Layered control method and system suitable for air conditioning cold source system

Abstract

The application provides a layered control method and system suitable for an air conditioner cold source system, an instruction containing an operation object, a target action and an execution time limit parameter is triggered through a system operation terminal, is transmitted in real time to a hierarchical control module through an industrial bus and an AI verification module is started synchronously; the AI verification module carries out multi-dimensional adaptability verification based on system panoramic operation data, verifies the feasibility of the instruction in combination with a historical working condition database, the instruction is hierarchically disassembled and the execution rule is specified if the verification is passed, an optimization suggestion is generated and re-verification is supported after parameter adjustment if the verification is not passed, feedback data is collected in real time during execution, the AI module intervenes in an emergency and gives an alarm in case of an exception, the system state is checked, closed-loop confirmation information is sent and whole-process data is stored after execution is completed; when the system is started or stopped or is regulated in terms of load increase or decrease, various types of data obtained by the AI module are integrated, the current equipment load state of each subsystem is associated, and it is ensured that the system regulation instruction drives the collaborative response of the subsystems.

Description

Technical Field

[0001] This invention relates to the field of hierarchical control technology, and more specifically, to a hierarchical control method and system applicable to air conditioning cold source systems. Background Technology

[0002] Traditional air conditioning cooling source system control often employs a centralized or simple distributed architecture, lacking a fine-grained hierarchical division of the system, subsystems, and equipment, resulting in insufficient coordination in command transmission and execution. In scenarios involving multiple devices operating in tandem and complex switching of operating conditions, problems such as delayed command response and parameter mismatch frequently occur. This not only reduces cooling efficiency but also exacerbates localized wear and tear due to uneven equipment load distribution, shortening the overall lifespan. Furthermore, manual command verification often relies on single-dimensional parameter detection, failing to integrate comprehensive system operation data and historical operating conditions for a complete assessment. This leads to higher command execution risks and a lack of rapid and accurate emergency intervention mechanisms in the event of anomalies, easily causing system fluctuations or even malfunctions.

[0003] Existing systems mostly employ a single-bus architecture for data transmission, resulting in weak resistance to electromagnetic interference and insufficient transmission stability. Bus failures can easily lead to control link interruptions. Furthermore, the lack of efficient real-time data acquisition, preprocessing, and hierarchical storage mechanisms makes it difficult to support dynamic control decisions. In addition, traditional control methods have a low level of intelligence, failing to autonomously optimize control strategies based on system operating status and equipment wear. In scenarios without human intervention, they struggle to achieve a balance between balanced equipment wear and optimal energy efficiency, failing to adapt to complex and ever-changing operational demands. This restricts the operational stability, energy-saving potential, and long-term maintenance benefits of air conditioning cooling source systems. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a hierarchical control method suitable for air conditioning cold source systems, the method comprising:

[0005] The system operation terminal triggers an instruction containing the operation object, target action, and execution time parameters, which is transmitted in real time to the hierarchical control module via the industrial bus and the AI ​​verification module is started simultaneously.

[0006] The AI ​​verification module performs multi-dimensional adaptability verification based on the system's panoramic operation data, and verifies the feasibility of instructions by combining historical operating condition databases. If the verification passes, the instructions are broken down into levels and the execution rules are clarified. If the verification fails, optimization suggestions are generated and re-verification is supported after parameter adjustment. Feedback data is collected in real time during execution. In case of anomalies, the AI ​​module intervenes in an emergency and issues an alarm. After the execution is completed, the system status is checked, closed-loop confirmation information is sent, and the entire process data is stored.

[0007] When the system performs power-on, power-off, and load-reduction adjustments, it incorporates various types of data acquired by the AI ​​module and correlates them with the current device load status of each subsystem to ensure that the system's control commands drive the subsystems to respond in a coordinated manner.

[0008] After receiving instructions from the system level, the subsystem decomposes them into appropriate sub-instructions based on its own equipment operating parameters and sends them to specific equipment. At the same time, it collects equipment operating status and feeds it back to the system level, forming a closed-loop feedback between the system and subsystems.

[0009] After receiving detailed instructions, the devices within the subsystem convert them into specific operating parameters, collect relevant data about their own operation in real time, and synchronously feed them back to the subsystem and the system-level AI module.

[0010] A real-time data link is established to acquire and store various relevant data in parallel. Verification and optimization are completed before the execution of manual instructions. When there is no human intervention, the AI ​​module autonomously formulates control strategies at the system, subsystem and equipment levels to achieve equipment wear balance. During the control execution, real-time verification and dynamic fine-tuning are performed. Data from all stages are stored afterward. The AI ​​module regularly analyzes historical samples, identifies the optimal control strategy and updates the model.

[0011] Furthermore, embodiments of the present invention also provide a layered control system suitable for air conditioning cold source systems, characterized in that it includes:

[0012] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described hierarchical control method for an air conditioning cooling source system by executing the machine-executable instructions.

[0013] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, and the processor executing the machine-executable instructions, causing the computer device to execute the above-described hierarchical control method applicable to air conditioning cold source systems.

[0014] Based on the above, a three-tiered control architecture of system, subsystem, and equipment, combined with AI multi-dimensional verification and dual-bus redundancy design, significantly improves the operational stability and command execution reliability of the air conditioning cooling source system. After command triggering, panoramic data verification and historical operating condition comparison effectively avoid risks such as parameter mismatch and equipment linkage conflicts. Coupled with real-time feedback and emergency intervention mechanisms during execution, the system failure rate and command execution deviation rate are significantly reduced. Dual-bus redundancy, interrupt triggering, and bandwidth priority allocation strategies ensure that command transmission latency is controlled at the millisecond level, and rapid switching is possible in case of bus failure to avoid control link interruption. Simultaneously, the layered decomposition and closed-loop feedback design ensure coordinated response between system control commands and subsystems and equipment, eliminating single-device overload or linkage lag issues.

[0015] Leveraging the autonomous control and model iteration capabilities of the AI ​​module, this invention achieves the dual goals of balancing equipment wear and optimizing system energy efficiency. Without human intervention, the AI ​​formulates differentiated control strategies based on three-dimensional evaluation results, which are then precisely implemented after simulation verification. Dynamic fine-tuning prevents unbalanced equipment wear and extends overall service life. The structured storage and regular analysis of the entire process data not only provide a complete basis for fault tracing but also support continuous optimization of the AI ​​model, enabling the control strategies to adapt to complex operating conditions. This significantly reduces manual maintenance costs and energy consumption, balancing the economic efficiency and long-term benefits of system operation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the execution flow of a hierarchical control method for air conditioning cold source systems provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of exemplary hardware and software components of a hierarchical control system for air conditioning cold source systems provided in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a hierarchical control method for air conditioning cold source systems according to an embodiment of the present invention. The hierarchical control method for air conditioning cold source systems will be described in detail below.

[0019] Step S110: Trigger an instruction containing the operation object, target action, and execution time parameters through the system operation terminal, transmit it to the hierarchical control module in real time via the industrial bus, and simultaneously start the AI ​​verification module.

[0020] The command is triggered by a domestic industrial touch screen equipped with the DevEco Studio development interface. The target object is selected as centrifugal chiller No. 1, the target action is to load to 60% load, and the execution time parameter is set to 300 seconds. The command is transmitted in real time to the domestic PLC hierarchical control module via the Profinet industrial bus. At the same time, a high-level signal is output through the module's DO interface to start the domestic AI edge computing module. After receiving the level signal, the AI ​​module automatically loads and initializes the verification rule base trained based on Baidu PaddlePaddle.

[0021] Step S111: The system operation terminal is equipped with a visual interface, with preset options related to the operation object, target action, and execution time. After the user inputs the parameters, the terminal verifies the completeness and legality of the parameters. After passing the verification, the user triggers the command, and the terminal synchronously records the operator, trigger time, and other additional information.

[0022] The system operation terminal uses DevEco Studio to develop a visual interface, which includes a drop-down menu for the operation object, a radio button for the target action, and an execution time input box (1-3600 seconds). The terminal program automatically verifies the integrity and legality of the parameters (e.g., the auxiliary water pump does not support loading actions). After the verification is successful, the user clicks the trigger command, and the terminal synchronously records the login account (e.g., OP001) and the trigger time (accurate to milliseconds), and stores them in a domestic industrial-grade storage module (capacity ≥8GB, supports power loss protection) and a cloud database to achieve dual backup and traceability.

[0023] Step S112: Encapsulate the instructions using a unified data frame format, including fields, and encapsulate them to complete data encoding according to the transmission protocol adapted to the industrial bus.

[0024] Instructions are encapsulated in 16-byte binary data frames, including a frame header (0xAA55), an operation object ID (e.g., 0x0001), a target action code (e.g., 0x03), an execution time limit (e.g., 0x012C), reserved fields, a CRC16 checksum (0x8005 polynomial), and a frame tail (0x55AA). After encapsulation, the instructions are converted to Profinet RT protocol data units according to the IEC 61158 standard, using ASCII encoding to ensure compatibility with industrial bus transmission protocols and guarantee data transmission accuracy (compatible with communication protocols of domestic PLCs and AI modules).

[0025] Step S113: Select an industrial Ethernet bus, configure the interface module and adopt a dual-bus redundancy design, and ensure fast command transmission through transmission mechanism and bandwidth allocation strategy;

[0026] The system utilizes the Profinet industrial Ethernet bus and is equipped with a Huawei S5735I-H series domestic industrial switch, employing a dual-bus redundancy design. Hardware adaptation is achieved through the Dongtu Technology KOM2100 domestic communication card and a domestic PN / IE compatible module. Interrupt triggering and bandwidth priority queue mechanisms are enabled, with a reserved 2Mbps fixed bandwidth. Command data is marked as the highest priority. A bus status monitoring module is configured to collect communication rate, bit error rate, and other indicators in real time. Transmission delay is controlled to ≤50ms, and retransmission (up to 3 times) or early warning is triggered in case of abnormalities to ensure fast and stable transmission of commands.

[0027] Step S1131: Select a PLC / DDC built-in bus communication card and an external Ethernet communication module for the terminal that are compatible with the system bus protocol, complete the hardware installation and connection, avoid strong electromagnetic interference, and use a specific wiring method to reduce signal attenuation;

[0028] The hierarchical control module has a built-in domestic Profinet compatible communication card, and the operation terminal has an external domestic PN / IE Ethernet communication module. It uses STP Cat5e shielded twisted pair cable crimped according to the T568B standard, and is laid through a metal cable tray with a distance of ≥30cm from the power cable. The communication module and the Huawei S5735I-H switch port are shielded and grounded (grounding resistance ≤4Ω). The cable length is controlled within 100 meters to avoid strong electromagnetic interference and reduce signal attenuation.

[0029] Step S1132: Deploy two independent primary and backup industrial Ethernet buses with separate cabling paths, configure independent network devices and maintain consistent topology, and connect the relevant control and verification modules via dual links;

[0030] Two independent Profinet buses with separate cabling paths are deployed (the main bus is along column A of the computer room, and the backup bus is along column B), each configured with one Huawei S5735I-H series domestic industrial switch, with port configuration and VLAN division completely consistent; the hierarchical control module, AI verification module, and operation terminal are all connected to the main and backup buses through dual network cards respectively, with a star topology and a transmission rate of 100Mbps full duplex to ensure the consistency of hardware configuration of the main and backup buses.

[0031] Step S1133: Embed a management program in the hierarchical control module, preset the priority of the primary and backup buses, monitor the status of the primary bus in real time, and quickly switch to the backup bus when the fault threshold is met.

[0032] The Huichuan H3U series hierarchical control module has a built-in organization block with an embedded bus management program. The main bus is preset to have a higher priority than the backup bus. The program reads the main bus communication status register every 10ms to monitor communication timeout (single threshold 50ms) and bit error rate (threshold > 0.1%). When the main bus times out three times in a row or the bit error rate exceeds the standard, the bus selection register is modified (0x01 is switched to 0x02) to achieve a fast switch within ≤100ms, ensuring uninterrupted instruction transmission.

[0033] Step S1134: Enable data synchronization function. When the main bus transmits data, it is synchronously stored in the backup bus buffer. After the link is switched, the backup bus data is read. After the main bus recovers, it is switched back and the backup bus buffer is cleared.

[0034] When the master bus data synchronization function is enabled, when the master bus transmits command data, it synchronously writes the data to the 8KB buffer of the backup bus (address 0x20000000-0x20001FFF) via DMA. After the master bus fails and switches over, the hierarchical control module (Inovance H3U series) immediately reads the data in the backup bus buffer. After the master bus recovers to normal (five consecutive status tests meet the requirements), it automatically switches back to the master bus and clears the backup bus buffer with a clear command to ensure the continuity of data transmission.

[0035] Step S1135: Enable interrupt triggering function on the relevant communication interface, the terminal encapsulates the instruction and initiates an interrupt request, prioritizes the allocation of transmission resources, and the control module switches to interrupt receiving mode to read the instruction.

[0036] In the Dongtu Technology KOM2100 domestic communication card, the interrupt trigger function is enabled, and the interrupt trigger register is configured as 0x01 (rising edge trigger). After the operation terminal (Huichuan IT6000 series) encapsulates the instruction, it sends a 5ms high-level interrupt request to the control module. After receiving the signal, the hierarchical control module (Huichuan H3U series) suspends low-priority tasks, switches to interrupt reception mode, prioritizes CPU resources to read instructions, and the interrupt response time is ≤10ms to ensure the priority of instruction transmission.

[0037] Step S1136: Configure a bandwidth priority queue on the switch, mark the instruction data as the highest priority and reserve a fixed bandwidth, and ensure the instruction transmission bandwidth through a dynamic adjustment algorithm;

[0038] Configure a 0-3 priority queue on the Huawei S5735I-H series domestic industrial switch, mark the instruction data as level 0 (highest priority), and reserve 2Mbps fixed bandwidth; enable the dynamic bandwidth adjustment algorithm, collect the bus load rate every 50ms, and automatically compress the bandwidth of level 1-3 data (such as status monitoring data) when the load rate is >80%; save the configuration parameters to the switch's non-volatile memory to ensure that the instruction transmission bandwidth is not affected by the bus load.

[0039] Step S1137: Add a monitoring module to the management program to collect bus status data in real time, verify the integrity of transmitted data, and trigger retransmission or push early warning information when an anomaly occurs.

[0040] A status monitoring module is added to the bus management program to collect data such as bus communication rate, bit error rate, and load rate every 100ms. The transmission command adopts double verification of checksum + CRC16. The checksum is the sum of the bytes of the data frame excluding the checksum code, and the CRC16 adopts the 0x8005 polynomial. If the verification fails, the retransmission is triggered (interval of 50ms, up to 3 times). If the retransmission fails, a "bus transmission abnormality" warning is pushed to the terminal (Huichuan IT6000 series), and the abnormality log is recorded to the local Flash storage of the hierarchical control module (Huichuan H3U series).

[0041] Step S114: Configure the module to adapt to the communication interface, open an independent buffer, verify the legality of the received data frame and parse the parameters, and mark the instruction status after verifying the validity.

[0042] The Huichuan H3U series hierarchical control module integrates a domestically compatible Ethernet interface chip, opens a 4KB independent buffer (0x20002000-0x20002FFF), and adopts a FIFO mechanism and semaphore mutex lock; the communication interface enables interrupt reception mode, and after receiving data, it sequentially completes frame header and frame tail verification, CRC16 verification, data parsing and parameter verification; if the parameters are valid, they are marked as pending verification and associated with a unique identifier; if they are invalid or the verification fails, the corresponding abnormal status is marked, and the buffer is cleared synchronously to ensure the standardization of instruction reception and parsing.

[0043] Step S1141: Select an interface chip that is compatible with the industrial bus transmission protocol and integrate it into the hierarchical control module. Configure the interface communication parameters to be consistent with the bus, load the adapter driver, initialize the registers and set the interrupt trigger conditions to ensure that the interface is compatible with the protocol.

[0044] A domestically produced compatible Ethernet interface chip is integrated into the control module expansion board, configured with the following communication parameters: baud rate 100Mbps, full-duplex, IP 192.168.0.10, subnet mask 255.255.255.0, consistent with the Profinet bus parameters. A domestic operating system adapter driver is loaded, and the data reception mode is configured through registers. The interrupt trigger condition is set to data frame reception completion (INT pin high) to ensure interface compatibility with the Profinet protocol and guarantee data reception stability.

[0045] Step S1142: Use static memory allocation to create an independent instruction receive buffer, adopt a FIFO storage mechanism, specify the initial position of the read and write pointers and configure a mutex lock to restrict single-process access and avoid conflicts;

[0046] A 4KB buffer (addresses 0x20002000-0x20002FFF) is allocated in the RAM of the hierarchical control module (Inovance H3U series) using static memory allocation, with the initial positions of the read and write pointers both being the starting address of the buffer; a semaphore-based mutex lock (initial value 1) is configured to restrict access to the buffer to only the receiving thread, avoiding data corruption caused by multi-threaded conflicts; a FIFO storage mechanism is adopted to ensure that instructions are processed in the order of reception, guaranteeing the orderly storage of data.

[0047] Step S1143: The communication interface enables interrupt reception mode. After detecting the data stream, an interrupt is triggered and the data is written to the buffer byte by byte, the pointer is updated, and the reception timeout threshold is set. If the data is not received within the timeout period, the reception is determined to have failed. The buffer is then cleared and the pointer is reset.

[0048] The communication interface is enabled in interrupt reception mode. Upon detecting the bus data stream, a reception interrupt is triggered. The interrupt service routine writes to the buffer byte by byte and updates the write pointer. A 200ms reception timeout threshold is set. If the write pointer is not updated within 200ms (the data frame has not been fully received), it is determined that the reception has timed out. The instruction status is marked as "reception failure", the buffer is cleared and the pointer is reset to avoid invalid data occupying the resources of the hierarchical control module (Inovance H3U series).

[0049] Step S1144: After receiving the data frame, first check the frame header and frame tail identifiers. If they are invalid, mark them as invalid frames and clean them up. If they are valid, extract the CRC check code and compare it. If they are inconsistent, mark them as verification failure frames, record the log and clean up the buffer.

[0050] After receiving a data frame, the frame header (0xAA55) and frame trailer (0x55AA) identifiers are first checked. If they do not match, the frame is marked as invalid and the buffer is immediately cleared. If the identifiers are valid, the CRC16 checksum is extracted and compared with the result calculated using the 0x8005 polynomial. If they do not match, the frame is marked as a check failure frame, the exception log is recorded to the KingbaseES V8R6 database, and the buffer is cleared to ensure the integrity and accuracy of the received data frames.

[0051] Step S1145: Extract data according to the preset data frame structure, convert it into a format that can be processed internally by the module, and adjust the byte order to adapt to the module CPU;

[0052] Based on the preset 16-byte data frame structure, the operation object ID, target action code, execution time, and other fields are extracted from the buffer; the binary data is converted into a decimal format that can be processed by the control module, and the byte order is adjusted to adapt to the Huichuan H3U series CPU (little-endian mode) to ensure that the parsed data format meets the module's processing requirements, laying the foundation for subsequent parameter verification.

[0053] Step S1146: Call the built-in system device information database to verify the validity of the parsed parameters. If any parameter fails, it is determined to be an illegal parameter and the specific item is recorded.

[0054] The built-in system device information database of the hierarchical control module (Inovance H3U series) is called to check whether the parsed operation object ID exists, whether the target action code matches the device's supported action, and whether the execution time is within the range of 1-3600 seconds; if any parameter is invalid, it is judged as "parameter invalid", and the specific illegal item (such as "execution time exceeds the range") is recorded in the DM8 database log to ensure that the instruction parameters meet the system operation requirements.

[0055] Step S1147: If the parameter verification passes, the instruction is marked as pending verification and associated with a unique identifier. If the verification or check fails, the corresponding abnormal state is marked. Regardless of the state, the pointer is updated and the buffer is cleared to prepare for the next reception.

[0056] After the parameter verification is successful, a unique identifier is assigned to the instruction using the UUID algorithm, marking the status as pending verification and associating it with the original parameters; if the verification or check fails, the corresponding abnormal status is marked; regardless of the status, the read and write pointers are updated to the initial position, the buffer is cleared, and resources are reserved for the next instruction reception to ensure the process is closed-loop (the identifier and status data are synchronously stored in the domestic caching middleware Alibaba Cloud PolarDB-X).

[0057] Step S115: The hierarchical control module and the AI ​​verification module are connected through an internal high-speed communication link. After successfully parsing the instruction, a start signal is sent. The AI ​​verification module completes initialization and returns the ready status.

[0058] The hierarchical control module (Inovance H3U series) and the AI ​​verification module (Huawei Atlas 200I DK A2) are connected via a PCIe 4.0 high-speed communication link. After the control module successfully parses the instruction, it sends a high-level start signal (lasting 20ms). After receiving the signal, the AI ​​module loads the verification rule base and dataset trained on Baidu PaddlePaddle. After completing the initialization, it feeds back a low-level ready signal through the link to ensure efficient communication between modules and a complete initialization process.

[0059] Step S116: After receiving the ready feedback, the hierarchical control module sends a success message back to the terminal. If no feedback is received within the timeout period, an abnormality is determined, an alarm is pushed and the execution of the command is suspended. The command will be retried after investigation.

[0060] After receiving the readiness feedback from the AI ​​verification module (Huawei Atlas 200I DK A2), the hierarchical control module (Inovance H3U series) sends a successful instruction reception prompt back to the operation terminal (Inovance IT6000 series) via the Profinet bus. If a 500ms timeout threshold is set and no readiness feedback is received within the timeout period, it is determined to be "module communication abnormal", an alarm message is pushed to the terminal and instruction execution is suspended. After the technicians have checked the hardware connection or module failure, the user can re-trigger the instruction to ensure that the abnormal scenario is traceable and handled (alarm logs are stored in KingbaseES V8R6).

[0061] Step S120: The AI ​​verification module conducts multi-dimensional adaptability verification based on the system's panoramic operation data, and verifies the feasibility of the instructions in conjunction with the historical operating condition database;

[0062] The Huawei Atlas 200I DK A2 domestic AI verification module was used to conduct feasibility verification based on the system's panoramic operation data and historical operating condition database. The module connects to the Advantech Adam-6000 series domestic data acquisition unit via the Profinet bus to acquire multi-dimensional operating condition data in real time. Simultaneously, it accesses the DM8 historical operating condition database, using the Euclidean distance algorithm to retrieve similar scenario records. Combining real-time verification rules with historical risk analysis results, the module determines the feasibility of instructions according to preset logic, ensuring that the verification balances real-time performance and reliability.

[0063] Step S121: The AI ​​verification module acquires key operating condition data of the host, water pump, subsystem, electric valve and the whole system through the data acquisition unit. After synchronous transmission through the industrial bus, it forms a structured panoramic operation dataset through outlier removal, missing value completion and standardization transformation.

[0064] Advantech Adam-6000 series domestic data acquisition units were selected and deployed at the main unit, water pumps, subsystem control cabinets, and electric valves. Key operating condition data were synchronously acquired at a frequency of 500ms / time via Profinet bus. Outliers were removed using the 3σ principle, and short-term missing data were filled in using linear interpolation. The data was mapped to the 0-1 range through Min-Max standardization, and finally a CSV format structured panoramic operation dataset was formed, which includes fields such as device ID, parameter name, value, and acquisition timestamp to ensure data standardization (the dataset is synchronously stored in the TDengine time series database).

[0065] Step S122: Preset three types of verification rule bases to clarify the host status adaptation requirements, water pump set matching standards and subsystem linkage adaptation conditions respectively;

[0066] The system includes three types of verification rule bases: host status, pump group matching, and subsystem linkage. It uses the domestic relational database DM8 to store the rules in tables (tb_host_rule / tb_pump_rule / tb_subsys_rule) according to rule type. The system is designed with a layered architecture: a basic layer (defining parameter units), a rule layer (storing verification logic), and an application layer (associating instructions and rules). It also reserves extended fields for adapting to device models and effective versions, supports differentiated configuration and iterative rule updates based on device type, and ensures the flexibility of the rule base.

[0067] Step S1221: Design a layered architecture of foundation layer, rule layer, and application layer. The foundation layer defines basic information, the rule layer stores verification logic, and the application layer associates instructions and rules. A relational database is used to store the rules in separate tables according to rule type, and extended fields are reserved to support iteration and differentiated configuration.

[0068] The rule base adopts a layered architecture of base layer, rule layer, and application layer: the base layer stores information such as the dimensions of the validation data and the units of the parameters (fields: parameter ID, name, unit, data type); the rule layer stores the core validation logic (fields: rule ID, validation SQL, threshold, exception conditions); the application layer associates instruction types and rules (fields: instruction ID, list of associated rule IDs); it uses DM8 table partitioning for storage, and reserves extended configuration JSON fields to support differentiated configurations and rule version iterations for different device models.

[0069] Step S1222: Review the verification dimensions, define thresholds and fault warning conditions based on relevant parameters, set risk scenario judgment rules and exception scenarios, enter key information into the rule base, and support differentiated configuration according to host model;

[0070] The system analyzes three verification dimensions: host load rate, runtime, and fault codes. For domestic centrifugal chillers (such as Gree LSBLG280H), it combines manufacturer parameters and on-site test data to define a high load threshold of 70% and a 15-minute stable period after startup. High-risk warning conditions are defined according to fault code levels (1-4). Logic risk judgment rules are set (any risk scenario triggering adaptation failure). Exception scenarios for high load verification exemption in emergency mode are configured, and the rules are entered into the tb_host_rule table, supporting filtering and calling by host model (rule data is synchronized to the domestic caching middleware Tencent Cloud CKV).

[0071] Step S1223: Determine the verification parameters, set the corresponding thresholds and grading judgment logic, configure differentiated parameters for different types of water pump groups, enter the relevant information into the rule base and mark the source of the parameters;

[0072] Three verification parameters were determined: pump frequency fluctuation, safe operating range, and flow supply and demand matching. For domestically produced cooling water pumps (such as Grundfos CR series compatible models and Wilo PW series domestic alternative models) and chilled water pumps, safe frequency fluctuation ranges (≤20Hz / ≤15Hz), safe frequency ranges (30-50Hz / 25-45Hz), and flow deviation thresholds (≤8% / ≤10%) were set respectively. A graded judgment + logical verification logic was adopted. The parameters were sourced from the manufacturer's energy efficiency curves / on-site measurements and entered into the tb_pump_rule table, which supports differentiated calls according to pump type.

[0073] Step S1224: Analyze the mapping relationship between instructions and linked devices, clarify the linkage status conditions and linkage priorities of associated devices, set judgment rules, and enter the rule base according to subsystem categories and support differentiated management;

[0074] System start-up, shutdown, and switching command types are categorized to create a command type and linked device mapping table (e.g., host start → cooling water pump + cooling tower fan); the linkage conditions (no fault, online, responsive) of associated devices (domestic frequency converter-driven water pump, domestic intelligent valve) are clearly defined, and the priority is configured according to core devices > auxiliary devices (host / main water pump is the core device); judgment rules are set (core devices that cannot be linked are directly rejected), and the data is entered into the tb_subsys_rule table according to the chilled water / cooling water subsystem, with fields including subsystem type, command ID, linked device list, and status requirements, supporting differentiated management (the mapping table is stored in the DM8 database).

[0075] Step S1225: Construct a mapping table to match the three types of verification rules for different instructions, avoiding invalid verification;

[0076] A JSON-formatted instruction and rule mapping table is constructed (stored in the application layer tb_cmd_rule_map table), and a corresponding verification rule ID is bound to each instruction type: for example, the host startup instruction (ID: 001) is associated with the host status rule (ID: 101), the cooling water pump group rule (ID: 201), and the cooling water subsystem linkage rule (ID: 301). After receiving the instruction, the AI ​​verification module (Huawei Atlas200IDKA2) accurately calls the corresponding rule according to the mapping table, avoiding verification redundancy caused by loading irrelevant rules and improving verification efficiency (the mapping table is cached in Alibaba Cloud PolarDB-X).

[0077] Step S1226: Develop a standardized calling interface. The rule base is stored using local caching and a remote database. When the module starts, it loads and verifies the integrity of the rule base. The accuracy is verified through simulation and the parameters are adjusted.

[0078] A standardized RESTful API (HTTP / HTTPS protocol) was developed. The rule base uses a dual storage mode of Alibaba Cloud PolarDB-X local cache + DM8 remote database. When the AI ​​module (Huawei Atlas200IDKA2) starts, it automatically loads the rule base and performs integrity verification (querying ≥10 rules in each table). A simulation verification script was written in Python, and typical operating conditions (such as host load 80% and water pump frequency 55Hz) were input to verify the accuracy of the judgment. When the false judgment rate is >3%, the threshold is automatically adjusted until the accuracy rate is ≥95% (the script runs in the domestic AI framework Baidu PaddlePaddle environment).

[0079] Step S123: Based on the preprocessed data, perform the host status, pump group matching degree, and subsystem linkage status verification in sequence according to the rule base, and mark the adaptability results of each link;

[0080] The AI ​​verification module (Huawei Atlas 200I DK A2) reads the pre-processed panoramic operation dataset and calls the corresponding rule base of DM8 in the order of host status → pump group matching degree → subsystem linkage status: it compares the host load rate with the threshold, pump frequency fluctuation with the safe range, and linkage equipment (domestic frequency converter, smart valve) status and requirements through SQL statements; after each verification is completed, it marks the pass / fail result and the reason (such as pump frequency fluctuation 25Hz > threshold 20Hz), and generates a JSON format verification report to ensure that the results of each link are traceable (the report is stored in KingbaseES V8R6).

[0081] Step S124: Call the historical operating condition database to retrieve historical records of similar system states and instruction types, and analyze the risk level of instruction execution;

[0082] The system calls the DM8 historical operating condition database (table tb_historical_conditions, fields: operating condition label, system load, equipment status, instruction type, risk level), and uses the Euclidean distance algorithm to retrieve the top 10 historical records that match the current system status (total load, supply and return water temperature difference) and instruction type. It analyzes indicators such as system fluctuation amplitude and equipment failure probability in the historical records, classifies the risk level into low (≤5% risk) / medium (5%-15%) / high (>15%), and outputs a risk analysis report (the report is synchronized to the TDengine time series database).

[0083] Step S125: Combining real-time verification results with historical analysis conclusions, determine whether the instruction is feasible or not according to preset logic, and assist in the determination of the redundancy capability of the accounting system in medium-risk scenarios.

[0084] If all real-time checks pass and historical low risk is detected, the instruction is feasible; if any real-time check fails or historical high risk is detected, the instruction is not feasible; if real-time checks pass but historical low risk is detected, the system redundancy capacity must be calculated (load redundancy = rated load - current load ≥ 20% is feasible). Redundancy capacity is calculated based on the rated load of the domestic host (e.g., Gree LSBLG280H 800RT) and the current operating load. The judgment result is output in the form of Boolean values ​​and cause descriptions to ensure that the judgment logic is quantifiable and executable.

[0085] Step S126: If feasible, send a verification pass signal; if not feasible, generate optimization suggestions including adjustment direction, parameters and execution order, display them on the terminal and provide a one-click optimization option.

[0086] If the verification is feasible, the AI ​​module (Huawei Atlas 200I DK A2) sends a 10ms high-level "verification passed" signal to the Huichuan H3U series hierarchical control module via the PCIe 4.0 link; if it is not feasible, it generates structured optimization suggestions (such as reducing the host load to 65% → adjusting the cooling water pump frequency to 45Hz → starting the cooling tower fan), including adjustment direction, parameter thresholds, and execution order, and pushes them to the Huichuan IT6000 series terminal via the Profinet bus. The DevEco Studio development interface displays the suggestions and sets a one-click optimization button, which automatically modifies the command parameters after clicking.

[0087] Step S127: Label the working condition tags of the entire verification process data and store them in the historical database. Analyze the historical data regularly and optimize the verification rule thresholds.

[0088] The entire verification process data (including panoramic dataset, verification report, judgment results, and optimization suggestions) is labeled with operating condition tags (such as high load, host startup) and stored in the DM8 historical database. The historical data is analyzed weekly using Python + Pandas (running on the domestic server operating system KylinOS), and the rule misjudgment rate is calculated (threshold ≤ 5%). If the misjudgment rate of a rule exceeds the standard, the threshold is automatically adjusted (such as optimizing the host high load threshold from 70% to 72%), the DM8 rule library is updated, and optimization logs are recorded (fields: optimization time, old threshold, new threshold, reason). The logs are synchronized to KingbaseES V8R6.

[0089] Step S130: If the verification passes, the instruction is broken down into levels and the execution rules are clarified. If it fails, optimization suggestions are generated and re-verification is supported after parameter adjustment. Feedback data is collected in real time during execution. In case of anomalies, the AI ​​module intervenes and issues an alarm. After the execution is completed, the system status is checked, closed-loop confirmation information is sent, and the entire process data is stored.

[0090] The system employs a collaborative processing approach using the Huichuan H3U series hierarchical control module and the Huawei Atlas 200I DK A2 AI verification module. Upon successful verification, the module breaks down instructions and defines execution rules at the system, subsystem, and device levels. If verification fails, the AI ​​module generates optimization suggestions with adjustment directions, supporting re-verification after parameter adjustments. During instruction execution, feedback data such as the frequency of the domestic inverter driving the water pump and the load of the domestic host unit are collected via the Profinet bus at a frequency of 300ms / time. The AI ​​module monitors this in real time, immediately pausing the process and implementing intervention measures such as switching to a backup pump and parameter callback when anomalies such as device response timeout (500ms) or pressure fluctuation ±15% are triggered. Simultaneously, alarms are pushed to the Huichuan IT6000 series terminals. After execution, the hierarchical control module checks the system pressure (0.4-0.6MPa), equipment fault codes, and other statuses, sends closed-loop confirmation information, and labels the entire process data with high load and instruction execution conditions, storing it in the DM8 historical database.

[0091] Step S131: After the AI ​​verification module outputs the results, the hierarchical control module automatically distributes the data. If the verification passes, the instruction hierarchical decomposition is initiated, clarifying the core content and execution rules of each level, assigning a unique identifier to the decomposed instruction and associating it with the original instruction. If the verification fails, the AI ​​module generates targeted optimization suggestions, provides a one-click optimization execution option, adjusts the parameters, and re-verifies until it passes.

[0092] The Huichuan H3U series hierarchical control module receives the verification results (pass flag 0x01 / fail flag 0x02) from the AI ​​module (Huawei Atlas 200I DKA2) via the PCIe 4.0 interface. Its built-in C-language-written traffic distribution logic (stored in the built-in organization block) automatically determines the status: if verification passes, it initiates a hierarchical instruction decomposition process, decomposing instructions by system-subsystem-device level and assigning unique UUID identifiers (e.g., generated by uuid.uuid1()), associating them with the original instruction parameters; if verification fails, the AI ​​module retrieves the DM8 historical operating condition database and DM8 rule database to pinpoint the failed dimension (e.g., host load 75% > threshold 70%), generating optimization suggestions to reduce the load to 65%. The terminal DevEco Studio interface provides a one-click optimization button, automatically correcting parameters and triggering re-verification (up to 3 times). Multiple failures trigger a manual intervention prompt. The module synchronously records the entire process status data, temporarily storing it in the Alibaba Cloud PolarDB-X cache, and archiving it to the DM8 historical database after successful execution.

[0093] Step S1311: The hierarchical control module receives the results from the AI ​​verification module through a standardized interface, completes the status determination with built-in diversion logic, and automatically triggers the corresponding processing flow;

[0094] The hierarchical control module (Inovance H3U series) is configured with a PCIe 4.0 standardized interface to receive verification result data frames (including verification status fields: 0x01 for pass, 0x02 for fail) output by the AI ​​module (Huawei Atlas200I DK A2). The built-in traffic splitting logic is written in C language and embedded in the built-in organization block. It scans interface data every 10ms and automatically triggers the corresponding process after completing the status determination: if the status is 0x01, it calls the instruction disassembly function (DisassembleCmd()); if the status is 0x02, it sends an "optimization suggestion generation" trigger signal (high level for 10ms) to the AI ​​module. The traffic splitting logic is set with a 100ms timeout protection; if the status is not identified, it marks "communication abnormal" and pushes an alarm to the Inovance IT6000 series terminal to ensure the reliability of process triggering.

[0095] Step S1312: Deconstruct the original instructions according to the three-level architecture of system level, subsystem level, and device level, clarify the core content of each level and filter redundant items, preset execution rule templates in combination with system characteristics, match the instruction type to set the execution order, interval and priority, assign a unique identifier to the deconstructed instructions, and establish an association mapping with the original instruction parameters to support traceability.

[0096] The system-level design clearly defines the collaborative goals of the chilled water system (e.g., "balanced total load distribution"). The subsystem-level design plans the linkage logic of the cooling water pump groups (e.g., "rotational start based on wear value"). The equipment-level design refines operating parameters (e.g., "adjusting the frequency of cooling water pump No. 1 (driven by Huichuan MD280 frequency converter) to 45Hz"). Redundant instructions such as "repeated frequency adjustment" are filtered during disassembly. An execution rule template is preset based on the characteristics of the air conditioning system: shutdown instructions are executed in reverse order of "equipment-subsystem-system," with a 20ms interval between equipment actions. Instructions from core equipment (host) have higher priority than those from auxiliary equipment (domestic intelligent valves). A unique identifier (e.g., f81d4fae-7dec-11d0-a765-00a0c91e6bf6) is assigned to each disassembly instruction using the UUID algorithm. An association mapping between the identifier and the original instruction parameters (operation object, target action) is established in the DM8 tb_cmd_mapping table, supporting reverse tracing.

[0097] Step S1313: The AI ​​module combines the historical database and rule base to locate the problem in the failed dimension and generate structured optimization suggestions containing adjustment direction, parameter range and priority. The terminal displays the suggestions and provides a one-click optimization function. After automatically replacing the command parameters, it triggers re-verification. The process is repeated until it passes. If it fails multiple times, it prompts for manual intervention.

[0098] The AI ​​module (Huawei Atlas 200I DK A2) accesses the DM8 historical operating condition database (including execution records of similar commands over the past year) and the DM8 verification rule base. Using an Euclidean distance algorithm, it identifies the core issues that failed (e.g., host high load 75% > threshold 70%) and generates structured optimization suggestions: adjustment direction (reduce host load), parameter range (60%-65%), and execution priority (prioritizing subsystem linkage). The suggestions are displayed in the DevEco Studio interface of the Huichuan IT6000 series terminal, with a blue one-click optimization button. Clicking this button automatically extracts parameters to replace the original command (e.g., changing the load from 75% to 65%) and triggers a re-verification via the Profinet bus. The re-verification process cycles a maximum of three times. If it fails three times, a red prompt requiring manual intervention appears on the terminal, pausing the automatic process.

[0099] Step S1314: The hierarchical control module records the key status and data of the entire process of diversion processing in real time, binds the instruction identifier and stores it in a temporary cache, and archives it to the historical database after the instruction is executed.

[0100] The hierarchical control module (Inovance H3U series) records key statuses of the traffic splitting process in real time: result reception time (accurate to milliseconds), dismantling completion time, optimization suggestion generation time, parameter adjustment records, and verification loop count. All data is uniquely identified by the instruction UUID. A temporary cache area (key is instruction UUID, value is JSON format status data) is built using Alibaba Cloud PolarDB-X, with a cache validity period of 1 hour to avoid data loss. After the instruction is successfully executed, the module automatically calls the archive function (ArchiveData()) to synchronously write the cached data to the DM8 tb_process_record table (fields: instruction UUID, status data, archive time, working condition label), ensuring that the entire process data is traceable and auditable.

[0101] Step S132: The hierarchical control module sends out the disassembled instructions at each level through the industrial bus. After receiving the instructions, the execution unit executes the action and reports the status. At the same time, it collects key feedback data and synchronizes it to the hierarchical control module and the AI ​​module to form a data link.

[0102] The Huichuan H3U series hierarchical control module uses the Profinet industrial bus (100Mbps transmission rate) to send the disassembled instructions at each level to the corresponding execution units: system-level instructions are sent to domestic system co-controllers (such as Xinje XC5 series), subsystem-level instructions are sent to Advantech Adam-6000 series cooling subsystem controllers, and device-level instructions are sent to Huichuan MD280 series domestic water pump frequency converters. After receiving the instructions, the execution unit provides real-time feedback on the execution status ("Received" 0x01 / "Executing" 0x02 / "Completed" 0x03 / "Failed" 0x04), and simultaneously collects key feedback data: actual water pump operating frequency (accuracy 0.1Hz), host load rate (accuracy 1%), and subsystem flow matching degree (accuracy 0.5%), which are synchronized to the hierarchical control module (local Flash storage) and AI verification module (Alibaba Cloud PolarDB-X real-time cache) via the bus at a frequency of 300ms / time, forming a closed-loop data chain.

[0103] Step S133: The AI ​​module monitors feedback data and execution status in real time, presets three types of abnormal judgment criteria, suspends the process after an abnormality is triggered, takes targeted intervention measures, generates alarm information to push to the operator, and supports manual intervention.

[0104] The Huawei Atlas 200I DK A2 AI module monitors feedback data and execution status in real time via the Profinet bus, and presets three types of anomaly judgment criteria: equipment execution anomaly (no response within 500ms after command issuance / action deviation > 10%), parameter out-of-bounds anomaly (system pressure 0.8MPa > safety threshold 0.6MPa / equipment frequency 55Hz > energy efficiency limit 50Hz), and system fluctuation anomaly (supply and return water temperature difference 5℃ > threshold 3℃). Upon triggering an anomaly, the AI ​​module immediately sends a "process pause" high-level signal (lasting 10ms) to the hierarchical control module (Inovance H3U series), matching the DM8 intervention strategy library to take measures: switching the No. 2 backup cooling water pump (driven by Inovance MD280 frequency converter) when the equipment does not respond; adjusting the pressure to 0.5MPa when parameters exceed limits; and fine-tuning the opening of the domestic intelligent valve to 80% when the system fluctuates. Simultaneously, standardized alarm information is generated and pushed to the Huichuan IT6000 series terminal (pop-up window + 3-second beep), supporting operators to manually intervene via terminal buttons. After the anomaly is handled, all data is recorded to the DM8 historical database.

[0105] Step S1331: The AI ​​module establishes a two-way communication link with the hierarchical control module and each execution unit through the industrial bus, receives feedback data and execution status in real time, synchronizes data at a preset frequency and temporarily stores it through the built-in data buffer queue to ensure real-time monitoring and data integrity.

[0106] The AI ​​module (Huawei Atlas 200I DK A2) establishes a bidirectional communication link with execution units such as the Huichuan H3U series hierarchical control module, Advantech Adam-6000 series subsystem controller, and Huichuan MD280 series frequency converter via a Profinet dual-bus (primary and backup redundancy). It receives two types of data in real time: execution status ("received / in execution / completed") and feedback data (actual device frequency, host load, system pressure, and subsystem flow matching degree). The data synchronization frequency is set to 300ms / time. The module has a built-in Python queue.Queue() buffer queue (capacity 1024) to temporarily store the received data stream and avoid data loss due to bus latency. The queue adopts a "first-in, first-out" mechanism, and the AI ​​module parses the data sequentially to ensure the real-time performance and integrity of the monitoring (Python environment runs on KylinOS).

[0107] Step S1332: Preset equipment execution abnormality standards, parameter out-of-bounds abnormality standards, and system fluctuation abnormality standards. All judgment thresholds can be adjusted as needed.

[0108] Three configurable anomaly judgment criteria are preset: ① Equipment execution anomaly: The response timeout threshold after the command is issued is 500ms, and the deviation between the actual action of the equipment and the command requirement is >10% (e.g., the command adjusts the frequency to 45Hz, but the actual operation is 50Hz); ② Parameter out-of-bounds anomaly: The system pressure safety threshold is 0.4-0.6MPa, and exceeding ±15% is considered out of bounds; the equipment operating frequency energy efficiency range is 30-50Hz, and exceeding it will trigger an alarm; ③ System fluctuation anomaly: The stable threshold for the supply and return water temperature difference is ≤3℃, and fluctuations >2℃ and lasting for 5 seconds are considered abnormal. All thresholds are stored in the DM8 tb_threshold table and can be manually modified through the DevEco Studio terminal interface (e.g., adjusting the pressure threshold to 0.3-0.7MPa) to adapt to different operating conditions.

[0109] Step S1333: The AI ​​module parses the buffer queue data in real time and compares it with the preset abnormal criteria. If any abnormal condition is met, the abnormal response mechanism is immediately triggered, and a process pause command is sent to the hierarchical control module. After receiving the command, the hierarchical control module pauses the issuance and execution of the current command and freezes the subsequent process.

[0110] The AI ​​module (Huawei Atlas 200I DK A2) parses the feedback data in the buffer queue at a frequency of 200ms / time, comparing each field with the preset standards in the DM8 tb_threshold table using a Python script (e.g., ifsys_pressure>max_pressure:trigger_exception()). When any abnormal condition is met, the module triggers an exception response mechanism within 10ms, sending a "process pause" command (high-level signal, lasting 10ms) to the Huichuan H3U series hierarchical control module. Upon receiving the signal, the hierarchical control module immediately freezes the issuance of subsequent commands (achieved by setting the "process pause" register 0x01), pausing the current execution process to prevent the exception from escalating and affecting system stability.

[0111] Step S1334: The AI ​​module uses the anomaly type matching intervention strategy library to take corresponding intervention measures such as device switching / command retry / parameter correction, parameter callback / associated device pause, and associated parameter fine-tuning / dynamic balancing.

[0112] The AI ​​module (Huawei Atlas 200I DK A2) parses the feedback data in the buffer queue at a frequency of 200ms / time, comparing each field with the preset standards in the DM8 tb_threshold table using a Python script (e.g., ifsys_pressure>max_pressure:trigger_exception()). When any abnormal condition is met, the module triggers an exception response mechanism within 10ms, sending a "process pause" command (high-level signal, lasting 10ms) to the Huichuan H3U series hierarchical control module. Upon receiving the signal, the hierarchical control module immediately freezes the issuance of subsequent commands (achieved by setting the "process pause" register 0x01), pausing the current execution process to prevent the exception from escalating and affecting system stability.

[0113] Step S1335: While the intervention is being executed, the AI ​​module generates standardized alarm information containing the anomaly type, location, cause, intervention measures taken, and suggested actions. This information is pushed through the system operation terminal via pop-up windows and sound prompts, and the alarm logs are simultaneously written to the storage and historical database.

[0114] While executing intervention measures, the AI ​​module (Huawei Atlas 200I DK A2) generates standardized alarm information (JSON format), including the anomaly type (e.g., "parameter out of bounds anomaly"), anomaly location ("No. 1 domestic centrifugal chiller"), anomaly cause ("overload causing system pressure to exceed limits"), intervention measures taken ("reduce host load to 60%), and suggested operation ("check host heat exchanger cleanliness"). The alarm information is pushed to the Huichuan IT6000 series terminal via the Profinet bus: a red-bordered alarm pop-up appears in the DevEco Studio interface, and the terminal buzzer sounds continuously for 3 seconds; simultaneously, the alarm log is written to the DM8tb_alarm_log table (fields: alarm time, anomaly details, handling measures, command UUID) to ensure anomaly traceability.

[0115] Step S1336: The system operation terminal is equipped with emergency stop, parameter adjustment, and resume execution interactive buttons, which support operators to terminate the process, modify associated parameters, and resume execution. All manual operations are recorded with the operator, time, and content and are bound to alarm information for archiving.

[0116] The Huichuan IT6000 series terminal DevEco Studio interface is configured with three types of manual intervention interaction buttons: ① Red Emergency Stop Button: Clicking this button immediately terminates all execution processes and sends a stop command to the hierarchical control module (Huichuan H3U series); ② Parameter Adjustment Area: Includes a numeric input box (e.g., adjustable water pump frequency from 30-50Hz) and a blue Confirm Submit button. After parameter modification, the changes are synchronized to the AI ​​module (Huawei Atlas 200I DK A2) and the hierarchical control module; ③ Green Resume Execution Button: Clicking this button after the anomaly has subsided triggers the process to continue. All manual operations are automatically recorded in the DM8 tb_manual_op table, with fields including the operator (login account OP001), operation time (accurate to milliseconds), and operation content (e.g., adjusting the frequency of water pump No. 1 to 40Hz), and are archived along with alarm information.

[0117] Step S1337: The AI ​​module continuously monitors the feedback data and execution status after intervention or manual operation. When the data returns to the normal range, it sends a process recovery instruction to the hierarchical control module to resume execution. If the abnormality is not resolved, it maintains the paused state and pushes alarms repeatedly on a regular basis.

[0118] The AI ​​module (Huawei Atlas 200I DK A2) monitors feedback data and execution status after intervention or manual operation at a frequency of 200ms / time. It determines the conditions for data to return to normal: system pressure stabilizes at 0.4-0.6MPa, equipment frequency deviation ≤5%, and supply and return water temperature difference ≤3℃ for 10 seconds. When these conditions are met, the module sends a low-level signal for process recovery to the hierarchical control module (Inovance H3U series). The hierarchical control module clears the process pause register and resumes the instruction execution process. If the anomaly is not resolved, the module maintains the process pause state and repeatedly pushes an alarm message to the Inovance IT6000 series terminal every 30 seconds until the anomaly is resolved or the operator confirms the handling.

[0119] Step S1338: Bind the time of occurrence, type, triggering condition, intervention measures, manual operation record, release time and processing result of the anomaly to the corresponding instruction unique identifier and store them in the historical database.

[0120] The AI ​​module (Huawei Atlas 200I DK A2) organizes the entire abnormal process data, binds it with the unique instruction UUID, and stores it in the DM8 tb_exception_record table. Specific fields include: abnormal occurrence time (accurate to milliseconds), abnormal type (e.g., "system fluctuation abnormality"), triggering condition ("supply and return water temperature difference 5℃ > threshold 3℃"), intervention measures ("fine-tuning the opening of the domestic intelligent valve to 80%), manual operation record (none / "adjusting the load to 60%), abnormal resolution time, and processing result ("system returned to stability"). Before data storage, MD5 encryption is used to verify integrity. After archiving, an abnormal handling report (PDF format) is generated and synchronized to a domestic cloud server (e.g., Huawei Cloud OBS) to provide data support for subsequent analysis of similar abnormalities.

[0121] Step S134: After each execution unit completes its feedback, the hierarchical control module checks the system status. If it passes, the execution is deemed complete and a closed-loop confirmation message is sent. The entire process data is then summarized, labeled with operating condition tags, and stored in the historical database.

[0122] After each execution unit (domestic frequency converter, intelligent valve, etc.) reports "execution completed" (marked 0x03), the Huichuan H3U series hierarchical control module initiates a system status check: It reads key indicators such as system pressure (0.4-0.6MPa), equipment fault codes (0 for no fault), and subsystem flow matching degree (≥90%) via the Profinet bus. If all indicators meet the criteria, execution is considered complete. The module immediately sends a closed-loop confirmation message ("Instruction execution completed, system status stable") to the Huichuan IT6000 series terminal, and simultaneously summarizes all process data (instruction breakdown details, execution feedback data, and anomaly records), labels it with the "High Load - Cooling Water Pump Frequency Adjustment" operating condition tag, and stores it in the Dameng DM8 tb_full_process table. The data is stored in a structured manner according to "Instruction UUID - Stage - Data Content," supporting rapid retrieval by operating condition tag and time range.

[0123] Step S140: When the system performs power-on, power-off, and load adjustment, it incorporates various types of data obtained by the AI ​​module, associates them with the current equipment load status of each subsystem, and ensures that the system adjustment commands drive the subsystems to respond in a coordinated manner.

[0124] The system employs a Huichuan H3U series system-level controller in collaboration with a Huawei Atlas 200I DK A2 AI module. During power-on, power-off, and load adjustment, the AI ​​module integrates multiple types of data via the Profinet bus: device-level (load of the domestic host, frequency of the water pump driven by the domestic inverter), subsystem-level (flow matching degree), system-level (total energy efficiency ratio), and wear data from the equipment health database. It correlates the current load status of subsystems such as cooling and chilled water (e.g., cooling subsystem load rate 65%) and formulates coordinated response logic: upon power-on, it first triggers the pre-start of the cooling water pump (frequency 40Hz), and after 30 seconds, it starts the domestic host to load to the target load; during load adjustment, it issues commands according to subsystem load balancing and equipment action sequence, ensuring that system commands drive subsystems to respond synchronously, avoiding single-device overload or linkage lag.

[0125] Step S150: After receiving the system-level instruction, the subsystem decomposes it into appropriate sub-instructions based on its own equipment operating parameters and sends them to specific equipment. At the same time, it collects the equipment operating status and feeds it back to the system level, forming a closed-loop feedback between the system and the subsystem.

[0126] The subsystem uses Advantech Adam-6000 series domestic controllers. After receiving system-level commands from Huichuan H3U series (such as loading the cooling subsystem by 10%), it combines its own equipment operating parameters (current water pump frequency 42Hz, domestic intelligent valve opening 75%) and decomposes them into detailed commands: Adjusting the frequency of cooling water pump No. 1 (driven by Huichuan MD280 frequency converter) to 47Hz, and adjusting the opening of the domestic intelligent valve in the cooling water pipeline to 85%. These commands are then sent to the Huichuan MD280 frequency converter and the domestic electric valve controller (Shanghai Automation Instrument Factory No. 11 ZJK series) via Modbus RTU protocol. Simultaneously, the system collects equipment execution status (frequency adjustment in progress, opening meets standard) and operating parameters (actual frequency, opening) at a frequency of 300ms / time, feeding them back to the system level via Profinet bus, forming a two-way closed-loop feedback between the system and subsystem.

[0127] Step S160: After receiving the detailed instructions, the devices in the subsystem convert them into specific operating parameters, collect relevant data about their own operation in real time, and synchronously feed them back to the subsystem and the system-level AI module.

[0128] After receiving detailed instructions, the equipment within the subsystem (such as domestically produced cooling water pumps and ZJK series electric valves from Shanghai Automation Instrumentation Factory No. 11) converts the instructions into specific operating parameters through the controller: the pump instruction is frequency-modulated to 47Hz and converted into the output frequency signal of the Huichuan MD280 frequency converter; the valve instruction "opening degree 85%" is converted into a 4-20mA current signal. The built-in sensors of the equipment collect operating data in real time (pump operating current 5.2A, vibration value 2.3mm / s, actual valve opening degree 84.7%), and synchronously feed it back to the subsystem controller (Advantech Adam-6000 series) via Modbus RTU, and then upload it to the system-level AI module (Huawei Atlas 200I DK A2) via the industrial bus to support real-time control and decision-making.

[0129] Step S170: Establish a real-time data link, acquire and store various relevant data in parallel, complete verification and optimization before the execution of manual instructions, and when there is no human intervention, the AI ​​module autonomously formulates control strategies at the system, subsystem and equipment levels to achieve equipment wear balance. During the control execution, real-time verification and dynamic fine-tuning are performed, and all data are stored afterward. The AI ​​module regularly analyzes historical samples, identifies the optimal control strategy and updates the model.

[0130] Domestic industrial edge data acquisition nodes (compatible with the Profinet protocol) collect multi-dimensional data and transmit it to the Alibaba Cloud PolarDB-X real-time library and the DM8 historical library via the Profinet bus. Before manual command execution, the AI ​​module (Huawei Atlas 200IDK A2) verifies, optimizes, and supports re-verification. Without intervention, the AI ​​generates three-dimensional evaluation results periodically, formulates system-subsystem-equipment level wear leveling strategies, and issues them after verification by the domestic simulation tool ZW3D. During execution, data is collected and dynamically fine-tuned at a frequency of 200ms / time, and all data are archived afterward. Every week, the AI ​​module analyzes historical samples through clustering algorithms, updates the Huawei MindSpore control model, and retains the old model for backup, realizing intelligent control of the entire process (the algorithm runs on the Baidu PaddlePaddle framework).

[0131] Step S171: Construct a three-level link consisting of edge acquisition nodes, industrial bus, and central storage layer. Edge nodes are directly connected to devices and subsystems, and multi-dimensional data is collected in parallel. After preprocessing, the data is stored in layers, and a retrieval index is established by associating identifiers.

[0132] A three-tiered link system was constructed, consisting of domestically developed industrial edge data acquisition nodes, a Profinet industrial bus, and a central storage layer. Edge nodes directly connect to devices and subsystems, acquiring multi-dimensional data at the device level (operating parameters, wear values), subsystem level (interconnection status), and system level (total load) in parallel. Differentiated acquisition frequencies were set (vibration value 100ms / time, load rate 500ms / time). After preprocessing including 3σ outlier removal, linear interpolation completion, and Min-Max standardization, the data was pushed to tiered storage: Alibaba Cloud PolarDB-X stores high-frequency data for the past hour, while TaoSi TDengine archives historical data daily. A retrieval index was established using associated device IDs and subsystem identifiers, supporting fast queries by operating condition tags (index stored in DM).

[0133] Step S172: After receiving the manual instruction, the AI ​​module retrieves real-time and historical data, verifies the adaptability of the instruction and the impact of equipment wear, generates optimization suggestions, executes them after confirmation, and adjusts the parameters for re-verification if they fail.

[0134] After receiving manual commands, the system operation terminal (Siemens TP177B) uses an AI module to retrieve current system status data from the Redis real-time database and similar command records from the MySQL historical operating condition database over the past year via a RESTful interface. It then uses an Euclidean distance algorithm to verify command compatibility (e.g., whether start-stop is allowed under high host load) and combines this with equipment health database data to assess wear impact (e.g., whether frequent start-stops exacerbate bearing wear), generating structured optimization suggestions (e.g., adjusting execution timing, replacing patrolling equipment). Commands are executed only after operator confirmation. If verification fails, parameters can be adjusted (e.g., extending the start-stop interval), and verification can be retried until compatibility and wear leveling requirements are met.

[0135] Step S173: AI evaluates the system status according to a preset cycle, generates three-dimensional evaluation results, clarifies the control target of prioritizing wear equalization, formulates strategies at different levels and verifies them through simulation;

[0136] The AI ​​module (NVIDIA Jetson Xavier NX) evaluates system status every 5 minutes by default, supplemented by a trigger mechanism (immediately triggered when load fluctuation > 10% or wear rate > 0.01 / h). It retrieves real-time and historical wear data through a standardized interface, preprocesses it to generate an evaluation dataset, and calculates three-dimensional results: system health (0-100 points), load distribution uniformity (0-1), and equipment wear evenness (0-1). Based on the principle of prioritizing wear evenness, it refines and quantifies control targets, formulates strategies at the system-subsystem-equipment level, and calls the MATLAB / Simulink simulation engine to simulate the execution effect. If the target is not met, parameters are adjusted and the simulation is repeated. Upon successful simulation, the results are pushed to the hierarchical control module; multiple failures trigger manual intervention.

[0137] Step S1731: The AI ​​module can be configured with a customizable evaluation cycle and supplementary triggering mechanism. When the cycle is reached or the triggering condition is met, the evaluation process will be started automatically and the dedicated data retrieval thread will be awakened.

[0138] The AI ​​module has a built-in Python period configuration script that supports manually customizing the evaluation period (adjustable from 1 to 60 minutes, default 5 minutes). It also includes a supplementary trigger mechanism: evaluation is automatically triggered when system load fluctuations exceed 10% or equipment wear rate exceeds 0.01 / h. When the period is reached or the trigger condition is met, the module wakes up a dedicated data retrieval thread (with the highest priority), which only uses CPU resources during the evaluation phase to avoid unnecessary resource consumption. Data retrieval preparation is completed within 10ms after thread startup, ensuring a rapid start of the evaluation process.

[0139] Step S1732: The AI ​​module retrieves multi-dimensional status data at the device level, subsystem level, and system level from the real-time database through a standardized interface, as well as historical wear data from the device health database. After cleaning, feature extraction, and dimensional unification processing, a standardized evaluation dataset is generated.

[0140] The AI ​​module (Huawei Atlas 200I DK A2) retrieves multi-dimensional status data at the device, subsystem, and system levels from a real-time database via standardized interfaces, as well as historical wear data from the device health database. After cleaning, feature extraction, and dimensional unification, a standardized evaluation dataset is generated. The AI ​​module also retrieves full data from the Alibaba Cloud PolarDB-X real-time database for the past period, including device-level (operating parameters, start / stop records), subsystem-level (load deviation), and system-level (total energy efficiency) data, and historical data such as cumulative wear values ​​and fault records from the DM8 device health database via a RESTful standardized interface. The data is then cleaned (outliers removed, missing values ​​filled), features extracted (wear growth rate and load deviation rate calculated), and dimensional unification performed (mapping to the 0-1 range), ultimately generating a standardized evaluation dataset in CSV format. This dataset includes parameter names, values, collection time, and associated identifier fields, supporting subsequent three-dimensional evaluation (the dataset is cached on Tencent Cloud CKV).

[0141] Step S1733: Calculate the system health, load distribution uniformity, and equipment wear uniformity respectively, and integrate them to generate a three-dimensional assessment report of health, load distribution, and wear to quantify the core issues;

[0142] The system health, load distribution uniformity, and equipment wear evenness are calculated separately, and integrated to generate a three-dimensional assessment report of health, load distribution, and wear, quantifying core issues. The AI ​​module (Huawei Atlas 200I DK A2) generates the three-dimensional assessment results according to a preset algorithm: System health is calculated using a weighted summation (failure rate 30%, energy efficiency ratio 40%, linkage response rate 30%), outputting a score of 0-100 and classifying it as excellent / good / requiring intervention; Load distribution uniformity is calculated by determining the load deviation rate of domestic equipment in the same group, outputting a value in the 0-1 range to identify overloaded / idle equipment; Equipment wear evenness integrates runtime, vibration value, and wear coefficients from domestic manufacturers to calculate the cumulative wear value and growth rate of a single piece of equipment, outputting an evenness value in the 0-1 range. Finally, a JSON-formatted three-dimensional assessment report of health, load distribution, and wear is generated, quantifying and annotating core issues (e.g., wear evenness of domestic host #1 is 0.65, requiring load adjustment), and the report is stored in the DM8 database.

[0143] Step S1734: Based on the wear equalization priority principle and the three-dimensional evaluation results, refine the core, secondary and auxiliary quantitative control targets;

[0144] Based on the wear leveling priority principle and the three-dimensional evaluation results, the core, secondary, and auxiliary quantitative control targets are refined. The wear leveling priority principle is solidified based on the DM8 preset rule base, and the quantitative control targets are further refined in conjunction with the three-dimensional evaluation results: when the wear leveling degree of domestic equipment is <0.7, the core target is a wear difference of ≤10% for a single group of equipment; when the system health score is <60, the secondary target is to improve the health score to above 60; when the load distribution uniformity is <0.6, an auxiliary target of load deviation rate ≤15% is added. All targets have clearly defined quantitative indicators and judgment criteria to ensure that the control process is assessable and verifiable, avoiding ambiguous requirements (target parameters are synchronized to the domestic cache middleware).

[0145] Step S1735: Develop control strategies at the system level, subsystem level, and equipment level, clarify the overall direction, linkage logic, and operating parameters, and ensure that the parameters meet the equipment safety requirements;

[0146] Control strategies are formulated at the system, subsystem, and equipment levels, clarifying the overall direction, linkage logic, and operating parameters to ensure that parameters meet equipment safety requirements. The control strategies are further structured at the system, subsystem, and equipment levels: at the system level, the overall direction is clearly defined (e.g., equalizing the load distribution of the chilled water system and reducing the wear difference of domestic main units to within 10%); at the subsystem level, linkage logic is planned (e.g., domestic cooling water pumps operate in rotation according to wear values, and the frequency of newly commissioned pumps is gradually increased by 5Hz / minute); at the equipment level, operating parameters are refined (e.g., when the load of domestic main unit 1 decreases from 65% to 55%, and the load of domestic main unit 2 increases from 50% to 60%, the corresponding opening of the domestic intelligent valve is adjusted to 82%). All operating parameters strictly adhere to the safety thresholds of domestic equipment (e.g., pump frequency 30-50Hz, main unit load 40%-80%) to avoid operating outside these ranges.

[0147] Step S1736: The AI ​​module calls the system simulation engine, imports the current state data and control strategy to simulate the execution effect. If the target is not met, the parameters are adjusted and the simulation is repeated until all control targets are met.

[0148] The AI ​​module (Huawei Atlas 200I DK A2) calls the system simulation engine, imports current state data and control strategies to simulate the execution effect. If the target is not met, the parameters are adjusted and the simulation is restarted until all control objectives are met. The AI ​​module also calls the domestic simulation tool ZW3D system simulation engine, imports current system state data (such as the load and wear values ​​of various domestic equipment) and the formulated control strategies, and simulates the strategy execution effect within 10 minutes. The simulation process focuses on verifying three indicators: whether the wear difference of domestic equipment is ≤10%, whether the system operating parameters exceed the safety threshold, and whether the secondary / auxiliary objectives are achieved. If the target is not met, the strategy parameters are automatically adjusted (such as reducing the frequency regulation rate of domestic water pumps to 3Hz / minute and fine-tuning the load distribution ratio of domestic host), the simulation is restarted, and the cycle continues until all control objectives are met, ensuring the feasibility of the strategy (the simulation log is stored in KingbaseES V8R6).

[0149] Step S1737: After the simulation passes, the standardized and encapsulated control strategy is pushed to the hierarchical control module. At the same time, the evaluation results, strategy content and simulation report are associated and stored. If the simulation fails to meet the target multiple times, a manual intervention prompt is triggered.

[0150] After a successful simulation, the standardized and encapsulated control strategy is pushed to the hierarchical control module. Simultaneously, the evaluation results, strategy content, and simulation report are stored together. Manual intervention is triggered if the simulation fails multiple times. After a successful simulation, the AI ​​module (Huawei Atlas 200I DK A2) encapsulates the control strategy in a standardized JSON format (including strategy ID, hierarchical instructions, execution parameters, and target value) and pushes it to the Huichuan H3U series hierarchical control module via a PCIe 4.0 link. Simultaneously, the 3D evaluation report, strategy content, and simulation results (including simulation curves) are stored together in the DM8 historical database and bound to a unique instruction identifier. If three consecutive simulations fail to meet the requirements, the module immediately pushes a red "Manual intervention required" alarm pop-up to the Huichuan IT6000 series terminal via the Profinet bus, pausing the automatic control process and awaiting operator intervention.

[0151] Step S174: After the strategy is issued, the execution data is collected in real time and compared with the target value. Dynamic fine-tuning is performed for scenarios such as parameter drift, wear loss balance, and system fluctuation.

[0152] After the strategy is issued, execution data is collected in real time and compared with the target value. Dynamic fine-tuning is performed for scenarios involving parameter drift, wear loss balance, and system fluctuations. After the control strategy is issued, domestic industrial edge acquisition nodes collect execution data at a frequency of 200ms / time: actual operating parameters of domestic equipment (pump frequency, valve opening), subsystem linkage status, and wear trend changes, which are transmitted back to the AI ​​module (Huawei Atlas200IDKA2) and the hierarchical control module (Huichuan H3U series) in real time. The AI ​​module compares and verifies the feedback data with the strategy target value (such as load deviation ≤5%, wear rate increase ≤0.01 / h): when equipment parameters drift (deviation >8%), a correction command is issued; when the wear rate increase is abnormally high (exceeding the average of the same group by 20%), domestic backup equipment is switched and the load is adjusted; when the system fluctuates (supply and return water temperature difference >2℃), parameter adjustment is paused, and subsystem linkage is used for stabilization and control. Execution resumes after the operating conditions stabilize (fine-tuning data is synchronized to Alibaba Cloud PolarDB-X).

[0153] Step S175: After the adjustment is completed, classify and sort the data of the entire process, label the working conditions, establish related indexes, and archive them to the corresponding database;

[0154] After the control is completed, all data is categorized and organized, labeled with operating conditions, and associated indexes are established and archived to the corresponding databases. After the control is executed, the system automatically categorizes and organizes all data across the entire process: instruction / strategy dimension (human instructions / AI autonomous strategies, verification records, optimization suggestions), execution process dimension (instruction issuance, response latency, parameter adjustment trajectory), status change dimension (system load, energy efficiency curve), and wear dimension (wear value changes, vibration trends). All data is labeled with operating conditions (e.g., high-load period - wear leveling control), and associated indexes are established for instructions / strategies, execution processes, and status results. Real-time data is archived to the TaoSi TDengine time-series database, and wear characteristic data is synchronized to the Dameng DM8 equipment health database to support subsequent model training and analysis (index information is stored in the Dameng DM8).

[0155] Step S176: Periodically screen samples, compare the effects of different strategies, identify the optimal strategy and update the AI ​​model, deploy it after simulation verification, and keep the old model for later use.

[0156] Regularly screen samples, compare the effects of different strategies, identify the optimal strategy and update the AI ​​model. Deploy the model after simulation verification, and retain the old model for backup. A fixed analysis cycle of once a week is set. The AI ​​module (Huawei Atlas 200I DK A2) automatically filters historical samples: removing invalid samples due to abnormal operating conditions, and retaining valid samples with successful strategy execution, wear leveling achieved, optimal energy efficiency, and typical abnormal samples. The K-means clustering algorithm (running on the Baidu PaddlePaddle framework) is used to compare the wear leveling, energy efficiency ratio, and stability of different strategies under the same operating conditions to identify the optimal control strategy (such as the combination of domestic equipment in a specific load range). Strategy features are extracted to update the Huawei MindSpore AI control model. The model's effectiveness is verified through offline simulation using the domestic simulation tool ZW3D (accuracy ≥95%). After verification, it is deployed to the online control module, while the old model is retained as a backup to ensure the iteration process is safe and controllable (model files are stored in Huawei Cloud OBS).

[0157] Based on the same inventive concept, please refer to Figure 2 This diagram illustrates a schematic block diagram of a hierarchical control system 100 for air conditioning cold source systems, provided in an embodiment of this application, for executing the aforementioned hierarchical control method for air conditioning cold source systems. The hierarchical control system 100 may include a communication unit 110, a machine-readable storage medium 120, and a processor 130. Alternatively, the machine-readable storage medium 120 may be integrated into the processor 130 and can communicate with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the hierarchical control method for air conditioning cold source systems provided in the aforementioned method embodiments.

[0158] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A hierarchical control method suitable for air conditioning cold source systems, characterized in that, The method includes: The system operation terminal triggers an instruction containing the operation object, target action, and execution time parameters, which is transmitted in real time to the hierarchical control module via the industrial bus and the AI ​​verification module is started simultaneously. The AI ​​verification module performs multi-dimensional adaptability verification based on the system's panoramic operation data, and verifies the feasibility of instructions by combining historical operating condition databases. If the verification passes, the instructions are broken down into levels and the execution rules are clarified. If the verification fails, optimization suggestions are generated and re-verification is supported after parameter adjustment. Feedback data is collected in real time during execution. In case of anomalies, the AI ​​module intervenes in an emergency and issues an alarm. After the execution is completed, the system status is checked, closed-loop confirmation information is sent, and the entire process data is stored. When the system performs power-on, power-off, and load-reduction adjustments, it incorporates various types of data acquired by the AI ​​module and correlates them with the current device load status of each subsystem to ensure that the system's control commands drive the subsystems to respond in a coordinated manner. After receiving instructions from the system level, the subsystem decomposes them into appropriate sub-instructions based on its own equipment operating parameters and sends them to specific equipment. At the same time, it collects equipment operating status and feeds it back to the system level, forming a closed-loop feedback between the system and subsystems. After receiving detailed instructions, the devices within the subsystem convert them into specific operating parameters, collect relevant data about their own operation in real time, and synchronously feed them back to the subsystem and the system-level AI module. Establish a real-time data link, acquire and store various relevant data in parallel, complete verification and optimization before manual instructions are executed, and when there is no human intervention, the AI ​​module autonomously formulates control strategies at the system, subsystem and equipment levels to achieve equipment wear balance. During the control execution, real-time verification and dynamic fine-tuning are performed, and data from all stages are stored afterward. The AI ​​module regularly analyzes historical samples, identifies the optimal control strategy and updates the model. The system operation terminal has a visual interface with preset options for operation objects, target actions, and execution time. After the user inputs the parameters, the terminal verifies the completeness and legality of the parameters. Once verified, the user triggers the command, and the terminal simultaneously records the operator, trigger time, and other additional information. Instructions are encapsulated using a unified data frame format, including fields, and the data is encoded according to the transmission protocol adapted to the industrial bus after encapsulation. An industrial Ethernet bus is selected, an interface module is configured, and a dual-bus redundancy design is adopted. The transmission mechanism and bandwidth allocation strategy ensure fast command transmission. The module is configured to adapt to the communication interface, open an independent buffer, receive data frames, verify their legality and parse the parameters, and mark the instruction status after verifying their validity. The hierarchical control module and the AI ​​verification module are connected through an internal high-speed communication link. After successfully parsing the instruction, a start signal is sent, and the AI ​​verification module completes initialization and returns the ready status. After receiving the ready feedback, the hierarchical control module sends a success message back to the terminal. If no feedback is received within the timeout period, an abnormality is determined, an alarm is pushed and command execution is suspended, and the command will be retried after investigation. The AI ​​verification module acquires operating condition data of the host, water pump, subsystem, electric valve and the whole system through the data acquisition unit. After being transmitted synchronously through the industrial bus, it forms a structured panoramic operation dataset through outlier removal, missing value completion and standardization transformation. Three types of verification rule bases are preset to clearly define the host status adaptation requirements, water pump set matching standards and subsystem linkage adaptation conditions respectively; Based on the preprocessed data, the host status, pump set matching degree and subsystem linkage status are checked sequentially according to the rule base, and the adaptability results of each link are marked. Access the historical operating condition database to retrieve historical records of similar system states and instruction types, and analyze the risk level of instruction execution. Combining real-time verification results with historical analysis conclusions, the feasibility of instructions is determined according to preset logic, and the redundancy capability of the accounting system is assisted in determining the redundancy capability in medium-risk scenarios. If feasible, a verification pass signal is sent; if not feasible, optimization suggestions including adjustment direction, parameters and execution order are generated, displayed on the terminal and provided with a one-click optimization option. The entire verification process data is labeled with working condition tags and stored in a historical database. Historical data is analyzed regularly to optimize verification rule thresholds.

2. The hierarchical control method for air conditioning cold source systems according to claim 1, characterized in that, The configuration interface module adopts a dual-bus redundancy design, ensuring fast command transmission through transmission mechanisms and bandwidth allocation strategies, including: Select a PLC / DDC built-in bus communication card compatible with the system bus protocol and an external Ethernet communication module for the terminal to complete the hardware installation and connection, avoid strong electromagnetic interference, and use a specific wiring method to reduce signal attenuation; Deploy two independent primary and backup industrial Ethernet buses with two cabling paths, configure independent network devices and maintain consistent topology, and connect relevant control and verification modules via dual links; A management program is embedded in the hierarchical control module to preset the priority of the primary and backup buses, monitor the status of the primary bus in real time, and quickly switch to the backup bus when the fault threshold is met. When the data synchronization function is enabled, the data is synchronously stored in the backup bus buffer when the main bus transmits data. After the link switch, the backup bus data is read. After the main bus recovers, the main bus switches back and the backup bus buffer is cleared. The interrupt triggering function is enabled on the relevant communication interface. After the terminal encapsulates the instruction, it initiates an interrupt request, prioritizes the allocation of transmission resources, and the control module switches to the interrupt receiving mode to read the instruction. Configure a bandwidth priority queue on the switch, mark command data as the highest priority and reserve a fixed bandwidth, and ensure the command transmission bandwidth through a dynamic adjustment algorithm; A monitoring module is added to the management program to collect bus status data in real time, verify the integrity of transmitted data, and trigger retransmission or push early warning information when an anomaly occurs.

3. The hierarchical control method for air conditioning cold source systems according to claim 1, characterized in that, The module is configured to adapt to the communication interface, allocate an independent buffer, receive data frames, verify their validity, parse parameters, and mark the instruction status after verifying validity, including: An interface chip compatible with the industrial bus transmission protocol is selected and integrated into the hierarchical control module. The interface communication parameters are configured to be consistent with the bus. The adapter driver is loaded, the registers are initialized, and the interrupt triggering conditions are set to ensure that the interface is compatible with the protocol. A static memory allocation is used to create an independent instruction receive buffer. A FIFO storage mechanism is adopted, the initial position of the read and write pointers is clearly defined, and a mutex lock is configured to restrict single-process access and avoid conflicts. The communication interface is enabled in interrupt reception mode. After detecting the data stream, an interrupt is triggered and the data is written to the buffer byte by byte, the pointer is updated, and the reception timeout threshold is set. If the data is not received within the timeout, the reception is determined to have failed, the buffer is cleared, and the pointer is reset. After receiving a data frame, first check the frame header and frame tail identifiers. If they are invalid, mark them as invalid frames and clean them up. If they are valid, extract the CRC check code and compare it. If they do not match, mark them as verification failure frames, log them and clean up the buffer. Data is extracted according to the preset data frame structure, converted into a format that can be processed internally by the module, and the byte order is adjusted to adapt to the module's CPU. The built-in system device information database is called to verify the validity of the parsed parameters. If any parameter fails, it is determined to be an illegal parameter and the specific item is recorded. If the parameter verification passes, the instruction is marked as pending verification and associated with a unique identifier. If the verification or check fails, the corresponding abnormal state is marked. Regardless of the state, the pointer is updated and the buffer is cleared to prepare for the next reception.

4. The hierarchical control method for air conditioning cold source systems according to claim 1, characterized in that, The three preset verification rule bases respectively clarify the host status adaptation requirements, pump group matching standards, and subsystem linkage adaptation conditions, including: The architecture is designed as a layered structure of foundation layer, rule layer, and application layer. The foundation layer defines basic information, the rule layer stores the verification logic, and the application layer associates instructions and rules. A relational database is used to store the rules in separate tables according to the rule type, and extended fields are reserved to support iteration and differentiated configuration. The verification dimensions are sorted out, thresholds and fault warning conditions are defined in combination with relevant parameters, risk scenario judgment rules and exception scenarios are set, and the information is entered into the rule base to support differentiated configuration according to host model. Determine the verification parameters, set the corresponding thresholds and grading judgment logic, configure differentiated parameters for different types of water pump groups, and enter the relevant information into the rule base and mark the source of the parameters; Analyze the mapping relationship between instructions and linked devices, clarify the linkage conditions and linkage priorities of associated devices, set judgment rules, and enter the rule base according to subsystems and support differentiated management; Construct a mapping table to match different commands with three types of verification rules to avoid invalid verification; A standardized API call interface was developed. The rule base is stored using local caching and a remote database. When the module starts, the integrity of the rule base is loaded and verified. The accuracy is verified and parameters are adjusted through simulation.

5. The hierarchical control method for air conditioning cold source systems according to claim 1, characterized in that, If the verification passes, the instruction is broken down into levels and the execution rules are clearly defined; if it fails, optimization suggestions are generated and re-verification is supported after parameter adjustment. Feedback data is collected in real time during execution. In case of anomalies, the AI ​​module intervenes and issues an alarm. After execution, the system status is checked, a closed-loop confirmation message is sent, and all process data is stored, including: After the AI ​​verification module outputs the results, the hierarchical control module automatically distributes the data. If the verification passes, it initiates hierarchical decomposition of the instructions, clarifies the content and execution rules of each level, assigns a unique identifier to the decomposed instructions and associates them with the original instructions. If the verification fails, the AI ​​module generates targeted optimization suggestions, provides a one-click optimization execution option, adjusts the parameters and re-verifies until it passes. The hierarchical control module sends out the disassembled instructions at each level through the industrial bus. After receiving the instructions, the execution unit performs the actions and reports the status. At the same time, it collects the feedback data and synchronizes it to the hierarchical control module and the AI ​​module to form a data link. The AI ​​module monitors feedback data and execution status in real time, presets three types of anomaly judgment criteria, suspends the process and takes targeted intervention measures after an anomaly is triggered, generates alarm information to push to the operator and supports manual intervention; After each execution unit completes its feedback, the hierarchical control module checks the system status. If it passes, the execution is deemed complete, and a closed-loop confirmation message is sent. The entire process data is then summarized, labeled with operating condition tags, and stored in the historical database.

6. The hierarchical control method for air conditioning cold source systems according to claim 5, characterized in that, After the AI ​​verification module outputs its results, the hierarchical control module automatically distributes the data. If the verification passes, it initiates hierarchical instruction decomposition, clearly defining the content and execution rules of each level, assigning a unique identifier to each decomposed instruction and associating it with the original instruction. If the verification fails, the AI ​​module generates targeted optimization suggestions, providing a one-click optimization execution option. After adjusting the parameters, the verification is repeated until it passes, including: The hierarchical control module receives the results from the AI ​​verification module through a standardized interface, and its built-in diversion logic completes the status determination and automatically triggers the corresponding processing flow. The original instructions are broken down into three levels: system level, subsystem level, and device level. The content of each level is clarified and redundant items are filtered out. Execution rule templates are preset in combination with system characteristics. The execution order, interval and priority are set according to the instruction type. A unique identifier is assigned to the decomposed instructions and an association mapping with the original instruction parameters is established to support traceability. The AI ​​module combines historical databases and rule bases to identify problems in failed dimensions and generate structured optimization suggestions including adjustment directions, parameter ranges, and priorities. The terminal displays the suggestions and provides a one-click optimization function. After automatically replacing the command parameters, it triggers a re-verification and repeats the process until it passes. If it fails multiple times, it prompts for manual intervention. The hierarchical control module records the status and data of the entire process of traffic diversion in real time, binds the instruction identifier and stores it in a temporary cache, and archives it to the historical database after the instruction is executed.

7. The hierarchical control method for air conditioning cold source systems according to claim 5, characterized in that, The AI ​​module monitors feedback data and execution status in real time, presets three types of anomaly judgment criteria, and pauses the process and takes targeted intervention measures after an anomaly is triggered. It generates alarm information to push to the operator and supports manual intervention, including: The AI ​​module establishes a two-way communication link with the hierarchical control module and each execution unit through the industrial bus, receives feedback data and execution status in real time, synchronizes data at a preset frequency and temporarily stores it through the built-in data buffer queue to ensure real-time monitoring and data integrity. Preset standards for device execution anomalies, parameter out-of-bounds anomalies, and system fluctuation anomalies; all judgment thresholds can be adjusted as needed. The AI ​​module parses the buffer queue data in real time and compares it with the preset anomaly criteria. When any anomaly condition is met, the anomaly response mechanism is immediately triggered, and a process pause command is sent to the hierarchical control module. After receiving the command, the hierarchical control module pauses the issuance and execution of the current command and freezes the subsequent process. The AI ​​module uses an anomaly type matching intervention strategy library to take corresponding intervention measures such as device switching / command retry / parameter correction, parameter callback / associated device pause, and associated parameter fine-tuning / dynamic balancing. While the intervention is being implemented, the AI ​​module generates standardized alarm information that includes the type of anomaly, location, cause, intervention measures already taken, and suggested actions. This information is pushed through the system operation terminal via pop-up windows and sound prompts, and alarm logs are simultaneously written to storage and the historical database. The system operation terminal is equipped with interactive buttons for emergency stop, parameter adjustment, and resumption of execution. It supports operators to terminate the process, modify associated parameters, and resume execution. All manual operations are recorded, including the operator, time, and content, and are bound to alarm information for archiving. The AI ​​module continuously monitors the feedback data and execution status after intervention or manual operation. When the data returns to the normal range, it sends a process recovery instruction to the hierarchical control module to resume execution. If the abnormality is not resolved, it maintains the paused state and pushes alarms repeatedly on a regular basis. The time of occurrence, type, triggering conditions, intervention measures, manual operation records, resolution time, and processing results of the anomaly are bound to the corresponding instruction's unique identifier and stored in the historical database.

8. A hierarchical control system suitable for air conditioning cold source systems, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the hierarchical control method for an air conditioning cold source system according to any one of claims 1 to 7 by executing the machine-executable instructions.

Images