针对高层建筑地基易产生蝶形沉降，通过在桩顶增设连梁，提出一种便于施工，整体性较强的连梁-桩结构基础。主要研究内容与结论如下：

（1）通过有限元计算方法建立不同连梁高度与格室数目的连梁-桩结构。对竖向与水平承载力进行研究。对比沉降、轴力、侧摩阻力、土压力、弯矩、经济效益，分析连梁-桩结构的承载特性，得到连梁高1/10桩长，格室数目2时为最佳连梁-桩结构形式。对比群桩与连梁-桩结构基础，相同沉降时连梁-桩结构几乎不产生蝶形沉降，且对底板的不均匀荷载较小。通过10×10桩数的分析发现，不同桩数的连梁均能改善承载性能，且最佳连梁布置方式具有普适性。

（2）通过研究模型试验结果，分析沉降、轴力、侧摩阻力和连梁轴力，并与有限元计算结果对比，两者基本吻合。连梁-桩结构的荷载-沉降曲线为缓变型；连梁使轴力产生突变，在连梁作用下不同位置桩体轴力大致相同；连梁内外侧摩阻力存在明显差异，内摩阻力接近于0，而外摩阻力较大；连梁能承担荷载并传递较大受拉荷载。

（3）对比连梁-桩结构与群桩基础在竖向-水平组合荷载、侧向堆载和地震作用下的承载能力。由于连梁的荷载分担与传递作用，连梁-桩结构承载能力均高于相应的群桩；在组合荷载作用下，连梁还能发挥桩体横向连接的加固作用，侧面堆载时，连梁的存在使桩顶出现负弯矩，地震作用下连梁兼具一定的隔震减震作用。

（4）建立二维随机有限元模型，研究连梁-桩结构的承载能力，通过卷积神经网络建立随机场图像与基础极限承载力之间的模型进行承载力预测。考虑土体空间变异性的基础承载力与试验结果相吻合，高于确定性分析。采用卷积神经网络建立的基础承载力预测模型具有较高的准确性，可以用于参数分析。

（5）建立多组有限元模型，分析连梁宽度、桩间距、桩长径比、连梁刚度对结构竖向承载能力的影响，并对连梁-桩结构易损性进行研究。连梁宽度、桩间距及桩长径比均与结构的承载能力成正比，连梁的刚度对承载能力几乎没有影响。连梁宽度应控制在3.5倍的桩径之内，桩间距应控制在4.5倍桩径以上，6倍桩径以上时可以不考虑群桩效应影响。桩长径比与连梁刚度可根据实际情况，因地制宜进行调整。结构的易损面为连梁的四角及桩与连梁的连接处，易受拉破坏，需对其进行加固。

（6）采用响应面法建立54组试验并运用有限元求解，建立了相应的数学模型，通过NSGAⅡ遗传算法进行考虑造价的多目标优化，并基于不同权重系数分析得到最佳结构参数，采用响应面法代替有限元建模可以快速提高优化效率，且精度较高。造价的权重系数不宜取太高，建议取值0.5。选取最佳结构参数时工程造价降低了57.99%，沉降量下降了34.87%。

In order to mitigate the differential settlement issue commonly found in high-rise building foundations, a coupling beam pile foundation structure is proposed, which features easy construction and strong integrity by introducing coupling beams at the pile heads. The main research objectives and conclusions are as follows:

(1) Different configurations of coupling beam pile structures with varying heights and number of compartments were modeled using finite element software. The vertical and horizontal bearing capacities were investigated. By comparing parameters such as settlement, axial force, lateral friction resistance, soil pressure, bending moment, and economic benefits, the bearing characteristics of the coupling beam pile structure were analyzed. The basic form of the coupling beam pile structure was determined to have a coupling beam height equal to 1/10 of the pile length and 2 compartments. A comparison between the group pile foundation and the coupling beam pile structure revealed that the latter exhibited little differential settlement under the same settlement condition and imposed a relatively uniform load on the foundation slab. The analysis with a 10×10 pile arrangement demonstrated that coupling beams could improve the bearing capacity for different numbers of piles, and the basic layout of coupling beams had general applicability.

(2) Indoor model tests were conducted on the coupling beam pile structure to monitor settlement, axial force, lateral friction resistance, and coupling beam axial force. The results were compared with the finite element analysis and showed good agreement. The load-settlement curve exhibited a gradual change. There was a significant discontinuity in the axial force distribution, but under the influence of coupling beams, the axial forces of different locations of the piles were approximately the same, indicating a uniform load distribution. There was a noticeable difference in the lateral friction resistance between the inner and outer sides of the coupling beams, with the inner side having negligible frictional resistance while the outer side exhibited significant resistance. The coupling beam was capable of carrying loads and transmitting substantial tensile force.

(3) Compare the bearing capacity of the coupling beam pile structure and the group pile foundation under vertical-horizontal combined loads, lateral pile loads, and seismic actions. Thanks to the load transfer function of the coupling beam, the coupling beam pile structure has a higher bearing capacity than the group pile foundation. Under combined loads, the coupling beam can enhance the reinforcement function of the pile body. When subjected to lateral pile loads, the presence of the coupling beam induces negative bending moments at the pile heads. Under seismic actions, the coupling beam can provide certain isolation and damping effects, mitigating the vibration of the foundation.

(4) Establish a two-dimensional random finite element model to study the bearing capacity of the coupling beam pile structure. A model is established using convolutional neural networks to predict the foundation's ultimate bearing capacity based on stochastic field images. The results show that the foundation's bearing capacity considering the spatial variability of the soil agrees well with experimental results and is higher than deterministic analysis. The foundation's bearing capacity prediction model established by convolutional neural networks has high accuracy and can be used for parameter analysis.

(5) Establish multiple sets of finite element models to analyze the influence of coupling beam width, pile spacing, pile length-diameter ratio, and coupling beam stiffness on the vertical bearing capacity of the foundation structure. Additionally, study the vulnerability of the coupling beam pile structure. The coupling beam width, pile spacing, and pile length-diameter ratio are all proportional to the structural bearing capacity, while the stiffness of the coupling beam has minimal effect on the bearing capacity. It is recommended to control the coupling beam width within 3.5 times the pile diameter and the minimum pile spacing above 4.5 times the pile diameter, with no consideration of the group pile effect when the pile spacing exceeds 6 times the pile diameter. The pile length-diameter ratio and coupling beam stiffness should be adjusted according to the actual conditions. The vulnerable areas of the structure are the four corners of the coupling beam and the connections between piles and the coupling beam, which are prone to tensile failure and require reinforcement.

(6) A total of 54 experiments were conducted using the response surface method, and finite element analysis was employed to obtain corresponding mathematical models. The NSGAII genetic algorithm was then utilized for multi-objective optimization considering cost, and the optimal structural parameters were obtained based on different weight coefficients. The results show that replacing finite element modeling with the response surface method significantly reduces the optimization time while maintaining high accuracy. The weight coefficient for cost should not be too high, and a recommended value is 0.5. When selecting the optimal structural parameters, the engineering cost was reduced by 57.99%, settlement decreased by 34.87%.