Container type submerged liquid cooling BMC centralized management device and method

Abstract

The application relates to the technical field of immersed liquid cooling BMC centralized management and control, and belongs to a container type immersed liquid cooling centralized management and control device which comprises a container body, a liquid cooling subsystem arranged in the container body, a power supply subsystem arranged in the container body, and a network exchange subsystem arranged in the container body. The liquid cooling subsystem comprises a cooling tower, at least one cooling capacity distribution unit and at least one Tank unit, the cooling capacity distribution unit and the Tank unit are arranged in one-to-one correspondence, the power supply subsystem comprises a power distribution cabinet and a high-voltage direct-current power supply (HVDC), and the network exchange subsystem has the following beneficial effects: centralized management and control and improved manageability. A centralized management and control module with the BMC as the core realizes centralized monitoring and management of various components in the container, an operator can centrally acquire the running states of various components through the BMC, unified management and maintenance are facilitated, and the manageability of the system is improved.

Description

Technical Field

[0001] This invention relates to the field of centralized control technology for immersion liquid-cooled BMCs, and specifically to a containerized immersion liquid-cooled BMC centralized control device and method. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence, high-performance computing, and edge computing, the demand for high-density, high-power computing units is increasing daily. Traditional air-cooling methods become inefficient, energy-intensive, and cause significant noise pollution once the power density exceeds a certain threshold. Immersion liquid cooling technology, by directly immersing heat-generating components in insulating coolant, leverages the liquid's extremely high specific heat capacity and thermal conductivity to achieve revolutionary heat dissipation performance, becoming a key technology for solving the heat dissipation problem of high-power chips. Containerized data centers, due to their modularity, rapid deployment, and mobility, offer significant advantages in edge computing, temporary expansion, and military applications. However, integrating a complete immersion liquid cooling system (including cooling towers, cooling capacity distribution units, liquid cooling tanks, power distribution, refrigeration, and monitoring) into a standard container introduces unprecedented complexity in management and control. Existing solutions typically employ decentralized, independent control systems (such as cooling towers with integrated PLCs, power distribution cabinets with integrated meter management systems, and network equipment with integrated network management systems), lacking effective data communication and intelligent collaboration between subsystems. This causes the system to operate in a locally optimal rather than globally optimal state, and its energy efficiency potential cannot be fully realized. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a containerized immersion liquid-cooled BMC centralized control device and method.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A containerized immersion liquid cooling centralized control device includes a container body; a liquid cooling subsystem disposed within the container body, the liquid cooling subsystem including: a cooling tower, at least one cooling capacity distribution unit and at least one Tank unit, the cooling capacity distribution unit and the Tank unit being arranged in a one-to-one correspondence; an electronic power supply system disposed within the container body, the electronic power supply system including a power distribution cabinet and a high-voltage direct current power supply (HVDC); a network switching subsystem disposed within the container body, the network switching subsystem including a switch cabinet and a switch disposed therein; and a baseboard management controller (BMC) centralized control module disposed within the container body.

[0005] Preferably, the centralized control module of the baseboard management controller includes a multi-protocol adaptation layer, including a processor, a memory, a multi-protocol communication interface, a local control database, and a control strategy library. The local control database includes configuration data, which includes: a preset normal value range, a warning threshold, an alarm threshold, and a set of correlation weights representing the degree of influence between parameters for each monitored parameter. The control strategy library includes predefined control strategies for different anomaly types and root cause combinations, and each control strategy contains at least one active action instruction template.

[0006] Preferably, the multi-protocol adaptation layer includes: an I2C interface, an SPI interface, and a CAN interface connected to the cooling capacity distribution unit; an RS-485 interface following the Modbus-RTU protocol connected to the cooling tower; an RS-485 interface connected to the Tank unit and the power distribution cabinet; an Ethernet interface connected to the high-voltage DC power supply and establishing a transmission control protocol communication connection; and an Ethernet interface following the SNMP protocol connected to the switch.

[0007] Preferably, the method is characterized by including the following steps: S1. Parameter Acquisition and Communication Verification: Through the multi-protocol adaptation layer, raw data frames of each monitoring parameter are periodically acquired from the cooling tower, cooling capacity distribution unit, Tank unit, power distribution cabinet, high-voltage DC power supply and switch. For each raw data frame, verification is performed according to its communication protocol specification, and a set of valid parameter values ​​that pass the verification is output. For data that fails the verification, the valid parameter values ​​corresponding to the previous cycle are used for replacement. S2. Data Fusion and Anomaly Detection: Perform digital filtering on the numerical parameters in the set of effective parameter values ​​to obtain a set of steady-state parameter values; compare each steady-state parameter value in the set of steady-state parameter values ​​with the corresponding alarm threshold read from the local control database; when any steady-state parameter value reaches or exceeds its alarm threshold, determine that an alarm-level abnormal event has occurred, generate and output an abnormal event identifier, which at least includes the abnormal parameter name and the event type; S3. Correlation Analysis and Root Cause Locator: Read the list of pre-associated correlation parameter names and their corresponding correlation weights from the local control database; obtain the current steady-state parameter value of each parameter in the list of correlation parameter names from the steady-state parameter value set; calculate the abnormal contribution of each correlation parameter; sort the correlation parameters according to their calculated abnormal contribution; and determine the correlation parameter with the highest contribution as the core root cause parameter and output it. S4. Strategy matching and instruction generation: Using the event type in the abnormal event identifier and the core root cause parameter as the joint query key, match the corresponding predefined control strategy in the control strategy library, parse the predefined control strategy, and generate a control instruction sequence by combining it with the current steady-state parameter value set; S5. Command Execution and Closed-Loop Feedback: Through the multi-protocol adaptation layer, the control command sequence is sent to the corresponding target device for execution. Within the set monitoring window after command execution, the execution step S1 is returned to collect new system operating parameters. Based on the new operating parameters, steps S2 to S3 are repeated to obtain a new set of steady-state parameter values ​​and core root cause parameters. By comparing the changes in steady-state parameter values ​​and core root cause parameters before and after command execution, the control effect is evaluated. If the control effect does not meet expectations, the cross-subsystem collaborative control process is triggered.

[0008] Preferably, in step S1, the outlet water temperature of the cooling tower and the pressure parameters of the cooling capacity distribution unit are filtered using a moving average filter, and the coolant temperature in the Tank unit is filtered using a Kalman filter algorithm. The input of the observed values ​​of this algorithm is the temperature sensor sampling value, and the current temperature state prediction input is based on the temperature value output in the previous filtering cycle and the current flow rate value of the cooling capacity distribution unit.

[0009] Preferably, in step S3, the calculation of the anomaly contribution degree specifically involves: for each correlation parameter i, its anomaly contribution degree... ,in The degree of abnormality of the parameter is calculated using the following formula: ,in The current value, These are the boundary values ​​of the normal range. Alarm threshold; The preset association weights for the parameters.

[0010] Preferably, in step S4, when the abnormal event type is: the liquid temperature of the Tank unit reaches a temperature threshold and the core root cause parameter is the flow rate of the corresponding cooling capacity distribution unit, the matched predefined control strategy is to adjust the speed of the variable frequency pump. The generated control command sequence includes: when the liquid temperature of the Tank unit reaches the temperature threshold, generating an I2C format command for adjusting the speed of the variable frequency pump of the corresponding customized cooling capacity distribution unit. The target flow rate parameter in the command is dynamically calculated based on the contribution of the core root cause. The following calculations were performed: ,in, The current flow rate value is read from the current set of steady-state parameter values. The abnormal contribution of the flow parameters calculated in step S3 is given by k, where k is a preset gain coefficient. The value should not exceed the rated maximum value of the equipment.

[0011] Preferably, in step S5, the cross-subsystem collaborative control process includes: S51: When collaborative control is triggered due to network traffic in the switch cabinet reaching the traffic threshold, a first collaborative instruction is sent to the high-voltage DC power supply at the same time as a security policy instruction is sent to the switch. The first collaborative instruction is a request message suggesting adjustment of non-critical load power. S52: When coordinated control is triggered due to the branch current of the distribution cabinet reaching the current threshold, a second coordinated command is sent to the high-voltage DC power supply, and a third coordinated command is sent to the associated Tank unit. The second coordinated command is a power supply optimization request, and the third coordinated command is an instruction to put the energy storage system in the Tank unit into a power supply standby state.

[0012] Preferably, in step S5, the judgment conditions for the evaluation control effect not meeting expectations include: after the monitoring window ends, the abnormal parameter values ​​read from the new steady-state parameter value set do not return to their normal range, or, judging from the new core root cause parameters, the main root cause of the abnormality has not changed.

[0013] Compared with the prior art, the present invention provides a containerized immersion liquid-cooled BMC centralized control device and method, which has the following beneficial effects: 1. Centralized control and improved manageability: The centralized control module with BMC as the core realizes centralized monitoring and management of various components inside the container. Managers can obtain the operating status of each component through BMC, which facilitates unified management and maintenance and improves the manageability of the system.

[0014] 2. Efficient heat dissipation and stable system operation: The immersion liquid cooling technology, combined with the synergistic operation of the cooling tower and CDU, can efficiently remove the heat generated by the computing nodes, ensuring that the computing nodes operate in a suitable temperature environment and improving the stability of the system operation.

[0015] 3. Energy-saving and efficient: The application of high-voltage direct current (HVDC) power supply improves power transmission efficiency and reduces losses; at the same time, the BMC centralized management module can optimize liquid cooling circulation and power supply according to the operating status of each component, further improving system energy efficiency, achieving energy saving, rapid deployment, and high flexibility. In addition, the system is integrated into the container, which facilitates overall transportation and rapid deployment, adapts to computing scenarios of different scales, and has high flexibility.

[0016] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the device of the present invention.

[0018] Figure 2 This is a flowchart of the parameter acquisition process for the Tank unit in this invention.

[0019] Figure 3 This is a logic diagram of the execution steps of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0021] Specific implementation method one: Combining Figures 1 to 3 This embodiment describes a containerized immersion liquid-cooled centralized control device, comprising: a container body; a liquid-cooling subsystem disposed within the container body, the liquid-cooling subsystem including a cooling tower, at least one cooling capacity distribution unit, and at least one Tank unit, the cooling capacity distribution unit and the Tank unit being arranged in a one-to-one correspondence; an electronic power supply system disposed within the container body, the electronic power supply system including a power distribution cabinet and a high-voltage DC power supply; a network switching subsystem disposed within the container body, the network switching subsystem including a switch cabinet and a switch disposed therein; and a baseboard management controller centralized control module disposed within the container body. Each CDU coolant distribution unit measures 0.6m × 0.5m × 0.172m. Container body: Serving as the carrier of the entire system, it provides space for the integrated installation of various components, facilitating overall transportation and rapid deployment; Cooling tower: Located inside the container, it is responsible for cooling the hot water after heat exchange by the CDU, so that it can regain good heat dissipation capacity and provide a cold source for the liquid cooling cycle; Tank unit: Used to house computing nodes. The computing nodes are immersed in coolant, and heat is transferred away through the coolant to achieve efficient heat dissipation. Power distribution cabinet: Provides power to various modules such as cooling tower, CDU, computing nodes, HVDC, and switch cabinet in the container body, ensuring the power needs of each component; HVDC: It adopts high-voltage direct current transmission and is placed inside the container. It takes advantage of the fact that direct current has no inductive reactance and no synchronization problem to improve power transmission efficiency, reduce losses, and achieve energy saving. Switch cabinet: Used in conjunction with an air conditioning cabinet for heat dissipation, it houses the switch and ensures network communication between various components inside the container and with the outside world. It can communicate with the BMC via the SNMP protocol. The BMC centralized control module, serving as the core of the entire system's centralized control, can collect real-time operating parameters of the cooling tower, CDU, Tank units, power distribution cabinet, HVDC, and switch cabinet, such as temperature, flow rate, power consumption, voltage, and current. Based on these parameters, it adjusts and controls each component to ensure stable system operation. Specific Implementation Method Two: Combining Figures 1 to 3 This embodiment describes a centralized control module for the baseboard management controller, which includes a multi-protocol adaptation layer, a processor, a memory, a multi-protocol communication interface, a local control database, and a control strategy library. The local control database includes configuration data, which includes: a preset normal value range, a warning threshold, an alarm threshold, and a set of correlation weights representing the degree of influence between parameters for each monitored parameter. The control strategy library stores predefined control strategies for different anomaly types and root cause combinations, and each control strategy contains at least one active action instruction template. Multi-protocol adaptation layer: This is a software middleware that integrates drivers and parsing libraries for various industrial communication protocols, including Modbus-RTU master library, I2C / SPI driver, CANopen protocol stack, SNMP agent and management terminal library, TCP / IP Socket communication library, etc. It is responsible for converting the heterogeneous hardware interfaces at the lower level into a unified data interface for upper layer calls.

[0022] Local control database: Implemented using an embedded Redis database, this database stores all configuration data, specifically storing the normal_low (lower normal limit), normal_high (upper normal limit), warning (warning threshold), and alarm (alarm threshold) for each monitoring parameter (such as "Tank1_Temp" and "CDU1_Flow"), as well as the associated weights of the parameters. For example, when the "Tank temperature too high" event occurs, the weight of the parameter "CDU flow" is recorded as W=0.8, and the weight of "cooling tower efficiency" is recorded as W=0.7. The database also stores the communication addresses of each device, parameter register mappings, etc.

[0023] Control policy library: stored in JSON format files. Each policy is a mapping rule, for example: { "fault_type":"TankOverTemp", "root_cause":"CduLowFlow", "primary_action":{ "target":"CDU", "action":"adjust_pump", "params_calc":"F_target=F_cur (1+C)" } } Control Engine: This is the core control program running on the processor, which includes, in order: Data acquisition and verification unit: Schedules data acquisition from the multi-protocol adaptation layer at regular intervals and calls the verification program.

[0024] Data analysis and processing unit: Receives valid data, executes filtering algorithms, and queries the database for threshold comparison.

[0025] Control Decision Execution Unit: Executes the root cause analysis algorithm, queries the policy library, and generates the final instruction.

[0026] Specific implementation method three: Combining Figures 1 to 3 This embodiment describes a multi-protocol adaptation layer comprising: an I2C interface, an SPI interface, and a CAN interface connected to the cooling capacity distribution unit; an RS-485 interface conforming to the Modbus-RTU protocol connected to the cooling tower; an RS-485 interface connected to the Tank unit and the power distribution cabinet; an Ethernet interface connected to the high-voltage DC power supply and establishing a transmission control protocol communication connection; and an Ethernet interface conforming to the SNMP protocol connected to the switch.

[0027] The working process of the device: Liquid cooling cycle process: After the cooling tower cools the coolant, the coolant enters the CDU. The CDU distributes the coolant to the corresponding Tank unit. The coolant comes into contact with the computing node in the Tank unit, absorbs the heat generated by the computing node, and its temperature rises. The heated coolant flows out of the Tank unit and back to the CDU. The CDU exchanges heat with the incoming hot water, transferring the heat to the internal circulation medium (such as cooling water). The coolant, whose temperature has dropped, re-enters the liquid cooling cycle, while the circulation medium (cooling water) that has absorbed heat flows back to the cooling tower for cooling. This forms a liquid cooling cycle that continuously dissipates heat from the computing node.

[0028] Data Acquisition: Temperature and flow sensors on the customized cooling distribution unit (CDU) transmit the temperature and flow data of the coolant within the CDU to the BMC management module via interfaces such as I2C and SPI. Pump status information from the CDU is also synchronously transmitted to the BMC. Simultaneously, pressure parameter data is acquired by the BMC via the CAN interface. Temperature sensors on the cooling tower transmit temperature data to the BMC management module using the Modbus protocol, providing a basis for heat dissipation control of the liquid cooling system. Tank units and power distribution cabinets interact with the BMC management module via the RS485 protocol, transmitting their own operating status information to the BMC. The switch cabinet uses the SNMP protocol to transmit relevant status data of network devices to the BMC management module. Centralized Management and Control Process: The BMC centralized management and control module collects real-time data on parameters such as cooling tower efficiency and coolant temperature, CDU coolant inlet and outlet temperatures and flow rates, tank unit coolant temperature and computing node power consumption, power distribution cabinet power supply voltage, current, and power consumption, HVDC power transmission status, and switch cabinet operating temperature and network status. Based on these collected parameters, the BMC centralized management and control module analyzes the operating status of each component. When it detects excessively high coolant temperature in a tank unit, the BMC centralized management and control module adjusts the corresponding CDU to increase coolant flow or adjusts the cooling tower to enhance cooling capacity, thereby reducing the coolant temperature in that tank unit. When it detects abnormal power supply parameters in the power distribution cabinet, it promptly issues an alarm and takes corresponding protective measures. Through this centralized management and control, the stable and efficient operation of the entire system is ensured.

[0029] Specific implementation method four: Combination Figures 1 to 3 This embodiment is described, and a preferred embodiment is characterized by including the following steps: S1. Parameter Acquisition and Communication Verification: Through the multi-protocol adaptation layer, raw data frames of each monitoring parameter are periodically acquired from the cooling tower, cooling capacity distribution unit, Tank unit, power distribution cabinet, high-voltage DC power supply, and switch. For each raw data frame, verification is performed according to its communication protocol specifications, and a set of valid parameter values ​​that pass the verification is output. For data that fails the verification, the valid parameter values ​​corresponding to the previous cycle are used for replacement. For example, the cooling tower temperature is read at a frequency of 1Hz via Modbus, the liquid temperature and flow rate of the CDU are read at a frequency of 5Hz via I2C / SPI, and the server power consumption is read at a frequency of 10Hz via IPMI (via the Tank unit gateway). SNMPTrap alarm information from the switch is monitored in real time. All raw data is stored in the BMC's memory database or time-series database after CRC verification, unit conversion, and filtering (such as moving average filtering). S2. Data Fusion and Anomaly Detection: The numerical parameters in the effective parameter value set are digitally filtered to obtain a steady-state parameter value set; each steady-state parameter value in the steady-state parameter value set is compared with the corresponding alarm threshold read from the local control database; when any steady-state parameter value reaches or exceeds its alarm threshold, an alarm-level abnormal event is determined to have occurred, and an abnormal event identifier is generated and output. This identifier at least includes the abnormal parameter name and event type. According to the predefined equipment association diagram, the system finds all related parameters that may cause the abnormality (such as the CDU flow rate, cooling tower outlet water temperature, ambient temperature, etc. corresponding to the Tank), and calculates the abnormal contribution of each related parameter. The parameter with the highest contribution is determined as the core root cause, and the second highest is the secondary root cause. For example, if the Tank liquid temperature exceeds the limit, the core root cause is the insufficient pump speed of its CDU, and the secondary root cause is the overall cooling capacity of the cooling tower decreases. S3. Correlation Analysis and Root Cause Locator: Read the list of pre-associated correlation parameter names and their corresponding correlation weights from the local control database; obtain the current steady-state parameter value of each parameter in the list of correlation parameter names from the steady-state parameter value set; calculate the abnormal contribution of each correlation parameter; sort the correlation parameters according to their calculated abnormal contribution; and determine the correlation parameter with the highest contribution as the core root cause parameter and output it. S4. Strategy Matching and Instruction Generation: Using the event type in the abnormal event identifier and the core root cause parameter as the joint query key, the corresponding predefined control strategy is matched in the control strategy library. The predefined control strategy is parsed, and combined with the current steady-state parameter value set, a control instruction sequence is generated. The preset control strategy is matched from the strategy library. The main action of this strategy (such as increasing the speed of the CDU-A variable frequency pump) and a backup action (such as simultaneously increasing the speed of the cooling tower fan) are matched. After matching, the BMC will generate specific and executable control instructions (such as I2C instruction frames containing target flow values) based on context information such as the contribution of the root cause, and send them to the CDU for execution through the corresponding communication interface (such as the I2C bus). S5. Command Execution and Closed-Loop Feedback: Through the multi-protocol adaptation layer, the control command sequence is sent to the corresponding target device for execution. Within the set monitoring window after command execution, the execution step S1 is returned to collect new system operating parameters. Based on the new operating parameters, steps S2 to S3 are repeated to obtain a new set of steady-state parameter values ​​and core root cause parameters. By comparing the changes in steady-state parameter values ​​and core root cause parameters before and after command execution, the control effect is evaluated. If the control effect does not meet expectations, a backup control strategy or cross-subsystem collaborative control process is triggered. For example, when an overload warning occurs in the distribution cabinet, the BMC will not only adjust the power distribution strategy but may also simultaneously notify the HVDC to optimize the output and order the backup energy storage unit in the Tank to prepare for connection, thereby achieving collaboration within the power supply system. Alternatively, when severe network congestion affects monitoring data, the BMC will, while controlling network devices, notify other systems (such as the HVDC) to temporarily reduce the generation of non-critical data, thereby achieving cross-subsystem collaboration.

[0030] In summary, a closed-loop, intelligent, and collaborative system-level autonomous management framework can be built, upgrading from "decentralized monitoring" to "unified intelligent collaborative control," achieving a leap in operational efficiency, system stability, and overall energy efficiency.

[0031] Specific Implementation Method Five: Combination of Filtering and Contribution Calculation Examples Figures 1 to 3 This embodiment describes step S2: For the cooling tower outlet water temperature and the pressure parameters of the cooling capacity distribution unit, a moving average filter is used. For the coolant temperature in the Tank unit, a Kalman filter algorithm is used. The observed inputs of this algorithm are the temperature sensor sampling values, and the state prediction inputs are based on the temperature value output from the previous filtering cycle and the current flow rate value of the cooling capacity distribution unit, further corrected by sensor readings, while dynamically updating the error covariance. (See Appendix) Figure 2 ; Filtering: A moving average filter is used for parameters such as cooling tower temperature and CDU pressure. For example, if the cooling tower outlet water temperature is [37.1, 37.3, 37.0]℃ for three consecutive cycles, then the current steady-state value = (37.1+37.3+37.0) / 3 = 37.13℃.

[0032] Kalman filtering is used for the coolant temperature of the Tank unit. As shown in Table 1, the algorithm requires two inputs: a) Observed value: the raw temperature sensor value (including noise) collected by S100. b) Predicted value: calculated using a simplified thermodynamic model based on the temperature value output by the previous filtering cycle and the current CDU flow rate value collected by S100. The Kalman filter fuses these two values ​​to output an optimal steady-state temperature estimate, while updating the internal covariance matrix. This step effectively filters out sensor noise and transient disturbances; Anomaly detection: After filtering, a high-quality set of steady-state parameter values ​​is obtained. The analysis unit reads the alarm (alarm threshold) corresponding to each parameter from the thresholds table in the local control database (Redis). Each value is compared; for example, if the steady-state value of "Tank1_Temp" is 46℃, while its alarm threshold is 45℃, an anomaly is detected. Output: Generate an exception event identifier, such as {"event_id": "E20250320120001", "type": "TankOverTemp", "param": "Tank1_Temp", "value": 46}. (This identifier is the only signal to trigger subsequent intelligent analysis.) Table 1 is the Redis database data table of this invention, as follows: Specific Implementation Method Six: The calculation of the abnormal contribution degree in the dynamic adjustment embodiment is as follows: for each correlation parameter i, its abnormal contribution degree ,in The degree of abnormality of the parameter is calculated using the following formula: ,in The current value, These are the boundary values ​​of the normal range. Alarm threshold; The preset association weights for the parameters; This formula compares the distance the current value deviates from the normal boundary with the critical distance that triggers the alarm, thus obtaining a scale that intuitively reflects the severity of the anomaly.

[0033] Specific Implementation Method Seven: Dynamic Adjustment of Examples Figures 1 to 3 In this embodiment, in step S4, when the abnormal event type is "Tank unit liquid temperature too high" and the core root cause parameter is "corresponding cooling capacity distribution unit flow rate", the predefined control strategy is to adjust the variable frequency pump speed. The generated control command sequence includes: Generate I2C format commands for adjusting the speed of the corresponding CDU variable frequency pump. The target flow parameters in the commands are dynamically calculated based on the contribution of the core root cause.

[0034] Target traffic The following calculations were performed: ,in, The current flow rate value is read from the current set of steady-state parameter values. The abnormal contribution of the flow parameters calculated in step S3 is given by k, where k is a preset gain coefficient. The value should not exceed the rated maximum value of the equipment, as shown in Table 2; Table 2 is the calculation table for the dynamic adjustment parameters of this invention, as detailed below: Example: Assuming the current flow rate Vcurrent is 15L / min, the core root cause contribution... ,but: If the maximum allowed flow rate of the CDU is 20L / min, then the final target flow rate is 20L / min. The BMC encapsulates the instruction frame according to the I2C protocol (device address 0x40, function code 0x03, data field encoding is 0x14, CRC16 check) and sends it to the CDU for execution.

[0035] Specific implementation method eight: Combination Figures 1 to 3 This embodiment describes the collaborative control example of this embodiment. Combination Figures 1 to 3 This embodiment describes a cross-subsystem collaborative control process, specifically step S5, which includes: S51. Network Congestion Coordination: When the switch cabinet network is severely congested, while issuing security policy instructions to the switches, a first coordination instruction is sent to the HVDC. This instruction is a request message suggesting adjustment of non-critical load power. The BMC generates a structured JSON instruction, for example: { "CommandID": "CMD-HVDC-2025XXXX-006", "TargetDevice": "HVDC-100kVA", "CommandType": "OptimizePowerDistribution", "AffectedBranch": "PDU-005-Branch-02", "CurrentLoad": 30.2, "RatedCurrent": 32.0, "RelatedDevices": ["CDU-Pump-003", "CoolingTower-Fan-002"], Request: "Improve the power supply stability of this branch without affecting the core power supply, allowing current fluctuations of ±0.5A." "Timeout": 5000, Checksum: "0x9E1C098C" } BMC sends this JSON command to HVDC via Ethernet TCP, and HVDC dynamically adjusts the inverter or filter strategy; S52. Distribution Overload Coordination: When an overload warning is issued for a branch current in the distribution cabinet, a second coordination command (power supply optimization request) is sent to the HVDC, and a third coordination command is sent to the associated Tank unit (causing the energy storage system within that Tank unit to enter power supply standby mode). For example, the command can be encapsulated in Modbus-RTU format. Slave address: Modbus address of the Tank unit (e.g., 0x08); Function code 0x06 (Write a single register); Register address 0x0010; Write data 0x0001 (representing "switch to temporary power standby mode"); CRC16 checksum.

[0036] Upon receiving the data, the Tank unit's energy storage management controller responds by closing the grid connection switch, putting the energy storage battery pack into standby mode, and monitoring the branch voltage / frequency in real time to prepare for seamless power replenishment when the mains power drops.

[0037] Detailed Implementation Method Nine: Effectiveness Evaluation Example Combination Figures 1 to 3 In this embodiment, step S5 includes the following criteria for determining that the control effect does not meet expectations: after the monitoring window ends, the abnormal parameter values ​​read from the new set of steady-state parameter values ​​do not return to their normal range. Alternatively, judging from the new core root cause parameters, the main root cause of the abnormality has not changed.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A containerized immersion liquid-cooled centralized control device, characterized in that, include: The container itself; And a liquid cooling subsystem installed inside the container body, the liquid cooling subsystem including a cooling tower, at least one cooling capacity distribution unit and at least one Tank unit, the cooling capacity distribution unit and the Tank unit being installed in a one-to-one correspondence; And an electronic power supply system installed within the container body, the electronic power supply system including a power distribution cabinet and a high-voltage DC power supply; And a network switching subsystem disposed within the container body, the network switching subsystem including a switch cabinet and switches disposed therein; It also includes a centralized control module for substrate management controller installed inside the container body.

2. The apparatus according to claim 1, characterized in that, The centralized control module of the baseboard management controller includes a multi-protocol adaptation layer, including a processor, memory, multi-protocol communication interface, local control database, and control strategy library. The local control database includes configuration data, which includes: preset normal value ranges, early warning thresholds, alarm thresholds, and a set of correlation weights representing the degree of influence between parameters for each monitored parameter. The control strategy library includes predefined control strategies for different anomaly types and root cause combinations, and each control strategy contains at least one main action instruction template.

3. The apparatus according to claim 2, characterized in that, The multi-protocol adaptation layer includes: the I2C interface, SPI interface, and CAN interface connected to the cooling capacity distribution unit; the RS-485 interface following the Modbus-RTU protocol connected to the cooling tower; the RS-485 interface connected to the Tank unit and the power distribution cabinet; the Ethernet interface connected to the high-voltage DC power supply and establishing a transmission control protocol communication connection; and the Ethernet interface following the SNMP protocol connected to the switch.

4. A control method based on the containerized immersion liquid-cooled BMC centralized control device according to any one of claims 1-3, characterized in that: Includes the following steps: S1. Parameter Acquisition and Communication Verification: Through the multi-protocol adaptation layer, the original data frames of each monitoring parameter are periodically acquired from the cooling tower, cooling capacity distribution unit, Tank unit, power distribution cabinet, high-voltage DC power supply and switch. For each original data frame, the communication protocol specification is verified, and a set of valid parameter values ​​that have passed the verification is output. For data that fails to be validated, replace it with the valid parameter value from the previous period; S2. Data Fusion and Anomaly Detection: Perform digital filtering on the numerical parameters in the set of effective parameter values ​​to obtain a set of steady-state parameter values; compare each steady-state parameter value in the set of steady-state parameter values ​​with the corresponding alarm threshold read from the local control database; when any steady-state parameter value reaches or exceeds its alarm threshold, determine that an alarm-level abnormal event has occurred, generate and output an abnormal event identifier, which at least includes the abnormal parameter name and the event type; S3. Correlation Analysis and Root Cause Locator: Read the list of pre-associated correlation parameter names and their corresponding correlation weights from the local control database; obtain the current steady-state parameter value of each parameter in the list of correlation parameter names from the steady-state parameter value set; calculate the abnormal contribution of each correlation parameter; sort the correlation parameters according to their calculated abnormal contribution; and determine the correlation parameter with the highest contribution as the core root cause parameter and output it. S4. Strategy matching and instruction generation: Using the event type in the abnormal event identifier and the core root cause parameter as the joint query key, match the corresponding predefined control strategy in the control strategy library, parse the predefined control strategy, and generate a control instruction sequence by combining it with the current steady-state parameter value set; S5. Command Execution and Closed-Loop Feedback: Through the multi-protocol adaptation layer, the control command sequence is sent to the corresponding target device for execution. Within the set monitoring window after command execution, the execution step S1 is returned to collect new system operating parameters. Based on the new operating parameters, steps S2 to S3 are repeated to obtain a new set of steady-state parameter values ​​and core root cause parameters. By comparing the changes in steady-state parameter values ​​and core root cause parameters before and after command execution, the control effect is evaluated. If the control effect does not meet expectations, a cross-subsystem collaborative control process is initiated.

5. The control method for the containerized immersion liquid-cooled BMC centralized control device according to claim 4, characterized in that, In step S1, the outlet water temperature of the cooling tower and the pressure parameters of the cooling capacity distribution unit are filtered using a moving average filter. The coolant temperature in the Tank unit is filtered using a Kalman filter algorithm. The input of the observed values ​​of this algorithm is the temperature sensor sampling value. The current temperature state prediction input is based on the temperature value output in the previous filtering cycle and the current flow rate value of the cooling capacity distribution unit.

6. The control method for the containerized immersion liquid-cooled BMC centralized control device according to claim 4, characterized in that, In step S3, the calculation of the anomaly contribution specifically involves: for each correlation parameter i, its anomaly contribution... ,in The degree of abnormality of the parameter is calculated using the following formula: ,in The current value, These are the boundary values ​​of the normal range. Alarm threshold; The preset association weights for the parameters.

7. The control method for the containerized immersion liquid-cooled BMC centralized control device according to claim 4, characterized in that, In step S4, when the abnormal event type is: the Tank unit liquid temperature reaches a temperature threshold and the core root cause parameter is the flow rate of the corresponding cooling capacity distribution unit, the matched predefined control strategy is to adjust the variable frequency pump speed. The generated control command sequence includes: when the Tank unit liquid temperature reaches the temperature threshold, generating an I2C format command to adjust the speed of the variable frequency pump of the corresponding customized cooling capacity distribution unit. The target flow rate parameter in the command is dynamically calculated based on the contribution of the core root cause. The following calculations were performed: ,in, The current flow rate value is read from the current set of steady-state parameter values. The abnormal contribution of the flow parameters calculated in step S3 is given by k, where k is a preset gain coefficient. The value should not exceed the rated maximum value of the equipment.

8. The control method for the containerized immersion liquid-cooled BMC centralized control device according to claim 4, characterized in that, In step S5, the cross-subsystem collaborative control process includes: S51: When collaborative control is triggered due to network traffic in the switch cabinet reaching the traffic threshold, a first collaborative instruction is sent to the high-voltage DC power supply at the same time as a security policy instruction is sent to the switch. The first collaborative instruction is a request message suggesting adjustment of non-critical load power. S52: When coordinated control is triggered due to the branch current of the distribution cabinet reaching the current threshold, a second coordinated command is sent to the high-voltage DC power supply, and a third coordinated command is sent to the associated Tank unit. The second coordinated command is a power supply optimization request, and the third coordinated command is an instruction to put the energy storage system in the Tank unit into a power supply standby state.

9. The control method for the containerized immersion liquid-cooled BMC centralized control device according to claim 4, characterized in that, In step S5, the judgment conditions for the evaluation control effect not meeting expectations include: after the monitoring window ends, the abnormal parameter values ​​read from the new steady-state parameter value set do not return to their normal range, or, judging from the new core root cause parameters, the main root cause of the abnormality has not changed.

Images