1. Problem Statement One research direction in elevator group control systems (EGCS) is how to employ optimized control strategies to coordinate the operation of multiple elevators, thereby improving elevator operating efficiency and service quality. Increased operating efficiency also means reduced passenger waiting time and energy consumption, directly impacting service quality. This inevitably involves considering passenger comfort. Research has found that factors affecting passenger comfort are multifaceted: acceleration and its changes during start/stop, car vibration, waiting time, travel time, car crowding, the number of stops and starts during the journey to the destination floor, and factors such as car lighting and decoration. Waiting time, travel time, crowding, and the number of stops during operation are determined by the group control scheduling algorithm. Therefore, how to improve passenger comfort within the elevator group control scheduling strategy is a major problem addressed in this research. Since Mitsubishi first applied fuzzy logic to elevator systems, intelligent control methods have become the most studied scheduling algorithms in elevator group control systems. For example, the fuzzy neural network elevator scheduling method can make full use of the learning ability and information processing ability of the neural network, and continuously adjust the network parameters through learning to achieve optimal control. The scheduling method based on genetic algorithm determines the control target according to customer requirements and uses genetic algorithm to optimize the parameters of the scheduling evaluation function based on the predicted elevator call generation and distribution data [2-3]. However, these methods only consider technical factors and ignore riding comfort. In addition, due to the lack of understanding of the number of passengers, when the number of people waiting on a certain floor is too large for one elevator to handle, the next elevator will be allocated only after the elevator has transported some passengers away, which prolongs the waiting time and causes multiple elevators to be idle, while passengers cannot get enough elevators. To solve the above problems, the design is carried out from the following two aspects: by configuring multiple sensors to obtain more passenger information, and from the perspective of riding comfort, a task allocation algorithm based on fuzzy reasoning is adopted to reduce congestion, shorten waiting time, and reduce the number of stops. 2 Structural Design of Elevator Control System The improved elevator control system structure is shown in Figure 1. The system comprises five main parts: an elevator dispatch controller, a main controller, a car controller, an instruction board, and a call board. The main controller primarily handles elevator operation control, door opening and closing control, and instruction processing. First, in learning mode, the main controller acquires hoistway height and elevator position information by recording pulses input from the rotary encoder and sensors at each floor. After learning, in operation mode, the main controller sends control signals to the frequency converter based on the recorded data to control elevator operation and respond to user commands and calls. The car controller receives commands from the instruction board and signals from sensors within the car, and controls the equipment in the car according to commands from the main controller and the user. The elevator dispatch controller receives call signals and passenger information from the call board. The instruction board and call board receive user commands and simultaneously feed back elevator status information to the user, enabling human-machine interaction. In this system, both communication and control signals are implemented digitally, improving control accuracy and simplifying software design. The controllers communicate with each other via a CAN bus for multi-master communication, which is fast, stable, and has a simple interface. 3. The sensors in the elevator group control system can be considered as two separate sensors: the external call buttons in the lobby and the internal call buttons in the car. These sensors receive call requests from passengers in the lobby and destination instructions from passengers in the car, and then send this information to the elevator dispatch controller as two input signals to the dispatch unit. Image sensors located in the lobby of each floor are used to calculate the number of people waiting to use the elevator on that floor. The image sensor consists of a camera and a connected microprocessor. To reduce computational load and minimize interference from useless information, the image sensor only starts working and calculating the number of people waiting when a call button in the lobby of that floor is pressed. When only the up or down button is pressed, the image sensor calculates the number of passengers waiting for the elevator as the number of passengers going up or down. If both buttons are pressed, the number of passengers going up and down is estimated based on the floor's position relative to the entire building. That is, assuming a building has N floors, and both the up and down call buttons in the Mth floor lobby are pressed, the image sensor on that floor calculates the number of passengers waiting for the elevator as X. Then, the number of passengers going up can be assumed to be X, where floor(x) is the largest integer not greater than x. By adding this sensor, the weight sensor in the shaft determines the number of passengers in the car by weighing them. The average weight of the passengers is taken, with a certain margin of 70 d/person. If the total weight of the passengers in the car is W, then the number of passengers in the car is n = W/70. This can be used to calculate the number of passengers the car can still hold, serving as the basis for whether the elevator responds to passenger calls. Speed ​​sensors measure the current operating speed of each elevator. Because the elevator speed command curve is often designed using a distance principle when decelerating to level, it ensures elevator comfort, leveling accuracy, and ease of on-site adjustments. Simultaneously, the elevator car's position is determined by position sensors by calculating the movement pulses generated by the pulse encoder and calculating the synchronization floor based on the floor heights stored in memory. Therefore, the elevator dispatch controller uses this speed and position to determine the nearest floor the elevator can respond to. Figure 2 shows the sensor configuration of the elevator group control system. 4. Task Algorithm Design Based on data acquired from various sensors, the following control parameters are selected: ① Waiting Time (WT): The time from when a passenger presses the call button to when they enter the elevator; ② Riding Time (RT): The time from when a passenger enters the elevator to when they reach their destination floor; ③ Number of Passengers (RN): The number of passengers in the elevator car, i.e., the crowding level; ④ Number of Waiting Passengers (WN): The number of passengers waiting for the elevator in the lobby; ⑤ Number of Stops (SN): The number of stops a passenger makes from entering the elevator to reaching their destination floor; ⑥ Relative Distance (RD): The number of floors the elevator passes through from the current floor to the destination floor. Then, the elevator dispatch controller calculates the control parameters for each elevator responding to a call. Because the response to an elevator call request is a multi-input, single-output process, considering the difficulty of multi-input, single-output processing and the many uncertainties in the system, a multi-rule weighted fuzzy algorithm is used. For each input and output of the system, fuzzy inference rules are established in a one-dimensional space. The final output is obtained by calculating the weighted sum of all rule outputs. This approach allows for the creation and modification of fuzzy rules effectively, offering greater flexibility than traditional fuzzy reasoning methods.