flooded; Severe damage: people and vehicles become unstable, roads become flooded and traffic is disrupted, and houses are submerged; Major disaster: people drowned, cars drowned, roads flooded, houses damaged.

4. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 3, characterized in that: Step 4: The method for calculating the economic losses of the disaster-affected body is shown in formula (1): Where L is the economic loss of the hazard-bearing body, I and J represent the total number of hazard-bearing body types and the total number of flooding depths, respectively; A ij is the inundation area corresponding to the jth inundation depth of the i-th hazard-bearing body type, D ij A ij Corresponding disaster level; Disaster level D ij Medium: slightly affected is 0.2, severely affected is 0.3, and severely affected is 0.5; i is the average economic value of the i-th disaster-prone unit type.

5. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 4, characterized in that: Step 5 is as follows: The roads with service functions in the urban area are divided into main roads and secondary roads, and the road intersections are abstracted into nodes using spatial complex networks. p ={v1,v2,…v p }, the road segments connected by the intersection are abstracted as a set of edges, E p ={e1,e2,…e p }, forming a road network graph G(V p ,E p );Adjacent intersection v i v j The required travel time T is used as the weight to construct the weight matrix Finally, the road network topology structure M(G) is formed; During the evolution of urban waterlogging disasters, if the water depth is greater than 50 cm, the road functionality will fail, resulting in changes in the road network topology. Based on the idea of ​​the shortest path, the regional road network traffic capacity is calculated as shown in formula (2): Where BC t (a) represents the traffic capacity of the road network and is the betweenness of node a at time t; n t is the total number of shortest paths connecting nodes b and c at time t, where a, b, and c represent arbitrary nodes; n t (a) is the total number of shortest paths connecting node a at time t, where the shortest path is calculated using the artificial bee colony optimization algorithm; If the depth of water on the road is greater than 50cm, the road is considered to be inoperable.

6. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 5, characterized in that: Step 6 is as follows: Taking the facility unit that provides emergency services as the center, the radiation range of emergency facilities is calculated according to the current status of the road network topology structure and the response arrival time threshold in different scenarios of waterlogging. Within the radiation range, the emergency response demand of the disaster-bearing bodies in high-risk areas is statistically analyzed, and the supply-demand ratio of emergency services is calculated. The calculation method of the supply-demand ratio of emergency services is as shown in formula (3): Where R j is the supply-demand ratio of emergency services; S j Represents the spatial area (d ij ,d0) service radiation range of emergency service facilities; N i is the number of high-risk units in the spatial area; G(d ij ,d0) is the Gaussian equation considering the effect of distance on the efficiency of emergency services; Taking the flood risk unit k as the center, search for rescue points within a given spatial distance threshold d0, and calculate the emergency service supply and demand ratio R of all risk units. j Gaussian equation is used for weighting, and weighted summation is used to obtain the overall emergency service accessibility A k , the emergency service accessibility is calculated as formula (4):

7. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 6, characterized in that: Step 7 is as follows: A refined assessment of urban flooding risk is carried out based on regional space, and an assessment index system considering hazard, exposure, vulnerability and disaster prevention and mitigation capabilities is constructed using a combination of the analytic hierarchy process and the entropy weight method. Scenario analysis is conducted in combination with the urban flooding process model, and the dynamic changes in urban flooding risk in disaster-bearing units are calculated by considering emergency drainage for road disaster prevention, emergency dispatch of flood control and drainage projects, and accessibility of emergency rescue services at different stages of the urban flooding process.

8. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 7, characterized in that: The specific steps for dynamic waterlogging risk assessment in step 7 are as follows: Step 7.1: Build an evaluation indicator system The evaluation index system includes the following 13 indicators: rainfall intensity, rainfall duration, flooding depth, flooding duration, flooding range, surface elevation, land use, population density, economic losses, road network traffic capacity, monitoring and early warning capabilities, flood control and drainage capabilities, and accessibility of emergency rescue services. The index weights are calculated using the hierarchical analysis method-entropy weight method. The economic losses of buildings involved are calculated using formula (1). a. The specific method of calculating the index weights by combining the analytic hierarchy process and the entropy weight method is as follows: (1) Constructing a judgment matrix: Using the 1-9 scaling method, compare each indicator pairwise to generate a judgment matrix; (2) Consistency test: Calculate the maximum eigenvalue of the judgment matrix and use the consistency ratio CR to determine whether the matrix meets the consistency requirements. When CR < 0.1, the matrix passes the consistency test. (3) Calculation of subjective weight By solving the eigenvector of the judgment matrix and normalizing it, the subjective weight of each indicator is obtained; b. Calculate objective weights using entropy weight method (1) Standardized data: All evaluation indicators are dimensionlessly standardized to ensure the consistency of the dimensions of each indicator; (2) Calculate information entropy: Calculate the information entropy value of each indicator based on the standardized value of each indicator. The information entropy formula is: Among them, p ij is the ratio of the standardized value of the i-th sample of the j-th indicator to the total, k′ is a constant, k′=1 / ln(n), n is the total number of samples; The objective weight is thus obtained as: in, is the objective weight of the jth indicator; s is the total number of evaluation indicators; c. Use the weighted average method to combine the subjective weight and the objective weight calculated by the entropy weight method, Among them, α is the subjective weight ratio, 0<α<1; Step 7.2: Dynamic calculation of the evolution process of extreme rainstorm waterlogging disaster: Based on the evolution of urban waterlogging, the dynamic evolution of urban waterlogging disaster scenarios is divided into five stages: initial scenario, evolution scenario, peak scenario, recovery scenario, and termination scenario. A waterlogging process model is used to simulate and calculate the inundation depth, inundation duration, and inundation range required by the indicator system in step 7.

1. The simulation considers three factors that contribute to reducing urban waterlogging: temporary drainage using portable road pumps, the rational allocation of emergency flood control and drainage vehicles to high-risk areas, and the emergency dispatch model of flood control and drainage projects. The rational allocation of emergency flood control and drainage vehicles is determined through the following three steps: (1) Ranking of high-risk road sections during waterlogging disasters: The water depth on the road is greater than 50 cm and the duration is greater than 15 minutes. The need for flood emergency drainage vehicles to carry out drainage operations is confirmed. High-risk road sections are ranked in the following order: key protection area > vulnerable industry > main road > time of occurrence of high-risk road sections. The scope of key protection areas, vulnerable industries, main roads, and high-risk sections can be obtained through observation of satellite image maps. The scope of key protection areas includes key protection objects and surrounding roads; The occurrence time of high-risk sections is calculated using a two-dimensional hydrodynamic model simulation; (2) Flood control and drainage vehicle route planning: Based on the road conditions under the waterlogging scenario, an artificial bee colony optimization algorithm is used to calculate the shortest path for flood control and drainage vehicles to reach the operation site and calculate the travel time. The new path needs to be re-planned and calculated after each drainage operation is completed; (3) Calculation of drainage by flood control drainage vehicles: Based on the peak water volume and the drainage volume of the flood control drainage vehicle, the drainage time is calculated. When the water depth in the area is less than 5 cm, the drainage operation is considered completed; Step 7.3: Calculate the road network traffic capacity and emergency rescue service accessibility at each stage of the waterlogging disaster evolution process in step 7.

2. The road network traffic capacity is calculated using formula (2), and the rescue service accessibility is calculated using formulas (3) and (4) in step 6. Calculate the waterlogging risk of different hazard-bearing bodies at each scenario stage. According to the combined weights in step 7.1, calculate the comprehensive flood risk score for each region or sample. Let the combined weight value of the i-th sample under the j-th indicator be x ij , then the comprehensive risk score R i for: Step 7.4, based on the standard deviation classification method, the mean and standard deviation of the risk score are used as the basis for classification. The mean μ and standard deviation σ of the comprehensive risk score are calculated, and the level is divided according to the standard deviation range of the score above and below the mean: Low risk: score is less than μ-σ, Medium risk: score between μ-σ and μ, Higher risk: scores between μ and μ+σ, High risk: score greater than μ+σ; On the basis of risk classification, experts are invited to optimize the risk level standards according to actual conditions and finally obtain the dynamic assessment results of urban waterlogging risk.

9. The method for dynamic assessment of extreme rainstorm waterlogging disaster risk for disaster prevention and mitigation according to claim 2, characterized in that: The specific steps for optimizing the LSTM network parameters in step 9 are as follows: Step 9.1: Taking the LSTM network prediction accuracy as the optimization target, construct the objective function, which is formula (5): Where y i is the i-th real data; is the average value of the real data; is the data prediction value; m is the number of data; Step 9.2: In the early stages of network parameter optimization iterations, the Harris Eagle optimization operator is integrated to improve global search capabilities. The improved energy equation of the prey in the transition phase is used to determine whether to enter the local search phase. The particle swarm optimization algorithm is used to accelerate convergence, and the mixed Cauchy-Gaussian mutation operator is introduced to avoid premature convergence. The details are as follows: (1) Use the Harris Hawk algorithm to perform global search. The search process is as shown in formula (6): Where, X t+1 and X t are the positions of the individuals in the population at the t+1th iteration and the tth iteration respectively; X rand,t is the random position of the population at the tth iteration; X m,t is the average position of the population at the tth iteration; X prey,t is the prey position at the tth iteration; r1, r2, r3, r4 are random numbers between 0 and 1; ub and lb are the upper and lower bounds of the search space respectively; q is a random number between (0, 1); (2) Using the improved energy equation of the prey in the transition phase, determine whether the algorithm has entered the local search phase; the energy equation is specifically as shown in formula (7): Where E is the energy of the prey to escape. When E≥1, the algorithm performs a global search, otherwise it performs a local search. E0 is the initial energy of the prey, E0=2×rand-1, rand is a random number between 0 and 1. max is the maximum number of iterations; (3) In the local search phase, the fast search capability of the particle swarm algorithm is used to accelerate convergence. The search process is as shown in formula (8)-formula (9): V t+1 =ω t V t +c1r1(pbest t -X t )+c2r2(gbest t -X t ) (8) X t+1 =X t +V t+1 (9) Where, X t+1 and V t+1 are the position and velocity of the particle in generation t+1 respectively; ω t is the weight; c1 and c2 are the individual learning and social learning factors of the particle respectively; r1 and r2 are random numbers between 0 and 1; pbest t is the individual optimal solution of the particle; gbest t is the global optimal solution for all particles in generation t; V t is the velocity of the particle in generation t; The Cauchy-Gauss mutation strategy is introduced to improve the global search capability of the algorithm. If the optimal solution of the population does not change after 10 iterations, the current best individual is selected for mutation, as shown in formula (10): X newt =gbest t ×(1+ζ1Cauchy(0,1)+ζ2Gauss(0,1)) (10) Where, X newt is the individual position after mutation; Cauchy(0,1) and Gauss(0,1) are random factors obeying Cauchy distribution and Gaussian distribution respectively; ζ1 and ζ2 are adaptive parameters of Cauchy distribution and Gaussian distribution respectively.

Images