Campus business one-stop service system based on AIoT

Abstract

The invention relates to the field of campus business management, discloses an AIoT-based campus business one-stop service system, and aims to solve the problems of low processing efficiency, serious information island, poor user experience, extensive energy consumption management and unintelligent equipment management of campus business in a traditional mode. An Internet of Things equipment integration unit of the system collects campus energy consumption, equipment state and people flow density data; the edge calculation unit preprocesses the data and identifies abnormity; the AI capability core unit provides natural language interaction and multi-modal analysis functions; the lightweight AI reasoning unit predicts equipment faults and gives maintenance suggestions; the dynamic model management unit realizes dynamic loading switching and framework expansion of multiple models; therefore, intelligent processing of campus business is realized, business processing efficiency is improved, information islands are broken, user experience is improved, energy consumption management is optimized, stable operation of equipment is guaranteed, and campus informatization management level is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of campus business management, and specifically relates to a one-stop service system for campus business based on AIoT. Background Art

[0002] With the continuous expansion of the scale of colleges and universities and the continuous promotion of informatization construction, various business services on campus have become increasingly complex and diverse. There are many problems in the traditional campus service model, which are difficult to meet the growing needs of teachers and students, specifically reflected in the following aspects:

[0003] 1. Low business processing efficiency: In the aspects of teaching management and general affairs and logistics affairs processing, the manual operation process is cumbersome. For example, for dormitory repair, teachers and students need to fill in forms offline, queue up to submit, and it is difficult to query the repair progress, resulting in a long transaction processing cycle and low efficiency, affecting the learning and living experience of teachers and students.

[0004] 2. Serious information island phenomenon: Various campus platforms such as the educational administration system, student affairs system, and financial system are independent of each other, and data cannot be shared, forming information islands. This makes it difficult for schools to obtain comprehensive and accurate data support when conducting comprehensive management and decision-making, reducing management efficiency and the scientific nature of decision-making.

[0005] 3. Poor user experience: When teachers and students use campus services, they need to switch between multiple systems, which is inconvenient to operate. Moreover, the traditional service method cannot provide personalized services, cannot accurately push and handle affairs according to the needs and preferences of teachers and students, and cannot meet the expectations of users for convenient, efficient, and personalized services.

[0006] 4. Extensive energy consumption management: The campus has a large energy consumption, but lacks effective monitoring and management means. The traditional energy management method cannot obtain energy consumption data in real time, making it difficult to conduct energy consumption analysis and optimization, resulting in energy waste and increasing the school's operating costs.

[0007] 5. Unintelligent equipment management: Various equipment on campus (such as cafeteria equipment, dormitory facilities, etc.) lacks intelligent management, and equipment failures cannot be discovered and processed in time, affecting the normal use and lifespan of the equipment, and also bringing inconvenience to the lives of teachers and students. Summary of the Invention

[0008] The purpose of the present invention is to provide a one-stop service system for campus business based on AIoT to solve at least one of the above technical problems.

[0009] The purpose of the present invention can be achieved through the following technical solutions:

[0010] A one-stop service system for campus business based on AIoT includes:

[0011] IoT device integration unit: used to establish connections with smart meters, sensors and campus card terminals to collect energy consumption data, equipment operation status information and crowd density data in campus scenarios in real time;

[0012] Edge computing unit: deployed at the IoT gateway node, performs local preprocessing on the collected raw data, completes abnormal data identification and real-time response logic execution;

[0013] AI capability core unit: Integrates a dedicated semantic understanding model for university business scenarios, provides a natural language interaction interface, and supports multimodal fusion analysis of text, images, and time series data;

[0014] Lightweight AI inference unit: Integrate artificial intelligence models in the IoT gateway to achieve localized fault prediction of equipment operation status, generate fault type labels and maintenance suggestions, and synchronize them to the management backend unit;

[0015] Dynamic model management unit: Through the model scheduling algorithm, it realizes the dynamic loading and switching of various artificial intelligence models according to the business load and model performance indicators, and provides extension interfaces for the open source frameworks PyTorch and HuggingFace. The artificial intelligence models include ERNIE and DeepSeek.

[0016] As a further technical solution, the IoT device integration unit includes a face recognition terminal device, which is used to work with the authority control sub-unit in canteen consumption and venue reservation scenarios to perform real-name identity authentication and authority management.

[0017] As a further technical solution, the edge computing unit synchronizes the pre-processed device status data to the management background in real time through the distributed log monitoring system and abnormal warning mechanism, and generates a visual data interface including user activity statistics and energy consumption analysis reports.

[0018] The edge computing unit collects device operation logs through a distributed log monitoring system (based on the ELK stack), generates warning messages in JSON format based on the abnormal warning mechanism, and pushes them to the visual data dashboard of the management background unit in real time. The data dashboard integrates the Grafana tool to display user activity heat maps, energy consumption trend analysis charts and equipment health status dashboards.

[0019] As a further technical solution, the AI ​​capability core unit includes: Multilingual interaction subunit: Based on pre-trained language models of different national languages ​​and dialects from different regions, it realizes the intention recognition of natural language input and the generation of multilingual responses; Context-aware engine: Associates the user's historical interaction records through the session ID, and uses the attention mechanism Transformer to achieve the context logical coherence of multi-turn conversations.

[0020] As a further technical solution, the system is also integrated with the robotic process automation (RPA) technology to achieve cross-platform data automatic synchronization and transaction linkage processing of the academic affairs management system, the student affairs system, and the financial system.

[0021] As a further technical solution, the management background unit includes:

[0022] Permission control sub-unit: Based on the role-based access control (RBAC) model, defines the role permission mapping tables for students, faculty, and administrators, and implements user identity authentication through the OAuth2.0 protocol;

[0023] Audit log sub-unit: Records user operation behaviors and system events, and supports log retrieval based on timestamps and operation types;

[0024] Data synchronization interface: Real-time synchronizes the user basic data with the academic affairs system and the student affairs system through the GraphQL API to ensure the dynamic update of the permission policy. The user basic data includes student numbers and department information.

[0025] As a further technical solution, the model scheduling algorithm of the dynamic model management unit includes:

[0026] Real-time monitors the business load, and calculates the business load Evaluates the business pressure when Exceeds the threshold Triggers scheduling, where Is the number of business requests, Is the average processing time, 、 Are preset weight coefficients and ;

[0027] Evaluates the model performance metrics, including the accuracy rate 、response time R and resource utilization rate ;

[0028] Based on the business type, selects the model and the model performance metrics to achieve dynamic model loading and smooth switching. Loads the new model according to the storage path and interface specifications, and verifies the compatibility after loading. When switching, migrates the business step by step according to the ratio Gradually migrates the business, Is the number of shunted requests, Is the total number of requests;

[0029] Adopt a model caching and elimination strategy based on the least recently used (LRU) principle, maintain a queue for recording model access times, and eliminate the model with the last access time when the cache is insufficient. The model with the minimum value.

[0030] The process of selecting a model based on business type and model performance metrics is as follows:

[0031] Calculate the comprehensive index for response time-sensitive services And select The model with the minimum value, and for accuracy-sensitive services, select the model with the Highest accuracy; is the weight of the response time, ; for balanced services, construct a balanced evaluation model and obtain the balanced evaluation index through the index balanced evaluation model And select The model with the largest value.

[0032] As a further technical solution, the process of constructing the index balanced evaluation model is as follows:

[0033] S1. Determine the key metrics for evaluating model performance, including accuracy, response time, and resource utilization;

[0034] S2. Determine the corresponding weight coefficients according to the importance of each metric to the business; among them, the accuracy weight is , the response time weight is , the resource utilization weight is , and satisfy ;

[0035] S3. Perform standardization processing on the response time and resource utilization, where the standardized response time is , and the standardized resource utilization is ;

[0036] S4. Construct the balanced evaluation index ; The expression is: ;

[0037] Among them, is the warning value of accuracy, is the warning value of the standardized response time, is the warning value of the standardized resource utilization, , , are preset coefficients used to control the change amplitude when the metric exceeds the warning value. The larger the value, the greater the impact after exceeding the warning value. It is an influence coefficient, which is used to measure the degree of mutual influence among accuracy rate, standardized resource utilization rate, and standardized response time.

[0038] Advantages of the present invention:

[0039] (1) The IoT device integration unit collects data in real time, and the edge computing unit preprocesses it quickly. Combining the intelligent analysis of the AI capability core unit and the lightweight AI inference unit, the service processing time is significantly shortened. For example, the dormitory repair process has changed from a traditional offline cumbersome process to an online process of quick submission, intelligent allocation, and efficient processing. The repair progress can be queried in real time, greatly improving the usage experience of teachers and students and reducing waiting time.

[0040] (2) By integrating the robotic process automation (RPA) technology of the system and the data synchronization interface of the management background unit, the automatic synchronization of data and the linkage processing of transactions in multiple systems such as academic affairs, student affairs, and finance are realized; school management can obtain comprehensive and accurate data for scientific decision-making, such as optimizing course arrangements and reasonably allocating resources according to students' comprehensive data.

[0041] (3) The multi-language interaction sub-unit and the context awareness engine of the AI capability core unit enable the system to understand multiple languages and user intentions, providing personalized services; for example, according to students' historical interaction records and preferences, course information, activity notifications, etc. are accurately pushed, with convenient operations, meeting the needs of teachers and students for efficient and personalized services. Description of the Drawings

[0042] The present invention will be further described below with reference to the drawings.

[0043] Figure 1 It is the system structure block diagram of the present invention. Detailed Embodiments

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.

[0045] To solve the above problems, the integration of artificial intelligence technology and Internet of Things technology brings new ideas to campus service. The campus business one-stop service system based on AIoT emerges as the times require. It can integrate various resources on campus, realize real-time data collection, analysis, and processing, provide intelligent and personalized services, improve the campus informatization level and management efficiency, and improve the campus life experience of teachers and students.

[0046] Please refer to Figure 1As shown in the figure, the present invention is a one-stop service system for campus services based on AIoT, including:

[0047] IoT device integration unit: used to establish connections with smart meters, sensors, and campus card terminals, and collect energy consumption data, device operation status information, and pedestrian flow density data in the campus scenario in real time; in this embodiment, connections are established with multiple devices to collect data such as energy consumption, device operation status, and pedestrian flow density in real time, providing rich and real-time data support for various business areas on campus. The energy consumption data helps the energy management department analyze the campus energy consumption pattern. For example, if it is found that the energy consumption in certain areas is too high during specific periods, targeted energy-saving strategies can be formulated to achieve optimized energy utilization and reduce operating costs. The device operation status data enables maintenance personnel to detect potential faults in a timely manner, arrange maintenance in advance, reduce device downtime, and ensure the normal operation of campus facilities. The pedestrian flow density data can assist in planning the campus space and rationally allocating resources. For example, according to the pedestrian flow in different teaching buildings during class hours, the opening hours and service capacity of public areas such as libraries and canteens can be adjusted to improve the utilization efficiency of campus resources.

[0048] Edge computing unit: deployed at the IoT gateway node, it performs local preprocessing on the collected raw data, completes abnormal data identification and real-time response logic execution; in this embodiment, local preprocessing of the raw data is performed at the IoT gateway node, reducing unnecessary data transmission and alleviating network congestion; abnormal data identification and real-time response logic execution can quickly process local data and respond in a timely manner to emergencies in campus services. When abnormal device operation data occurs, the edge computing unit can immediately trigger an alarm and execute preset operations, such as shutting down the faulty device to prevent the expansion of the fault, and at the same time send an abnormal report to the management background, improving the stability and reliability of the system; through abnormal data identification, incorrect or abnormal values are removed, providing a more accurate data basis for subsequent data analysis, improving the analysis accuracy of the AI capability core unit and the lightweight AI inference unit, and ensuring more reliable data-based decision-making.

[0049] Core Unit of AI Capability: Integrate a dedicated semantic understanding model for university business scenarios, provide a natural language interaction interface, and support multi-modal fusion analysis functions for text, images, and time-series data; the integrated dedicated semantic understanding model and natural language interaction interface support multiple languages, facilitating natural and smooth interaction between teachers, students, and the system. Teachers and students can quickly query course information, campus activity arrangements, life services, etc. through voice or text, without complex operation steps, greatly enhancing the user experience and improving the convenience and accessibility of campus services; at the same time, support multi-modal fusion analysis of text, images, and time-series data, enabling in-depth understanding of campus business scenarios from multiple dimensions. By combining students' course text records, classroom performance images, and learning time-series data, it can comprehensively evaluate students' learning situations, providing a basis for personalized teaching and tutoring. It can also conduct comprehensive analysis of campus security, activity organization, etc. by integrating different types of data, enhancing the scientificity and accuracy of campus management.

[0050] Lightweight AI Inference Unit: Integrate an artificial intelligence model in the IoT gateway to achieve local fault prediction of device operation status, generate fault type labels and maintenance suggestions, and synchronize them to the management background unit; the integrated artificial intelligence model realizes local fault prediction of device operation status, discovers potential problems of devices in advance. By continuously monitoring device operation data, the model can predict when a device is likely to fail and generate detailed fault type labels and maintenance suggestions; transform campus device maintenance from passive repair to active prevention, reduce the impact of sudden device failures on teachers' and students' learning and life, extend the service life of devices, and reduce device maintenance costs. For example, predict the fault risk of cafeteria equipment and arrange maintenance personnel in advance to ensure the normal supply of catering services.

[0051] Dynamic Model Management Unit: Through the model scheduling algorithm, according to the business load and model performance metrics, it realizes the dynamic loading and switching of multiple artificial intelligence models, and provides extension interfaces for the open-source frameworks PyTorch and HuggingFace. The artificial intelligence models include ERNIE and DeepSeek. According to the business load and model performance metrics, it dynamically loads and switches artificial intelligence models to ensure that the system can maintain efficient operation in different business scenarios. During the business peak period, the system automatically switches to a model with high processing speed to reduce the response time and improve the service efficiency; in business scenarios with high accuracy requirements, such as academic research data processing, it switches to a model with high accuracy to ensure the accuracy of data processing, make full use of the advantages of different models, optimize the system resource configuration, and improve the overall performance of the system; it provides extension interfaces for the open-source frameworks PyTorch and HuggingFace, which is convenient for integrating more advanced artificial intelligence models, can keep up with the pace of technological development, continuously introduce new models and algorithms, and enhance the functions and adaptability of the system. For example, the latest natural language processing models can be integrated to improve semantic understanding ability; or more advanced image recognition models can be introduced to optimize the campus security monitoring function to meet the changing needs of campus business.

[0052] The Internet of Things device integration unit includes a face recognition terminal device, which is used to link with the permission control sub-unit in scenarios such as cafeteria consumption and venue reservation for real-name identity verification and permission control. It guarantees the security and standardization of campus business, prevents illegal access and operations, and ensures the reasonable use of campus resources; for example, only users who have passed the identity verification can use specific venues to avoid resource abuse.

[0053] The edge computing unit, through the distributed log monitoring system and the anomaly warning mechanism, synchronizes the preprocessed device status data to the management background in real time and generates a visual data interface including user activity statistics and energy consumption analysis reports.

[0054] The edge computing unit collects device operation logs through a distributed log monitoring system (based on the ELK stack) and generates an alarm message in JSON format based on the anomaly warning mechanism, which is pushed to the visual data dashboard of the management background unit in real time. The data dashboard integrates the Grafana tool to display the user activity heat map, energy consumption trend analysis chart, and device health status dashboard.

[0055] The AI Capability Core Unit includes:

[0056] Multi - language Interaction Sub - unit: Teachers and students on campus come from different regions and have diverse language usage habits. The multi - language interaction sub - unit plays an important role based on pre - trained language models of different national languages and dialects in different regions. For example, in the freshman enrollment consultation service, a student from a minority area asks about the dormitory location and registration process in his / her native language. The multi - language interaction sub - unit accurately identifies his / her intention and uses the corresponding language model to generate a detailed guide in that ethnic language. In this process, the system first performs speech recognition on the input speech, converts it into text form, then judges the language type through a language classification model, and then inputs the text into the corresponding pre - trained language model for intention understanding. When generating a response, relevant information is retrieved from the knowledge base according to the intention, and the language generation model is used to convert the information into text in that ethnic language. Finally, the speech is output through the speech synthesis module. Through such a process, the intention recognition of natural language input and multi - language response generation are realized, greatly facilitating teachers and students with different language backgrounds to obtain campus service information.

[0057] Context - aware Engine: Taking the campus intelligent customer service system as an example, when a student asks "I want to query this semester's course schedule", the system records the session ID. When the student then asks "Where can I buy the textbooks for these courses", the context - aware engine associates these two interaction records through the session ID and analyzes the previous query "course schedule" and the current query "textbook purchase" using the attention mechanism Transformer. It will notice that both queries are related to course - related information, thus understanding that the student's intention is to further understand textbook purchase information based on the queried course schedule. In this way, the system can accurately provide accurate responses such as the purchase location and purchase method of the corresponding course textbooks according to the course schedule information, achieving the context logical coherence of multi - turn conversations and enhancing the fluency and satisfaction of the user - system interaction.

[0058] System Integrated with Robotic Process Automation (RPA) Technology: At the beginning of each semester, the academic affairs management system needs to update students' course information, the student affairs system needs to synchronize students' academic progress, and the financial system needs to obtain students' payment information to confirm the registration status. By integrating RPA technology, the system can automatically execute cross - platform data synchronization and transaction linkage processing according to the preset process. For example, when the academic affairs management system updates a student's course schedule, the RPA robot will automatically obtain this updated information, extract the data related to academic progress from the course information according to the preset rules, and synchronize it to the student affairs system. At the same time, the RPA robot will check the payment record of this student in the financial system. If the student has paid, it will mark the student as having completed registration in the academic affairs management system and the student affairs system, realizing the automatic synchronization of data between multiple systems and the linkage processing of transactions, avoiding the cumbersome and error - prone manual operations, and greatly improving work efficiency and data accuracy.

[0059] The management background unit includes:

[0060] Permission control subunit: In campus management, different roles have different permissions. Based on the Role-Based Access Control (RBAC) model, the permission control subunit defines a detailed role-permission mapping table. For example, a student role can only view personal course grades, course selection information, campus card consumption records, etc.; in addition to viewing their own relevant information, a faculty and staff role can also view the grades of students in the courses they teach and perform operations such as course management; the administrator role has the highest permissions and can configure the system, manage user information, view various data statistics of the whole school, etc. In actual applications, when a student tries to access the course management page of a teacher, the system performs user authentication through the OAuth2.0 protocol. The system will verify the identity token of the student, confirm that it is a student role, and judge according to the role-permission mapping table that the student does not have the permission to access this page, thereby blocking the access and returning a prompt message of insufficient permissions, ensuring the security of system data and the standardization of operations.

[0061] Audit log subunit: In the daily operation of the campus system, the audit log subunit plays an important role in recording and tracing. For example, when an administrator modifies user information in the system, the audit log subunit will record the timestamp of the operation, the operation type (such as modifying user information), the identity information of the operator, and the specific modification content (such as modifying the major information of a certain student). When data anomalies or security problems occur, the administrator can retrieve the logs through the audit log subunit according to the timestamp and operation type. For example, if it is found that the grades of a certain student have changed abnormally, the administrator can retrieve the operation logs related to the student's grades during the period of grade change to check whether there are illegal operations or misoperations, providing a strong basis for troubleshooting and solving problems.

[0062] Data synchronization interface: With the changes in campus personnel, such as students changing majors and faculty and staff transferring departments, the user basic data needs to be updated in a timely manner to ensure the dynamic update of the permission policy. Through the GraphQL API, the data synchronization interface synchronizes the user basic data with the educational administration system and the student affairs system in real time. For example, when a student transfers from the School of Computer Science to the School of Mathematics, the educational administration system will update the major information of the student. After receiving this update information, the data synchronization interface will synchronize the new major information to the student affairs system and adjust the permission policy of the student in the management background according to the new department information (School of Mathematics). For example, if the student had access rights to certain specific course resources when in the School of Computer Science, after transferring to the School of Mathematics, these permissions will be adjusted according to the new permission policy to ensure that the permissions are consistent with the actual situation of the student, improving the accuracy and standardization of campus management.

[0063] The model scheduling algorithm of the dynamic model management unit includes:

[0064] Monitor the business load in real time, and calculate the business load Evaluate the business pressure when Exceed the threshold Trigger scheduling, where Is the number of business requests, Is the average processing time, 、 Are preset weight coefficients and ; In the business load monitoring, N is the number of business requests statistically counted within a specific time period, T is the average processing time of all requests within this time period, 、 Are preset by the system administrator according to business characteristics and experience;

[0065] Evaluate the model performance indicators, including accuracy 、response time R and resource utilization rate ; When evaluating the model performance, the number of correctly predicted samples and the total number of predicted samples in the calculation of accuracy A are obtained based on the processing results of the model for the validation dataset; the response time R is obtained by measuring the time-consuming of the model to process requests multiple times and taking the average value; the model resource occupancy and the total system resources in the resource utilization rate U are statistically counted for key resources such as CPU usage or memory occupancy rate;

[0066] Select the model and model performance indicators based on the business type, realize the dynamic loading and smooth switching of the model, load the new model according to the storage path and interface specification, verify the compatibility after loading, and gradually migrate the business according to the ratio When switching, Is the number of shunted requests, Is the total number of requests; The process of selecting the model based on the business type and model performance indicators is:

[0067] For response time-sensitive services, calculate the comprehensive index And select The model with the smallest value, for accuracy-sensitive services, select the model with the highest accuracy ; Is the response time weight, ; For balanced services, an equilibrium evaluation model is constructed, and the equilibrium evaluation index Is obtained through the index equilibrium evaluation model, and select The model with the largest value.

[0068] When selecting the model, the business type is determined according to the business requirements document or system configuration information, The value is adjusted according to the sensitivity of the business to the response time. The closer the value is to 1, the more sensitive it is to the response time;

[0069] During the dynamic loading and switching process of the model, the compatibility verification includes checking the adaptability of the interfaces of the new model with other modules of the system and the consistency of the data formats; the shunt ratio The initial value is small and gradually increases as the verification of the running stability of the new model progresses until the full-scale switch is completed;

[0070] Adopt a model caching and elimination strategy based on the Least Recently Used (LRU) principle, maintain a record queue of model access times, and eliminate the model with the last access time when the cache is insufficient The smallest model. When caching and eliminating models, the value of each model in the access time record queue is updated every time the model is accessed value.

[0071] In this embodiment, scheduling is triggered by real-time monitoring of the business load. When exceeds the threshold it indicates that the current business pressure is relatively high and the existing model may not be able to handle the business efficiently. At this time, scheduling is triggered, and a suitable model is selected based on the business type. For example, for business that is sensitive to response time, calculate the comprehensive index and select the model with the smallest value, which can ensure that the system can still quickly respond to business requests under high load; during the peak period of campus course selection, a large number of students request the course selection service at the same time. The system monitors that the business load exceeds the standard, and selects a model with a fast response speed through scheduling, enabling students to complete the course selection operation quickly, avoiding long waits, and improving the user experience;

[0072] In different business scenarios, evaluate the model performance indicators (accuracy, response time, resource utilization rate), select the most suitable model and implement dynamic loading and smooth switching. After loading the new model, verify the compatibility, and gradually migrate the business according to the ratio when switching to reduce the switching risk; for balanced business, build a balanced evaluation model to obtain the balanced evaluation index , select the model with the largest value to ensure that the model achieves balance in multiple aspects of performance; when processing the campus financial reimbursement review business, it is necessary for the model to have a high accuracy rate to judge the compliance of the reimbursement, and at the same time ensure that the response time and resource utilization rate are within a reasonable range. Select a suitable model through balanced evaluation to ensure the stable and accurate processing of the business.

[0073] Adopt a model caching and elimination strategy based on the Least Recently Used (LRU) principle to maintain a record queue of model access times; when the cache is insufficient, eliminate the model with the smallest last access time t, and use the limited cache space to store models frequently used recently. In daily campus operations, certain models are frequently used during specific time periods, such as the performance analysis model during exams, while they are used less during other time periods. Through the LRU strategy, the cache content can be dynamically adjusted to improve the utilization rate of cache resources, avoid resource waste, and ensure the efficient operation of the system;

[0074] Finally, a flexible model management mechanism is provided. Through clear business load monitoring, performance metric evaluation, model selection, and switching strategies, the system can easily integrate new models; as campus business develops and technology progresses, newly developed or optimized artificial intelligence models can be conveniently incorporated into the system and scheduled for use according to business requirements and model performance; when a new intelligent device management model is introduced on campus, it can be integrated into the system according to this algorithm to enhance the system's adaptability to new business scenarios and maintain the system's advancement and competitiveness.

[0075] The process of constructing the balanced evaluation model of the metrics is as follows:

[0076] S1. Determine the key metrics for evaluating model performance, including accuracy, response time, and resource utilization rate;

[0077] S2. Determine the corresponding weight coefficients according to the importance of each metric to the business; among them, the accuracy weight is the response time weight is and the resource utilization rate weight is , and it satisfies ;

[0078] S3. Standardize the response time and resource utilization rate, where the standardized response time and the standardized resource utilization rate ;

[0079] S4. Construct the balanced evaluation metric ; the expression is: ;

[0080] Among them, is the warning value of accuracy, is the warning value of the standardized response time, is the warning value of the standardized resource utilization rate, , , are preset coefficients determined based on historical data analysis, used to control the change range when the metric exceeds the warning value. The larger its value, the greater the impact after exceeding the warning value, The influence coefficient is determined based on historical data analysis and is used to measure the degree of mutual influence among accuracy rate, standardized resource utilization rate, and standardized response time.

[0081] For the accuracy rate A, when A is close to or lower than then the value of is close to 0, and the contribution to the formula is mainly determined by ; when A exceeds this part of the value will increase rapidly with the increase of A, highlighting the importance improvement of the accuracy rate for the evaluation index when it exceeds the warning value; similarly, for the standardized resource utilization rate

[0082] Weight and respectively reflect the degree of importance of the business for each index. By weighting different indexes are incorporated into the comprehensive evaluation. The second half of the formula comprehensively considers the standardization and warning processing results of the three indexes in the form of scores. The first half introduces the cross terms of the three indexes to reflect the mutual influence relationship between them. For example, when all three indexes perform well, the value of the cross term will increase, further improving the value of the evaluation index ; if a certain index performs poorly, the cross term will have a negative impact on the value.

[0083]

[0084] Through the above technical solution, a method for constructing an index balanced evaluation model is provided. Since campus services are diverse and different services have different requirements for the model performance indexes, this model can comprehensively consider indexes such as accuracy rate, response time, and resource utilization rate, and accurately select a suitable model for balanced services. When processing the student grade query service, it is necessary for the model to respond quickly, ensure accurate results, and reasonably utilize system resources. Through the calculation of the balanced evaluation index model, the most suitable model that can balance these requirements can be determined among multiple candidate models, improving the service processing effect.

[0084] If the system detects high business load during a certain period, and the utilization rate of some model resources is high while the response time is long, the balanced evaluation index model can switch to a model with better performance in a timely manner. This can not only relieve the system resource pressure, but also shorten the business response time, enhance the overall operation efficiency of the system, and ensure the smooth progress of campus business. Moreover, the balanced evaluation index model monitors the model performance in real time. When the model indicators deviate from the expectations, such as a decrease in accuracy or an extension of the response time, the system can detect it in a timely manner and take measures. By switching the model or optimizing the existing model, it ensures that the business is not affected, provides continuous and stable services for campus business, and avoids business interruption caused by model failures.

[0085] Finally, campus business changes over time, such as the addition of new courses and facility updates. The balanced evaluation index model can flexibly adapt to these changes and adjust the weights and parameters of the evaluation indicators according to the new business requirements. When new intelligent teaching equipment is introduced to the campus and the requirements for the accuracy and timeliness of monitoring the operation status of the equipment are increased, the model can correspondingly adjust the weights of accuracy and response time, re-evaluate and select a suitable model to ensure that the system always fits the actual needs of campus business.

[0086] It should be noted that: the calculation formulas and each parameter participating in the operation in the present invention have been pre-processed by dimensionless processing, and the process of dimensionless processing is well-known in the industry and will not be described here.

[0087] The above has described a detailed implementation example of the present invention, but the content described is only the preferred implementation example of the present invention and cannot be considered as used to limit the implementation scope of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the patent coverage scope of the present invention.

Claims

1. A one-stop service system for campus business based on AIoT, characterized in that: include: IoT device integration unit: used to establish connections with smart meters, sensors and campus card terminals to collect energy consumption data, equipment operation status information and crowd density data in real time in campus scenarios; Edge computing unit: deployed at the IoT gateway node, performs local preprocessing on the collected raw data, completes abnormal data identification and real-time response logic execution; AI capability core unit: Integrates a dedicated semantic understanding model for university business scenarios, provides a natural language interaction interface, and supports multimodal fusion analysis of text, images, and time series data; Lightweight AI inference unit: Integrate artificial intelligence models in the IoT gateway to achieve localized fault prediction of equipment operation status, generate fault type labels and maintenance suggestions, and synchronize them to the management backend unit; Dynamic model management unit: Through the model scheduling algorithm, it realizes the dynamic loading and switching of various artificial intelligence models according to the business load and model performance indicators, and provides an extension interface for the open source framework.

2. According to the AIoT-based campus business one-stop service system according to claim 1, it is characterized in that: The IoT device integration unit includes a face recognition terminal device, which is used to work with the authority control subunit in canteen consumption and venue reservation scenarios to perform real-name identity authentication and authority management.

3. According to claim 2, the one-stop service system for campus business based on AIoT is characterized in that: The edge computing unit synchronizes the pre-processed device status data to the management background in real time through the distributed log monitoring system and abnormal warning mechanism, and generates a visual data interface including user activity statistics and energy consumption analysis reports.

4. According to claim 3, the one-stop service system for campus business based on AIoT is characterized in that: The AI ​​capability core units include: Multilingual interaction subunit: Based on pre-trained language models of different national languages ​​and dialects from different regions, it realizes the intention recognition of natural language input and the generation of multilingual responses; Context-aware engine: associates user historical interaction records through session IDs and uses an attention mechanism to achieve contextual logic coherence for multiple rounds of conversations.

5. The one-stop campus business service system based on AIoT according to claim 1 or 3 is characterized in that: The system also integrates robotic process automation technology to achieve cross-platform data automatic synchronization and transaction linkage processing of the educational management system, student work system and financial system.

6. According to claim 1, the one-stop service system for campus business based on AIoT is characterized in that: The management backend unit includes: Permission control subunit: Based on the role access control model, define the role permission mapping table for students, faculty and administrators, and implement user identity authentication through the OAuth2.0 protocol; Audit log subunit: records user operation behaviors and system events, and supports log retrieval based on timestamps and operation types; Data synchronization interface: synchronize user basic data with the academic affairs system and student affairs system in real time through GraphQLAPI to ensure dynamic update of permission strategies.

7. The one-stop campus business service system based on AIoT according to claim 1 is characterized in that: The model scheduling algorithm of the dynamic model management unit includes: Monitor business load in real time and calculate business load Assess business pressures when Exceeding the threshold When the scheduling is triggered, is the number of business requests, is the average processing time, , is the preset weight coefficient and ; Evaluate model performance indicators, including accuracy , response time R and resource utilization ; Select models and model performance indicators based on business types, realize dynamic model loading and smooth switching, load new models according to storage paths and interface specifications, verify compatibility after loading, and switch according to the proportion Gradually migrate business, is the number of diversion requests, is the total number of requests; Adopt the model cache and elimination strategy based on the least recently used LRU principle, maintain the model access time record queue, and eliminate the last access time when the cache is insufficient The smallest model; The process of selecting a model based on business type and model performance indicators is as follows: Comprehensive indicators for response time-sensitive business computing Select The model with the smallest value is selected for accuracy-sensitive businesses. The highest model; is the response time weight, For balanced business, a balanced evaluation model is constructed, and the balanced evaluation index is obtained through the indicator balanced evaluation model. , and select A model with a large value.

8. The one-stop campus business service system based on AIoT according to claim 7 is characterized in that: The process of constructing the indicator balanced evaluation model is as follows: S1. Identify key metrics for evaluating model performance, including accuracy, response time, and resource utilization; S2. Determine the corresponding weight coefficient according to the importance of each indicator to the business; among which, the accuracy weight is , the response time weight is , resource utilization weight is , and satisfies ; S3: Standardize the response time and resource utilization. , standardized resource utilization ; S4. Constructing balanced evaluation indicators ; The expression is: ; in, is the warning value of accuracy, is the warning value of the standardized response time, is the warning value of the standardized resource utilization rate, , , is the preset coefficient, is the influence coefficient.

Images