Long-term Practice of Mathematical Modeling

1. Introduction

 I once entered the mathematical modeling team of wuhan university through several rounds of selection, and participated in various mathematical modeling competitions on behalf of wuhan university. In the process of participating in mathematical modeling competitions, I systematically studied the classical statistical model, graph theory model, differential equation model, optimization theory and machine learning algorithm, and wrote nine complete mathematical modeling papers and more than 10,000 lines of relevant codes represented by matlab. My experience in mathematical modeling has comprehensively improved my independent scientific research ability from the perspectives of model building, problem solving and paper writing.

2. Summary of my work

• Systematically studied mathematical modeling and data mining, and finished nine papers on mathematical modeling.
• Applied optimization models to production problems, and solved these models using different heuristic algorithms.
• Adopted fuzzy analytic hierarchy process and principal components regression to solve evaluation problems.
• Built two cellular automatons which simulate traffic flow in road network and “Fan-in” area of tollbooth respectively, in order to evaluate the influence of breaking up gated communities and find better design of tollbooth.
• Used statistical modeling and machine learning to find key factors and predict in steelmaking and aquaculture.
• Developed an innovative algorithm based on cluster analysis and greedy algorithm for shredded paper stitching.

3. Typical abstracts

3.1 Prediction model of key factors for steelmaking based on optical information

 Temperature has significantly affections on the content of key element C in the steel making process, and the temperature can’t be measured directly. The paper extracts the feature of measured optical data by principal component analysis and spectral feature analysis. Based on the extracted features, the initial temperature and initial key elements of steel are introduced, and the prediction model based on BP neural network is established. Furthermore, based on the cross experiment and error analysis of the three steelmaking processes, the paper optimize our model and accuracy of prediction is improved.
 In the first problem, in order to maintain as much original optical information as possible, the paper makes use of principal component analysis to extract the principal components. Our selection criterion is that the selected principal components include at least 85% of the original information content. 12 principal components are extracted in process 1, 7 principal components are extracted in process 2, 11 principal components are extracted in process 3. In addition, through the correlation analysis of light intensity and temperature, light intensity and the key element content, the change analysis of light intensity and time, we find that lights from f_280 to f_2048 has a strong thermal effect. At the same time, the amplitude of light intensity is correlated with the temperature and the content of key element. In order to describe the findings above, we select the change quantity of average light intensity of lights with thermal effect at a certain moment as a supplementary characteristic value. The supplementary characteristic value and the results of principal component analysis are combined to form the result of optical information feature extraction.
 In the second problem, we find that we can’t predict the current temperature or the current content of the key elements precisely by the characteristic value at a certain time. But the characteristic value can be used to reflect the current temperature and the rate of change of the key element. So the paper takes the characteristic value as the input, take the temperature, the change of the key elements as the output, select the Bayesian training function to establish neural network with a single hidden layer.
In the third problem, cross table experiments show that the model based on its own data has a high accuracy in the prediction of its own data. But the error is relatively large in the prediction of other groups of data. Consequently, we step back to optimize our prediction model. We speculate that the accuracy will be higher if we extract the feature vector with the same method and formula. We notice that there is data which has stable correlation with temperature T in three bands in each group of data-f_400 to f_670, f_730 to f_1170, f_1280 to f_2048. So we should adopt the data based on the light intensity of these three bands. Furthermore, we see that the amplitude of light intensity has a very significant correlation with the slope of T-t line in the figure, so we give two necessary factors for solving the amplitude: Spectral information at present time, the spectral information of the last moment. At last, we modify our model’s input to the average of light intensity at present time and light intensity of the last moment in three bands- f_400 to f_670, f_730 to f_1170, f_1280 to f_2048.
 The results show that our model can be used to speculate the temperature and the content of key elements by the optical information, and it is reasonable and practical.

3.2 Modeling and simulation about “Fan-in” Area after toll barrier

 We aim to find out the optimal design of the “fan-in” area. We construct the index of different objectives respectively in terms of the size and shape of the lanes, and integrate these indexes into a composite one using the Catastrophe Theory. Through maximizing the composite index, we obtain the optimal solution of the design.
 The building-block of our analysis is the Lane-Transferring Model. We assume that the vehicles change lanes only in certain parts of the lanes (called the “transferring area”). Then we apply the Queuing Theory to this transferring area to get the total throughput of the toll plaza.
 We also take into consideration the safety performance and construction costs. We construct the index, the “transferring pressure” to measure the riskiness of the design and we select area as the index to measure the construction costs. Then we aggregate the three indexes using the Catastrophe Theory. With the Genetic algorithm, we obtain the optimal lane-design(i.e. the length and size of each lane) with the optimal combination of the three objectives, throughput, safety performance and total costs.
 We change the throughput at the position of tollbooths to see the changes in the solution and find that the smoothness of the traffic flow in the “fan-in” area are compromised as the traffic becomes heavier. Furthermore, we vary the proportion of autonomous vehicles, and come to the conclusion that the increase of autonomous vehicles adds to the smoothness and throughput of the traffic flow and helps reduce the construction costs with an optimal design of lanes with shorter lengths. Also, we check the efficiency of different tollbooths, resulting in different lane design with the electronic toll collection booths being both the most smooth and the most risky one.
 Furthermore, we test the applicability of the theoretical solutions in the cellular automaton model, and perform the simulation analysis. We make detailed comparisons between situations of different vehicles and and of different tollbooths in the simulation process. In particular, we find that the times of the lane-transferring process increases as the number of vehicles go up, boosting the risk of traffic accidents in the process of transferring. Therefore, we proposed that speed limit should be set in the”fan-in” area in avoidance of emergencies.

4. My papper list

[1] Modeling and solving of wallpaper production problem based on improved genetic algorithm. (In Chinese)
[2] Comprehensive study on aquaculture pond. (In Chinese)
[3] Solving gantry crane programming model based on genetic algorithm. (In Chinese)
[4] Blanking problem model and its settlement. (In Chinese)
[5] Fragment splicing problem model and its solution. (In Chinese)
[6] Discussion on some contest net evaluation questions. (In Chinese)
[7] Study on the influence of opening of residential area on road passage. (In Chinese)
[8] Prediction model of key factors for steelmaking based on optical information.
[9] Modeling and simulation about “Fan-in” Area after toll barrier.