Automatic routing inspection and monitoring alarm method for central machine room in bank

Abstract

The invention discloses an automatic routing inspection and monitoring alarm method for a central machine room in a bank, and the method comprises the steps: formulating a routing inspection plan based on the importance of equipment, collecting the data of the machine room through a routing inspection robot and a sensor network, and carrying out the anomaly detection through a machine learning algorithm with the historical data of the equipment in the next month as a dynamic baseline. The alarms are subjected to grading treatment, work order automatic circulation is achieved through an ITSM platform, and finally fault repairing and knowledge base updating are completed. 24-hour automatic inspection of the machine room is realized, the downtime probability of the equipment is reduced, the service continuity is improved, the operation and maintenance cost is reduced, the service life of the equipment is prolonged, and an effective scheme is provided for intelligent operation and maintenance of the bank machine room.

Description

Technical Field

[0001] This invention relates to the fields of computer technology, Internet of Things and artificial intelligence, and specifically to a full-process method for automated inspection, anomaly detection, hierarchical alarm and closed-loop handling in a bank's central computer room. It is applicable to intelligent operation and maintenance scenarios of key facilities such as bank central computer rooms and data centers. Background Technology

[0002] With the deepening of the digital transformation of banking business, the central computer room, as the core of business operations, has expanded its equipment scale to thousands of units (including UPS, precision air conditioners, server clusters, power distribution cabinets, etc.). Environmental parameters (temperature, humidity, power load) need to be maintained within a high-precision range of ±0.3℃, ±2%RH, and ±1% load fluctuation.

[0003] Traditional manual inspection methods have three major drawbacks: 1) Low efficiency: Manual inspections are conducted only twice daily, covering less than 30% of the data center per inspection, with data collection intervals for critical equipment reaching up to 12 hours; 2) Delayed response: Fault detection relies on manual troubleshooting, with an average response time exceeding 60 minutes. A 3-hour business interruption occurred due to the failure to detect UPS battery leakage in a timely manner; 3) High subjectivity: Equipment status judgment depends on the experience of maintenance personnel, resulting in a 25% failure rate to detect hidden anomalies such as excessive temperature and humidity. Meanwhile, financial regulators require banks' data centers to maintain 99.99% business continuity throughout the year (annual downtime ≤ 53 minutes). Traditional methods can no longer meet these requirements, necessitating the construction of an automated and intelligent operation and maintenance system. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for automated inspection and monitoring alarms of a bank's central computer room. By deploying automated inspection equipment, constructing an intelligent anomaly detection model, and establishing a hierarchical alarm closed-loop process, the method enables 24-hour inspection of the bank's central computer room, accurate anomaly identification, and rapid handling, thereby improving the continuity of core business to over 99.99%, while reducing operation and maintenance costs, extending equipment lifespan, and optimizing data room energy consumption.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0006] This invention discloses a method for automated inspection, monitoring, and alarm reporting in a bank's central computer room, comprising the following steps:

[0007] S1. Start the inspection process and determine the inspection frequency based on the importance of the equipment to be monitored. The inspection includes manual inspection and automatic inspection. Automatic inspection is executed 24 hours a day without interruption, and manual inspection is performed once a month.

[0008] S2. Develop an inspection plan; the inspection plan includes equipment status, environmental parameters, and safety facilities, where equipment status includes UPS, air conditioning, batteries, water leakage, and power supply status; environmental parameters include temperature and humidity, and smoke alarms; and safety facilities include access control and fire protection.

[0009] S3. Deploy an automated inspection robot and sensor network. The automated inspection robot collects data from the computer room in real time through lidar and various types of sensors, and works collaboratively through a multi-machine management platform.

[0010] S4. Record the inspection results and integrate the collected heterogeneous data sources into the system. The heterogeneous data sources include UPS logs, air conditioning control panel records, and temperature and humidity data.

[0011] S5. Call the DeepSeek big model to identify anomalies in the collected data: Based on the device's historical data of the past month, the DeepSeek big model learns the device's operating rules and outputs a dynamic baseline threshold percentage. The threshold percentage can be adjusted by the administrator, with an adjustment range of 5%-20%. The DeepSeek big model combines time series analysis and rule association mining to compare and analyze the real-time data with the dynamic baseline. When the data fluctuation exceeds the threshold, an alarm is triggered.

[0012] S6. If an anomaly is determined, generate alarm classification results and automated handling suggestions through the DeepSeek big model, automatically process and report to the service desk, triggering the alarm handling process; if no anomaly is determined, generate an inspection report and archive it, and return to step S2 to redo the inspection plan.

[0013] S7. In the alarm handling process, alarms are classified based on the classification results output by the DeepSeek large model: Emergency alarms: including smoke, UPS input power failure, triggering personnel evacuation and fire-fighting procedures; Important alarms: including high temperature, communication interruption, generating work orders and dispatching them to engineers.

[0014] S8. Verify the authenticity of the alarm by having back-end personnel or automated inspection robots perform a secondary verification of the alarm data, which includes temperature and humidity data;

[0015] S9. If an anomaly is confirmed, perform equipment repair or replacement, verify the repair results, update the record, and archive it; if no anomaly is confirmed, mark it as a false alarm and archive it.

[0016] S10. The entire alarm handling process data is aggregated into the knowledge base. The DeepSeek big data model is used to perform semantic analysis, tag classification and similar case association on the entire process data to complete the intelligent update and closure of the knowledge base. The entire process data includes the cause of the fault and the handling measures.

[0017] Preferably, the behavior anomaly identification algorithm in step S5 includes time series analysis and rule association mining. Specifically, it uses the DeepSeek large model to construct a dynamic model of the mutual influence between devices, and the model expression is:

[0018]

[0019] Where R i (t) represents the risk value of device i at time t, α and β are weighting coefficients and α + β = 1, W ij D represents the association weights between device i and device j obtained by the DeepSeek large model based on historical operation and maintenance data. j (t) represents the real-time data of device j at time t, B i (t) represents the dynamic baseline value of device i at time t, output by the DeepSeek large model.

[0020] Preferably, the automatic inspection robot described in step S3 supports multiple obstacle avoidance functions with an obstacle avoidance response time of ≤0.5 seconds. It also has an automatic fire extinguishing device triggering function. The DeepSeek large model can generate the optimal inspection route in real time based on the location of the automatic inspection robot and the alarm priority, and distribute it to the multi-machine management platform.

[0021] Preferably, the integration format of the heterogeneous data sources in step S6 is JSON, the data transmission delay is ≤1 second, and the DeepSeek large model performs unified vector representation processing on the integrated heterogeneous data sources to realize the fusion analysis of structured and unstructured data.

[0022] Preferably, the work order mentioned in step S7 is automatically created through the bank's ITSM platform, with a work order dispatch delay of ≤30 seconds. The DeepSeek big model can optimize the alarm classification rules based on the financial industry's regulatory requirements and the importance weight of the equipment, refining the alarm levels into four levels: emergency, important, general, and alert.

[0023] Preferably, the dynamic baseline in step S5 is updated once a day, and abnormal data is removed during the update. The criterion for judging abnormal data is a deviation from the historical mean by 3 times the standard deviation. The DeepSeek large model can dynamically adjust the baseline update rules based on the characteristics of scenarios such as holidays and peak business periods.

[0024] Preferably, the method for verifying the repair results in step S9 is as follows: the automatic inspection robot re-collects the corresponding equipment data, and if the data is within the dynamic baseline threshold for three consecutive times, the repair is determined to be complete. The DeepSeek big model can generate an equipment health assessment report based on the repaired data.

[0025] Preferably, step S3 also includes a mobile inspection system, which includes an inspection data acquisition module and an inspection data processing module. The inspection data acquisition module supports real-time data recording on mobile phones and tablets. After the data is entered, the inspection data processing module automatically synchronizes it to the database. The DeepSeek big model can perform semantic extraction and preliminary anomaly judgment on the text data entered by the mobile inspection.

[0026] Preferably, the knowledge base in step S10 is updated in real time after each alarm is processed. The knowledge base supports keyword retrieval, with a retrieval response time of ≤2 seconds. The DeepSeek large model can periodically analyze the knowledge base data and output operation and maintenance strategy optimization suggestions.

[0027] An apparatus for implementing the method according to the present invention includes:

[0028] Data acquisition module: Composed of an automated inspection robot and a sensor network, used to collect data from the computer room;

[0029] Model access module: Establishes API communication connection with DeepSeek large models to realize data interaction and receive model inference results;

[0030] Data analysis module: Deploy machine learning algorithms and combine them with the output of the DeepSeek large model for anomaly detection;

[0031] Alarm handling module: Implements alarm classification and work order dispatch based on the DeepSeek large model classification results; Data storage module: Stores inspection data, alarm records and knowledge base;

[0032] Interaction module: Supports human-computer interaction on mobile terminals and management platforms;

[0033] The data acquisition module, data analysis module, alarm handling module, data storage module, and interaction module are interconnected.

[0034] Beneficial effects:

[0035] Technological advantages: Real-time monitoring and accurate early warning: By building a 24-hour data collection network through fixed sensors and inspection robots, combined with edge computing and dynamic baseline models, potential risks before failure can be identified, reducing the probability of equipment downtime by more than 40%, achieving an anomaly detection accuracy of 95%, and a missed detection rate of ≤5%.

[0036] Intelligent anomaly diagnosis: Based on time series analysis and equipment correlation model, heterogeneous data is integrated to establish dynamic influence relationships between devices, reducing the false alarm rate by 60%. It can accurately distinguish the root causes of faults such as hardware aging and software errors, providing data support for differentiated maintenance.

[0037] Enhancing Business Value and Business Continuity: During peak banking periods, overload risks are mitigated through load forecasting and automatic adjustments (such as backup power activation and load redistribution), improving the availability of core business systems to over 99.99% and reducing annual downtime to within 53 minutes.

[0038] Optimize operation and maintenance processes: Alarm information is seamlessly integrated with the ITSM platform, realizing full automation of work order creation, dispatch, and tracking. Engineer response time has been reduced from 60 minutes to within 15 minutes, and work order processing efficiency has been improved by 50%.

[0039] Economic benefits include reduced operation and maintenance costs: automated inspections replace manual routine inspections, reducing labor costs by 30% annually; root cause analysis functions reduce reliance on senior experts, allowing intermediate engineers to handle more than 80% of faults; predictive maintenance extends the lifespan of major equipment by 20-30%.

[0040] Energy savings: The intelligent temperature control system dynamically adjusts the air conditioning power, combined with the hot aisle enclosure design, reducing the PUE value of the computer room from 1.5 to below 1.3, saving more than 15% in annual energy consumption. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the process of this invention.

[0042] Figure 2 This is a schematic diagram of the alarm process of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Invention Technology Principle / Content: This invention provides a method for automated inspection and monitoring alarms in a bank's central computer room, achieving full-process intelligence through a four-layer architecture of equipment layer, perception layer, analysis layer, and processing layer.

[0045] Two types of equipment are deployed at the equipment layer. Fixed sensing equipment includes temperature and humidity sensors (3 per area, deployed at equipment air inlets, air outlets, and in the middle of passageways) distributed in 12 areas of the computer room, 24 current sensors (deployed at each output terminal of the power distribution cabinet), and 8 sets of smoke sensors (deployed on the top of the computer room and in densely populated equipment areas); Mobile inspection equipment includes 3 automatic inspection robots, each equipped with LiDAR, infrared thermal imager, and acoustic sensor, with a cruising speed of 0.5m / s and a coverage radius of 15 meters.

[0046] The perception layer enables real-time data acquisition and transmission. Fixed equipment data is acquired at a frequency of 1 second and uploaded to the edge computing node via wired Ethernet (transmission rate 1000Mbps); robot data is acquired at a frequency of 0.5 seconds using infrared thermal imaging (resolution 640×512) and equipment noise (sampling rate 44.1kHz) data, and uploaded via 5G network (latency ≤20ms).

[0047] The analysis layer deploys two types of algorithms. The dynamic baseline algorithm uses the device's historical data from the past 30 days as a sample, calculating the mean and standard deviation through a sliding window (24 hours) to generate a dynamic baseline (e.g., a UPS current baseline of "50A±5A"). The anomaly detection algorithm uses the dynamic model described in claim 2, combined with a time-series ARIMA model to predict data trends. When real-time data simultaneously meets the criteria of "exceeding the dynamic baseline threshold" and "the predicted trend deviating from the normal range," it is determined to be an anomaly.

[0048] The handling layer implements hierarchical alarms and automated workflow: Alarm classification rules: based on the scope of fault impact and urgency, it is divided into 3 levels (urgent, important, and general); Work order workflow: it connects to the API interface of the internal ITSM platform to realize the mapping of alarm information and work order system fields (such as alarm ID corresponding to work order ID, abnormal equipment corresponding to work order associated assets).

[0049] This invention enables 24 / 7 inspection of bank central computer rooms, reducing the data collection interval from 12 hours to less than 1 second; increasing the accuracy of anomaly detection to over 95% and reducing the missed detection rate to below 5%; reducing fault response time from 60 minutes to within 15 minutes, achieving 99.99% continuity of core business; reducing the PUE value of the computer room from 1.5 to below 1.3; and extending the average lifespan of equipment by more than 20%.

[0050] Example description:

[0051] The electrical components and connection methods of the automatic inspection robot in this invention:

[0052] The core structure and electrical components are categorized as follows:

[0053] 1. Sensing Unit (Core of Data Acquisition)

[0054] LiDAR (e.g., SICKTIM571): Used for positioning and obstacle avoidance, collecting spatial coordinate data of the computer room; Infrared thermal imager (e.g., FLIRC2): Detects the surface temperature of equipment and captures thermal anomalies; Acoustic sensor (e.g., SGM3770): Collects equipment operating noise and identifies abnormal noise frequencies; Multi-parameter sensor group: Includes temperature and humidity sensor (e.g., SHT30), current sensor (e.g., ACS712), and voltage sensor (e.g., LV25-P), collecting environmental and equipment electrical parameters; High-definition camera (e.g., OV5640): Captures equipment indicator lights and panel display information.

[0055] 2. Motion control unit (walking and posture control)

[0056] Main controller (e.g., STM32H743): The core control center of the robot; Drive module (e.g., L298N motor drive board): Receives signals from the main controller and drives the walking motors; Walking motors (e.g., 60BLDC): 2 units, driving the left and right wheels of the robot respectively; Encoder (e.g., E6B2-CWZ6C): Coaxially connected to the walking motors, providing feedback on rotational speed to achieve closed-loop speed control; Gyroscope (e.g., MPU6050): Detects the robot's posture and assists in navigation and positioning.

[0057] 3. Data Processing and Transmission Unit (Data Processing and Interaction)

[0058] Edge computing module (e.g., NVIDIA Jetson Nano): preprocesses sensing data and filters invalid information; Communication module: 5G module (e.g., SIM7600) + Ethernet module (e.g., W5500) to realize data uploading and command reception; Storage module (e.g., TF card 128GB): temporarily stores locally collected data.

[0059] 4. Auxiliary Functional Units (Emergency and Operational Maintenance Assistance)

[0060] Automatic fire extinguishing trigger module (e.g., model: DS-2CD3T46DWD-I3 linkage module): receives alarm signals to trigger the fire extinguishing device; Charging management module (e.g., model: TP4056): controls the charging logic and connects to the charging base station; Backup battery pack (e.g., model: 18650 lithium battery pack 12V / 10Ah): provides power to the whole unit; Buzzer (model: SMD0905): provides local audible and visual alerts in case of abnormality.

[0061] In addition, the connection methods for electrical components include the following:

[0062] 1. The sensing unit connects to the main controller via the following components: LiDAR: connected to the main controller's UART1 interface via RS485 bus, with a baud rate of 9600bps, transmitting positioning data in real time; Infrared thermal imager / HD camera: connected to the edge computing module via USB2.0 interface, transmitting pre-processed data to the main controller via SPI interface; Acoustic sensor / multi-parameter sensor group: connected to the main controller's UART2 interface via I2C bus, with a sampling frequency of 1Hz, synchronously transmitting data such as temperature, humidity, and current.

[0063] 2. The motion control unit is internally connected to the main controller: it outputs PWM signals to the L298N driver board through the GPIO interface to control the forward and reverse rotation and speed of the walking motor; the encoder is connected to the TIM1 timer of the main controller through the AB phase pulse interface to feed back the motor speed and realize speed closed loop; the gyroscope is connected to the UART3 interface of the main controller through the I2C bus to output attitude data to assist navigation.

[0064] 3. Data processing and transmission unit connects to the edge computing module: It communicates with the main controller through the PCIe interface, receives raw data and feeds back preprocessing results; Communication module: The 5G module accesses the network through the SIM card slot, and the Ethernet module connects to the data center LAN through the RJ45 interface. Both communicate bidirectionally with the main controller through the UART4 interface; Storage module: It connects to the main controller through the SPI interface to store local data backup.

[0065] 4. Auxiliary Unit and Core Unit Connection: Automatic Fire Extinguishing Trigger Module: Connected to the GPIO1 pin of the main controller via a relay interface, triggering upon receiving an alarm signal; Charging Management Module: Connected to the backup battery pack via a Type-C interface, and simultaneously transmitting battery voltage data to the main controller via an ADC interface; Buzzer: Driven by an NPN transistor, controlled by the GPIO2 pin of the main controller.

[0066] 5. The backup battery pack (12V) of the power supply link outputs 5V (for the sensing unit and communication module), 3.3V (for the main controller and edge computing module), and 24V (for the walking motor) respectively through a DC-DC converter. Each link is connected in series with fuses (2A / 5A) to achieve overcurrent protection.

[0067] It should be noted that the automatic inspection robot's travel device in this application can adopt existing technologies such as electrically driven wheeled or tracked types, without specific limitations.

[0068] The ITSM (IT Service Management) platform in this invention is an IT service management system built based on ITIL (IT Infrastructure Library) best practices. Its core function is to standardize and automate IT service processes, covering the entire lifecycle management from fault reporting and work order processing to service review. It serves as a core support system for the IT operations and maintenance systems of enterprises such as banks. In banking scenarios, the ITSM platform typically interfaces with various internal IT systems (such as data center monitoring systems and server management systems) to achieve unified scheduling and management of operational events.

[0069] In this invention, the multi-machine management platform serves as the core system for controlling multiple automated inspection robots. Its components can be divided into two parts: software functional modules and hardware deployment carriers, as detailed below:

[0070] The robot status monitoring module collects and displays real-time data on the location, remaining battery power, sensor operating status, and cruising speed of each inspection robot. It supports real-time alarms for abnormal states (such as low battery or obstacle avoidance failure) and serves as the platform's "perception center" for the robots. The task scheduling and path planning module can automatically allocate inspection areas and tasks according to the inspection plan using built-in algorithms. In case of an alarm, it can dynamically adjust robot routes, dispatching the nearest robot to the abnormal location to enable multi-robot collaborative work. The data interaction and preprocessing module receives heterogeneous data such as infrared images and temperature / humidity data uploaded by the robots. After standardizing the format and filtering invalid data, it synchronizes the data to the intelligent management platform in the data center. Simultaneously, it sends control commands from the upper-level system (such as restarting sensors or adjusting inspection frequency) to the corresponding robots. The remote control module provides a manual robot operation interface, allowing engineers to remotely control robot movement, initiate special data collection, trigger fire extinguishing devices, and perform batch remote firmware upgrades. The task and report management module supports the creation, start, and stop of preset / temporary inspection tasks, automatically records task completion status, and generates reports such as inspection coverage and data collection completeness, achieving standardized management of inspection work. The interface adaptation module provides RESTful API and MQTT protocol interfaces to connect with the intelligent data center management platform and ITSM platform, ensuring data interoperability between systems. Servers: Deployed locally in the data center or at cloud edge nodes, using high-performance industrial servers (such as Huawei RH2288HV5) to host platform software operation and data storage; Visualization terminals: Includes a large data center monitoring screen and engineer operation terminals (computers / tablets) to display robot status and inspection data, supporting human-machine interaction; Communication equipment: 5G base stations and industrial switches to ensure wireless / wired communication between the platform and the robot, ensuring real-time data transmission.

[0071] Example 1: Specific Execution of Automated Inspection Process

[0072] Inspection plan configuration: Select "Core Equipment Group" (including UPS and core switch) and set the automatic inspection frequency to "once every 10 minutes"; Select "Auxiliary Equipment Group" (including air conditioner and power distribution cabinet) and set the automatic inspection frequency to "once every 30 minutes"; Configure inspection content: For UPS, collect "input voltage, output current, and battery temperature"; For air conditioner, collect "return air temperature, supply air temperature, and compressor power"; For power distribution cabinet, collect "output current of each circuit and total load rate".

[0073] Data Acquisition Execution: Fixed Sensors: The UPS input voltage sensor collects data every 1 second, with the data format "Device ID: UPS-001; Acquisition Time Setting; Input Voltage: 220V", and transmits it to the edge node via Ethernet; Inspection Robot: The robot patrols along a preset route (covering 12 areas of the server room), stops at the UPS-001 location for 5 seconds, collects the battery temperature (data is "35℃") using an infrared thermal imager, and collects the equipment operating noise (data is "55dB") using an acoustic sensor, and uploads it to the system simultaneously.

[0074] Data storage: Edge nodes convert the collected raw data into JSON format (example:

[0075] {"device_id":"UPS-001","time":"2025-11-2510:00:00","voltage":220,"battery_temp":35}) are stored in a distributed database with a storage period of 365 days.

[0076] Example 2: Specific Implementation of Anomaly Detection and Alarm Handling

[0077] Dynamic baseline generation: The system retrieves the input voltage data of UPS-001 for the past 30 days, calculates the mean to be 220V and the standard deviation to be 2V, and sets the threshold percentage to 10%. Therefore, the dynamic baseline is "220V ± 2.2V".

[0078] (i.e., 217.8V-222.2V).

[0079] Anomaly Detection: At 10:10:00 on November 24, 2025, the input voltage of UPS-001 was 215V. The system calculated the risk value using a dynamic model.

[0080] R UPS-001 (10:10:10)=0.6×(W UPS-空调 ×D 空调 (t)+W UPS-配电柜 ×D 配电柜 (t))+0.4×220;

[0081] in,

[0082] W UPS-空调 =0.3; D 空调 (t) = 25℃W UPS-配电柜 =0.2; D 配电柜 (t) = 80, and R is calculated. UPS-001 =216V, exceeding the lower limit of the dynamic baseline, and is judged as abnormal.

[0083] Alarm classification and dispatch: The system determines that the anomaly is an "important alarm" (input voltage anomaly), automatically calls the work order creation interface of the ITSM platform to generate a work order (work order content: "Equipment UPS-001 input voltage anomaly, current value 215V, baseline range 217.8V-222.2V"), ​​and dispatches it to the mobile terminal of the on-duty engineer.

[0084] Verification and Closure: After receiving the work order, the engineer arrived at the site at 10:15:00 and retested the UPS-001 input voltage with a multimeter, finding it to be 215V, confirming the abnormality. The engineer then performed the "switch to backup power" operation, completing the operation at 10:20:00. The system then used a robot to re-collect the UPS-001 input voltage, which was 220V. With three consecutive collections of values ​​within the baseline range, the repair was deemed complete. The engineer entered "Cause of Fault: Mains Power Fluctuation; Handling Measures: Switch to Backup Power" into the system, and the system synchronized this record to the knowledge base, completing the closure.

[0085] Finally, it should be noted that the present invention is not limited to the above embodiments, and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for automated inspection, monitoring, and alarm reporting in a bank's central computer room, characterized in that, Includes the following steps: S1. Start the inspection process and determine the inspection frequency based on the importance of the equipment to be monitored. The inspection includes manual inspection and automatic inspection. Automatic inspection is executed 24 hours a day without interruption, and manual inspection is performed once a month. S2. Develop an inspection plan; the inspection plan includes equipment status, environmental parameters, and safety facilities, where equipment status includes UPS, air conditioning, batteries, water leakage, and power supply status; environmental parameters include temperature and humidity, and smoke alarms; and safety facilities include access control and fire protection. S3. Deploy an automated inspection robot and sensor network. The automated inspection robot collects data from the computer room in real time through lidar and various types of sensors, and works collaboratively through a multi-machine management platform. S4. Record the inspection results and integrate the collected heterogeneous data sources into the database. The heterogeneous data sources include UPS logs, air conditioning control panel records, and temperature and humidity data. S5. Call the DeepSeek big model to identify anomalies in the collected data: Based on the device's historical data of the past month, the DeepSeek big model learns the device's operating rules and outputs a dynamic baseline threshold percentage. The threshold percentage can be adjusted by the administrator, with an adjustment range of 5%-20%. The DeepSeek big model combines time series analysis and rule association mining to compare and analyze the real-time data with the dynamic baseline. When the data fluctuation exceeds the threshold, an alarm is triggered. S6. If an anomaly is determined, generate alarm classification results and automated handling suggestions through the DeepSeek big model, automatically process and report to the service desk, triggering the alarm handling process; If no abnormalities are found, generate an inspection report and archive it, then return to step S2 to redo the inspection plan. S7. In the alarm handling process, alarms are classified based on the classification results output by the DeepSeek large model: Emergency alarms: including smoke, UPS input power failure, triggering personnel evacuation and fire-fighting procedures; Important alarms: including high temperature, communication interruption, generating work orders and dispatching them to engineers. S8. Verify the authenticity of the alarm. The alarm data is verified a second time by the back-end operator or the automatic inspection robot. The alarm data includes temperature and humidity data. S9. If an abnormality is confirmed, perform equipment repair or replacement, verify the repair results, update the record, and archive it. If no abnormality is confirmed, mark it as a false alarm and archive it; S10. The entire alarm handling process data is aggregated into the knowledge base. The DeepSeek big data model is used to perform semantic analysis, tag classification and similar case association on the entire process data to complete the intelligent update and closure of the knowledge base. The entire process data includes the cause of the fault and the handling measures.

2. The method for automated inspection and monitoring alarm of a bank's central computer room according to claim 1, characterized in that, Step S5, the behavior anomaly identification algorithm, includes time series analysis and rule association mining. Specifically, it uses the DeepSeek large model to construct a dynamic model of the mutual influence between devices. The model expression is: Where R i (t) represents the risk value of device i at time t, α and β are weighting coefficients and α + β = 1, W ij D represents the association weights between device i and device j obtained by the DeepSeek large model based on historical operation and maintenance data. j (t) represents the real-time data of device j at time t, B i (t) represents the dynamic baseline value of device i at time t, output by the DeepSeek large model.

3. A method for automated inspection, monitoring, and alarm reporting in a bank's central computer room according to claim 1 or 2, characterized in that, The obstacle avoidance response time of the automatic inspection robot described in step S3 is ≤0.5 seconds. The DeepSeek large model can generate the optimal inspection route in real time based on the position of the automatic inspection robot and the alarm priority, and then distribute it to the multi-machine management platform.

4. The method for automated inspection and monitoring alarm of a bank's central computer room according to claim 1, characterized in that, The heterogeneous data source integration format in step S6 is JSON, with a data transmission delay of ≤1 second. The DeepSeek large model performs unified vector representation processing on the integrated heterogeneous data source to achieve the fusion analysis of structured and unstructured data.

5. The method for automated inspection, monitoring, and alarm reporting of a bank's central computer room according to claim 1, characterized in that, The work orders mentioned in step S7 are automatically created through the bank's internal ITSM platform. The work order dispatch delay is ≤30 seconds. The DeepSeek big model can optimize the alarm classification rules based on the financial industry regulatory requirements and the importance weight of the equipment, and refine the alarm levels into four levels: emergency, important, general, and alert.

6. The method for automated inspection and monitoring alarm of a bank's central computer room according to claim 1, characterized in that, The dynamic baseline mentioned in step S5 is updated once a day. Abnormal data is removed during the update. The criterion for judging abnormal data is a deviation from the historical mean by 3 times the standard deviation. The DeepSeek large model can dynamically adjust the baseline update rules based on the characteristics of scenarios such as holidays and peak business periods.

7. The method for automated inspection and monitoring alarm of a bank's central computer room according to claim 2, characterized in that, The method for verifying the repair results in step S9 is as follows: the automatic inspection robot re-collects the corresponding equipment data, and if the data is within the dynamic baseline threshold for three consecutive times, the repair is determined to be complete. The DeepSeek large model can generate an equipment health assessment report based on the repaired data.

8. The method for automated inspection and monitoring alarm of a bank's central computer room according to claim 1, characterized in that, Step S3 also includes a mobile inspection system, which includes an inspection data acquisition module and an inspection data processing module. The inspection data acquisition module supports real-time data recording on mobile phones and tablets. After the data is entered, the inspection data processing module automatically synchronizes it to the database. The DeepSeek large model can perform semantic extraction and preliminary anomaly judgment on the text data entered by the mobile inspection.

9. The method for automated inspection, monitoring, and alarm reporting of a bank's central computer room according to claim 1, characterized in that, The knowledge base mentioned in step S10 is updated in real time after each alarm is processed. The knowledge base supports keyword retrieval, and the retrieval response time is ≤2 seconds. The DeepSeek large model can periodically analyze the knowledge base data and output operation and maintenance strategy optimization suggestions.

10. An apparatus for implementing the method according to any one of claims 1-9, characterized in that, include: Data acquisition module: Composed of an automated inspection robot and a sensor network, used to collect data from the computer room; Model access module: Establishes API communication connection with DeepSeek large models to realize data interaction and receive model inference results; Data analysis module: Deploy machine learning algorithms and combine them with the output of the DeepSeek large model for anomaly detection; Alarm handling module: Implements alarm classification and work order dispatch based on the classification results of the DeepSeek large model; Data storage module: Stores inspection data, alarm records, and a knowledge base; Interaction module: Supports human-computer interaction on mobile terminals and management platforms; The data acquisition module, data analysis module, alarm handling module, data storage module, and interaction module are interconnected.

Images