Air conditioner flow divider intelligent distribution control method based on flow data dynamic adjustment

Abstract

The application provides an air conditioner shunt intelligent shunt control method based on dynamic adjustment of flow data, relates to the technical field of local area network channel configuration and heat management, and comprises the following steps: monitoring network flow data and environmental temperature of a sub-region in real time; calculating a dynamic heat load coefficient based on the flow data; generating a network flow balance index in combination with the temperature; constructing an air duct opening matrix and dynamically optimizing the same through gradient descent; driving an actuating mechanism to adjust the air duct; and activating software-defined network rerouting to migrate overload flow when the index is lower than a threshold value. The application realizes dynamic air supply optimization by coupling network flow and air conditioner parameters, and establishes network rerouting and refrigeration linkage control, thereby solving the local overheating problem caused by the response lag of a static strategy and the fragmentation of cross-systems, and improving the energy efficiency and equipment safety of a data center.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of local area network channel configuration and thermal management, in particular to an intelligent shunt control method of an air conditioner shunt based on dynamic adjustment of traffic data. BACKGROUND

[0002] With the continuous expansion of data centers, the heat generated by server clusters has increased dramatically, and the energy consumption of the refrigeration system has accounted for more than 40% of the total energy consumption. The traditional air conditioner control method is mainly based on static temperature threshold or simple time sequence strategy, which cannot perceive the real-time changes of network traffic, resulting in low refrigeration efficiency. Especially in the traffic burst scenario, local areas form hot spots due to the heat generated by dense network equipment, and the air conditioning system responds with a lag, causing the risk of overheating and shutdown of equipment. The existing technology attempts to alleviate the problem by increasing the density of temperature sensors or increasing the fan speed, but it does not solve the problem of dynamic matching of network traffic and refrigeration demand, but instead exacerbates energy waste.

[0003] At the same time, the fragmentation of network traffic and thermal environment management is prominent. Network operation and maintenance systems can monitor traffic fluctuations, but cannot link air conditioning equipment; the refrigeration system lacks the ability to perceive network characteristics such as TCP throughput and UDP packet loss rate. This fragmentation leads to two serious consequences: first, the air conditioner shunt air duct opening is fixed and cannot dynamically adjust the air supply ratio according to business load; second, when local overheating occurs, only the overall refrigeration intensity can be passively increased, and the temperature cannot be reduced through traffic migration. Existing solutions such as SDN-based traffic scheduling can migrate loads, but they are not deeply coupled with temperature control algorithms, and the migration strategy is disconnected from the thermodynamic state, which can easily cause secondary hot spots.

[0004] In addition, the existing air duct control model has significant defects. Most systems use open-loop control or simple feedback mechanisms and do not establish a quantitative relationship between network state and refrigeration efficiency. The adjustment of air duct opening depends on human experience and lacks the ability to optimize multiple variables collaboratively. In complex traffic patterns (such as burst traffic caused by BGP routing oscillation), traditional methods are difficult to quickly converge to the optimal solution, which cannot guarantee the safety of equipment cooling and causes a large amount of cold waste. SUMMARY

[0005] In order to solve the technical problems of "static air conditioner control in the prior art cannot respond to dynamic changes in network traffic, leading to local overheating and a sharp increase in energy consumption" and "fragmentation of network traffic scheduling and refrigeration system, resulting in low thermal management efficiency", the present application provides an intelligent shunt control method of an air conditioner shunt based on dynamic adjustment of traffic data.

[0006] The technical solutions provided by the present application are as follows:

[0007] The intelligent shunt control method of an air conditioner shunt based on dynamic adjustment of traffic data provided by the present application comprises:

[0008] S1, real-time monitoring of network traffic data and environmental temperature data of each sub-area in the local area network, wherein the network traffic data includes transmission control protocol throughput, user datagram protocol packet loss rate and border gateway protocol route fluctuation frequency;

[0009] S2, calculating a dynamic thermal load coefficient based on the network traffic data, the coefficient representing the influence of network traffic changes on the dynamic thermal load of the refrigeration system;

[0010] S3, generating a network flow balance index according to the environmental temperature data and the dynamic thermal load coefficient, the index comprehensively reflecting the matching degree of network state and thermodynamic environment;

[0011] S4, constructing a multi-dimensional air duct opening matrix of the air conditioner flow divider, the matrix elements representing the opening proportion of each air duct to the sub-area;

[0012] S5, taking the network flow balance index as the optimization target, and dynamically adjusting all element values of the air duct opening matrix by using the gradient descent algorithm;

[0013] S6, generating a control instruction according to the optimized air duct opening matrix to drive the physical actuator of the air conditioner flow divider to complete the air duct opening adjustment;

[0014] S7, when the network flow balance index is lower than the alarm threshold, activating a network traffic rerouting mechanism to migrate the traffic data of the overloaded sub-area through a software-defined network controller.

[0015] Preferably, the S2 specifically comprises:

[0016] S201, collecting network traffic peak value and traffic arithmetic mean value within a preset time window;

[0017] S202, calculating a theoretical traffic entropy value based on the Shannon channel capacity formula:

[0018]

[0019] wherein, is the theoretical traffic entropy value, is the physical channel bandwidth, unit: Hz, is the network traffic peak value, unit: bps, is the network noise power, unit: W;

[0020] S203, calculating a traffic change gradient module value:

[0021]

[0022] wherein is the traffic change gradient module value, is the sampling time window length, unit: s, is the start time of the time window, is the network traffic, is the time variable;

[0023] S204, generating a dynamic thermal load coefficient:

[0024]

[0025] wherein, is the arithmetic mean of the traffic, in bps, is the dimensionless dynamic thermal load coefficient.

[0026] Preferably, the calculation of the traffic change gradient modulus in S203 further comprises:

[0027] S2031, constructing a cubic spline interpolation function for the discrete traffic data within the time window ;

[0028] S2032, solving the time domain integral of the absolute value of the second derivative by numerical integration:

[0029]

[0030] wherein, is the end time of the time window, is the start time of the time window, is the sampling point index, is the cubic spline interpolation function, is the sampling time point, is the total number of sampling points, and the interpolation function satisfies the natural boundary condition .

[0031] Preferably, the calculation of the network flow balance index in S3 specifically comprises:

[0032] S301, calculating the temperature variance between sub-regions:

[0033]

[0034] wherein, is the temperature variance between sub-regions, is the sub-region index, is the number of sub-regions, is the temperature of the sub-region, in ℃, is the average temperature, in ℃;

[0035] S302, generating a network flow balance index:

[0036]

[0037] wherein, is the network flow balance index, is the inter-subregion temperature variance, is the dimensionless dynamic thermal load coefficient, is the preset optimal temperature, is the temperature vector, is the preset optimal temperature, in ℃, is the maximum temperature difference allowed by the system, in ℃, is the all-ones vector, is the Euclidean norm, i.e., L2 norm, is the inter-subregion temperature variance, is the dynamic thermal load coefficient, is the dimensionless network flow balance index.

[0038] Preferably, the temperature variance calculation of S301 employs a dynamic sliding window mechanism, specifically comprising:

[0039] The window size is determined by the following formula:

[0040]

[0041] wherein, is the dimensionless dynamic thermal load coefficient, is the basic sampling period, in s; is the lower threshold of the thermal load coefficient; is the adaptive window size, in sampling points.

[0042] Preferably, the initialization of the air duct opening matrix in S4 satisfies:

[0043]

[0044] wherein, is the air duct index, is the subregion index, is the air duct to the subregion , the initial opening value, is the air duct to the subregion , the opening ratio, is the physical area of the th subregion, in , is the physical area of the th subregion, in ; is the total number of air ducts, is the total number of subregions.

[0045] Preferably, the gradient descent algorithm of S5 adopts a proportional-integral-derivative control mechanism, specifically including:

[0046] Bias input quantity is defined as:

[0047]

[0048] wherein, is the bias input quantity at time t, is the current time, is the network flow balance index varying with time t, is a preset target threshold, and is an integral time constant Self-adaptive adjustment with ambient temperature:

[0049]

[0050] wherein, is an integral time constant, and is a calibration coefficient, is the current ambient temperature, is a preset optimal temperature.

[0051] Preferably, the network flow re-routing mechanism of S7 specifically includes:

[0052] S701, identifying a sub-region satisfying as a migration source;

[0053] S702, selecting a sub-region satisfying as a migration target;

[0054] S703, modifying the switch forwarding table item through the OpenFlow protocol, and migrating the flow as:

[0055]

[0056] wherein is the real-time flow load of the migration source sub-region, in bps, is the real-time flow load of the migration target sub-region, in bps, is a preset flow balance threshold, in bps, is an actual migration flow value, in bps, and the above formula ensures that the flow of the source region after migration is not lower than the threshold and the flow of the target region does not exceed the threshold.

[0057] Preferably, the collection of the ambient temperature data in S1 specifically includes:

[0058] ​​S101, deploying at least three temperature sensors in each sub-area to form a redundant array;

[0059] S102, using weighted median filtering to process the original temperature data:

[0060]

[0061] wherein, is the filtered sub-area temperature, is the measurement value of the sensor is the sensor accuracy weight, is the to-be-solved filtered temperature variable, is the sensor index;

[0062] S103, performing time synchronization compensation on the filtered data, and the compensation amount , is the synchronization coefficient, is the first-order derivative of the flow with respect to time, is the small change amount of the network flow, is the small change amount of time.

[0063] Preferably, the cubic spline interpolation function further comprises:

[0064] The cubic spline interpolation function satisfies the continuity condition at the node :

[0065]

[0066] wherein, denotes the derivative of the function, the constraint ensures the smoothness of the flow change gradient calculation, is the left limit time of the th sampling time point, is the right limit time of the th sampling time point, is the derivative order.

[0067] The technical solution provided by the present application has at least the following beneficial effects:

[0068] (1) In the present application, through the coupling model of the dynamic thermal load coefficient and the network flow balance index, the network flow characteristics and the air conditioning refrigeration parameters are first realized in real time linkage, so that the air duct opening degree can be dynamically optimized according to the network state such as transmission control protocol throughput, user datagram protocol packet loss rate, etc., so as to fundamentally eliminate local hot spots and significantly reduce refrigeration energy consumption;

[0069] ​(2) In the present application, the gradient descent algorithm is used to optimize the multi-dimensional air duct opening matrix, and the adaptive mechanism of proportional-integral-derivative control is combined to effectively solve the response delay problem of the traditional control strategy, quickly complete accurate air supply adjustment under sudden scenes such as border gateway protocol routing fluctuations, and greatly reduce the risk of equipment overheating;

[0070] (3) In the present application, the software-defined network rerouting triggered by the network flow balance index is linked with the air conditioning control cross-system linkage to build a network refrigeration integrated closed-loop control: when the air duct regulation reaches the limit, automatically migrate the overload area flow to the low-temperature area, break through the single refrigeration bottleneck, and overall improve the energy efficiency level of the data center. BRIEF DESCRIPTION OF DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0072] Figure 1 The flowchart of the intelligent shunting control method of the air conditioner shunt based on dynamic adjustment of flow data provided by the embodiment of the present application is shown.

[0073] Figure 2 The dynamic thermal load coefficient calculation flowchart of the intelligent shunting control method of the air conditioner shunt based on dynamic adjustment of flow data provided by the embodiment of the present application is shown.

[0074] Figure 3 The three-level degradation strategy execution flowchart of the intelligent shunting control method of the air conditioner shunt based on dynamic adjustment of flow data provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0075] The technical solutions in the present application will be described below with reference to the drawings.

[0076] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be either one of the two.

[0077] In the embodiments of the present application, "image" and "picture" can be used interchangeably, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0078] In the embodiments of the present application, sometimes the subscript is It may be a non-subscript form such as W1 by mistake, and the meanings expressed are consistent when the distinction is not emphasized.

[0079] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0080] Reference is made to the accompanying drawings and Figure 1 , which shows a flowchart of the intelligent shunting control method of the air conditioner shunt provided by the embodiments of the present application based on dynamic adjustment of traffic data.

[0081] The embodiments of the present application provide an intelligent shunting control method of an air conditioner shunt based on dynamic adjustment of traffic data, and the processing flow can include the following steps:

[0082] S1, real-time monitoring of network traffic data and environmental temperature data of each sub-area in the local area network, wherein the network traffic data includes transmission control protocol throughput, user datagram protocol packet loss rate and border gateway protocol routing fluctuation frequency.

[0083] Specifically, the temperature sensor adopts a PT1000 platinum resistance temperature sensor with an accuracy of ±0.1℃, and transmits data through a Modbus RTU protocol at a sampling frequency of 1Hz. The three sensors are arranged in an equilateral triangle in the sub-area, and are located at the front door, the middle and the rear door of the cabinet, respectively, at a distance of 1.2 meters from the ground. The weights of the weighted median filter are set according to the sensor calibration certificate: the weight of the sensor with an accuracy of ±0.1℃ is 0.4, and the weight of the sensor with an accuracy of ±0.2℃ is 0.2. The time synchronization compensation coefficient =0.1, and the compensation amount The flow rate change rate dF / dt is obtained by forward difference method, for example, when the flow rate change rate is 10Gbps / s, the time compensation value is 1ms.

[0084] In one possible implementation, the collection of the environmental temperature data in S1 specifically includes:

[0085] S101, deploying at least three temperature sensors in each sub-area to form a redundant array;

[0086] S102, the original temperature data is processed by weighted median filtering:

[0087]

[0088] wherein, is the filtered sub-region temperature, is the measurement value of the sensor , is the sensor precision weight, is the to-be-solved filtered temperature variable, is the sensor index;

[0089] S103, time synchronization compensation is performed on the filtered data, and the compensation amount , is the synchronization coefficient, is the first derivative of the flow with respect to time, is the small change amount of network flow, is the small change amount of time.

[0090] It should be noted that the high-precision temperature data obtained by the above environmental temperature data acquisition method will be combined with real-time network flow data to form the basis input of the system. These synchronously collected data will be directly used in the calculation process of the dynamic thermal load coefficient, which is a key indicator for measuring the influence of network flow on the thermal load of the refrigeration system, and provides a quantitative basis for subsequent intelligent control decisions.

[0091] S2, calculate the dynamic thermal load coefficient based on the network flow data, which represents the influence of network flow changes on the dynamic thermal load of the refrigeration system.

[0092] In one possible implementation, as shown in Figure 2 S2 specifically includes:

[0093] S201, collect the network flow peak value and flow arithmetic mean value in a preset time window;

[0094] S202, calculate the theoretical flow entropy value based on the Shannon channel capacity formula:

[0095]

[0096] wherein, is the theoretical flow entropy value, is the physical channel bandwidth, in Hz, is the network flow peak value, in bps, is the network noise power, in W;

[0097] S203, calculate the flow change gradient module value:

[0098]

[0099] wherein, is the flow change gradient modulus, is the sampling time window length, in s, is the time window starting time, is the network flow, is the time variable;

[0100] S204, generating a dynamic thermal load coefficient:

[0101]

[0102] wherein, is the flow arithmetic mean, in bps, is the dimensionless dynamic thermal load coefficient.

[0103] It should be noted that the time window is fixed at 60 seconds. The flow peak uses the 99.9 percentile statistical method to effectively exclude abnormal value interference such as Distributed Denial of Service (DDoS). The network noise power is measured during the business low period, and the typical value of the 10G Ethernet port is -90dBm. The theoretical flow entropy value When calculating, the channel bandwidth B takes the physical negotiation rate of the switch port (such as 10Gbps / 25Gbps / 100Gbps). The flow second derivative uses the central difference method for approximation: wherein is the sampling interval.

[0104] In one possible implementation, the calculation of the flow change gradient modulus in S203 further includes:

[0105] S2031, constructing a cubic spline interpolation function for the discrete flow data in the time window ;

[0106] S2032, solving the time domain integral of the second derivative absolute value by numerical integration:

[0107]

[0108] wherein, is the time window ending time, is the time window starting time, is the sampling point index, is the cubic spline interpolation function, is the sampling time point, for the total number of sampling points, the interpolation function satisfies the natural boundary condition .

[0109] In a possible implementation, the cubic spline interpolation adopts the natural boundary condition, and the spline coefficients are solved using the three-moment method. The node number N = 60 corresponds to 60 sampling points per minute. The numerical integration adopts the composite Simpson formula: for each sub-interval , the integral value , where is the length of the sub-interval. In the FPGA hardware implementation, the Q2.30 fixed-point number format is used to ensure that the cumulative calculation error is <0.1%.

[0110] In a possible implementation, the cubic spline interpolation function further comprises:

[0111] The cubic spline interpolation function satisfies the continuity condition at the node :

[0112]

[0113] wherein denotes the order derivative of the function, and the constraint ensures the smoothness of the gradient of the flow change calculation, is the left limit time of the th sampling time point, is the right limit time of the th sampling time point, is the derivative order.

[0114] Specifically, the node continuity is ensured by the following mathematical constraints:

[0115] Function value continuity

[0116] First-order derivative continuity

[0117] Second-order derivative continuity

[0118] When solving the three-moment equation, a 52x52 band matrix is formed (N = 60), and the LU decomposition algorithm is used, and the iteration tolerance is set to . The natural boundary condition is realized by forced constraints .

[0119] S3, generating a network flow balance index according to the environmental temperature data and the dynamic thermal load coefficient, which comprehensively reflects the matching degree of the network state and the thermodynamic environment.

[0120] In one possible implementation, the calculation of the network flow balancing index in S3 specifically includes:

[0121] S301. Calculate the temperature variance between sub-regions:

[0122]

[0123] in, The temperature variance between sub-regions. For sub-region indexing, The number of sub-regions. For the first Sub-region temperature, in °C The average temperature is expressed in °C.

[0124] S302, Generation Network Flow Equilibrium Index:

[0125]

[0126] in, This is the network flow balance index. The temperature variance between sub-regions. This is a dimensionless dynamic thermal load coefficient. To preset the optimal temperature, For temperature vectors, The preset optimal temperature is expressed in °C. The maximum allowable temperature difference for the system, expressed in °C. It is a vector of all 1s. This is the Euclidean norm, also known as the L2 norm. The temperature variance between sub-regions. For dynamic thermal load coefficient, It is a dimensionless network flow equilibrium index.

[0127] Understandably, data validity verification is necessary before calculating temperature variance: when A value at ℃ is considered a sensor fault, and the average value of adjacent sensors is used as a replacement. L2 norm The calculation is performed using the `cblas_dnrm2` function from the OpenBLAS library. (Exponential term, denominator) Set a lower limit of 0.01 to prevent division by zero errors. Preset the optimal temperature during actual deployment. =24℃±0.5℃, the maximum allowable temperature difference of the system =5℃. The calculation results were normalized to the [0,1] interval using the sigmoid function.

[0128] In one possible implementation, the temperature variance calculation of S301 employs a dynamic sliding window mechanism, specifically including:

[0129] The window size is determined by the following formula:

[0130]

[0131] wherein, is a dimensionless dynamic thermal load coefficient, is a basic sampling period, in seconds; is a lower threshold of the thermal load coefficient; is an adaptive window size, in sampling points.

[0132] Specifically, the basic sampling period = 5 seconds, and the lower threshold of the thermal load coefficient = 0.1. The window size W is rounded up after calculation, and the maximum value is limited to 300 (corresponding to 25 minutes of data). A ring buffer is used to store temperature data, and the buffer depth is 300. The variance calculation is triggered once every W sampling periods, and when the window update is suspended until the threshold is above.

[0133] It can be understood that the network flow balance index As a core parameter reflecting the matching degree of network state and thermodynamic environment, it provides a decision basis for precise regulation and control of the air conditioning system. Based on the regulation and control demand generated by the index, the system will start the construction process of the air conditioning flow divider multi-dimensional air duct opening matrix, and through mathematical modeling, it will convert the regulation and control demand into executable control parameters, laying a foundation for the subsequent gradient optimization algorithm.

[0134] S4, a multi-dimensional air duct opening matrix of an air conditioning flow divider is constructed, and the matrix elements represent the opening proportion of each air duct to a sub-region.

[0135] In one possible implementation, the initialization of the air duct opening matrix in S4 satisfies:

[0136]

[0137] wherein, is an air duct index, is a sub-region index, is the initial opening value of the air duct to the sub-region , is the opening proportion of the air duct to the sub-region , is the physical area of the first sub-region, in , is the physical area of the first sub-region, in ; Total number of sub-regions.

[0138] In a possible implementation, the total number of air ducts =8 corresponds to the number of air conditioner terminals, the number of sub-regions =48 corresponds to the number of standard cabinets. The physical area is calculated according to the cabinet floor area (width 0.6m x depth 1.2m = 0.72 ). The initial opening value is written to the holding register of the PLC, and the normalization constraint is calculated in real time and the deviation is checked. When the deviation >1%, the audible light alarm is triggered and the automatic re-normalization is triggered.

[0139] Specifically, the re-normalization operation is realized by the following steps:

[0140] Calculate the current opening of each air duct and: ;

[0141] Perform scaling on each matrix element: ;

[0142] Write the normalized matrix to the PLC holding register through the Modbus TCP protocol;

[0143] Check the error after scaling: if , trigger the secondary alarm;

[0144] The above process is completed within 200ms, ensuring the real-time requirements of the control system.

[0145] Specifically, after the initialization and verification of the air duct opening matrix are completed, the system will enter the dynamic optimization phase. This phase takes the network flow balance index as the core optimization target, and adjusts the air duct opening parameters in real time through intelligent algorithms, so that the cooling capacity of the air conditioning system and the heat generation characteristics of the network equipment are in the best matching state, and the precise regulation and control of the data center thermal environment is realized.

[0146] S5, taking the network flow balance index as the optimization target, using the gradient descent algorithm to dynamically adjust the value of all elements of the air duct opening matrix.

[0147] In a possible implementation, the gradient descent algorithm of S5 adopts a proportional-integral-derivative control mechanism, specifically including:

[0148] The deviation input quantity is defined as:

[0149]

[0150] Wherein, is the The deviation of the time instant input quantity, is the current time, is the network flow balance index changing with time t, is the preset target threshold, and the integral time constant Self-adaptive adjustment with ambient temperature:

[0151]

[0152] wherein, is the integral time constant, and is the calibration coefficient, is the current ambient temperature, is the preset optimal temperature.

[0153] It should be noted that the target threshold is set to 0.85. The calibration coefficient , . The integral time constant is dynamically adjusted: when the temperature difference is > 3℃, = 1s (fast response mode), and when the temperature difference is < 1℃, = 3s (anti-oscillation mode). The gradient descent step size , and the control period is synchronized with the data acquisition period to be 5 seconds.

[0154] It should be noted that the specific implementation process of the gradient descent algorithm is as follows:

[0155] Initialize the Jacobian matrix to be a zero matrix;

[0156] Perform perturbation analysis on each matrix element :

[0157] a. Positive perturbation:

[0158] b. Negative perturbation:

[0159] c. Calculate the partial derivative:

[0160] Update the opening matrix:

[0161] Constraint processing: if , set it to 0, and if , set it to 1.

[0162] wherein the learning rate , and the perturbation analysis is performed every 5 iterations to reduce the computational load.

[0163] S6, generating control instructions according to the optimized air duct opening matrix, and driving a physical actuator of an air conditioner flow divider to complete air duct opening adjustment.

[0164] In a possible implementation, the optimized air duct opening matrix is issued through an OPC UA protocol. The actuator is an electric air valve controlled by a 0-10V analog signal, and the voltage-opening degree is linearly corresponding (10V=100% opening degree), the opening degree resolution is 0.1%, and the mechanical response time is less than 2 seconds. A full calibration process is performed every 15 minutes: first, reset to the initial opening degree , then adjust to the current optimal solution step by step at a step length of 0.5%, and finally record the position sensor feedback value to compensate for the mechanical backlash.

[0165] It should be noted that, while the air conditioning system is performing air duct adjustment, the system continues to monitor the change trend of the network flow balance index. When the index is continuously lower than the safety threshold, it indicates that single air conditioning regulation cannot meet the thermal management demand, at which time the collaborative regulation mechanism at the network level is activated, and the spatial transfer of the thermal load is realized through flow rerouting, forming a linkage control closed loop of the air conditioning system and the network system.

[0166] S7, when the network flow balance index is lower than the alarm threshold, activating a network flow rerouting mechanism, and migrating the flow data of an overloaded sub-region through a software-defined network controller.

[0167] In a possible implementation, the network flow rerouting mechanism of S7 specifically includes:

[0168] S701, identifying a sub-region satisfying as a migration source;

[0169] S702, selecting a sub-region satisfying as a migration target;

[0170] S703, modifying a switch forwarding table item through an OpenFlow protocol, and migrating the flow as follows:

[0171]

[0172] Wherein is the real-time flow load of the migration source sub-region, in bps, is the real-time flow load of the migration target sub-region, in bps, is a preset flow balance threshold, in bps, is an actual migration flow value, in bps, and the above formula ensures that the flow of the source region is not lower than the threshold and the flow of the target region is not higher than the threshold after migration.

[0173] Specifically, the alarm threshold =0.7, and the temperature deviation threshold =1℃. Migration source selection uses a minimum heap data structure, which prioritizes processing the lowest sub-region. Traffic balancing threshold Take 70% of the switch port rate (7Gbps for a 10G port). Migration is performed in three steps:

[0174] 1) Modify the forwarding table through the Flow-Mod message of OpenFlow;

[0175] 2) Batch migration (single ≤350Mbps);

[0176] 3) Post-migration verification Promote the value, and if it does not meet expectations, trigger VMware vMotion virtual machine migration.

[0177] It can be understood that the Flow-Mod message contains the following key fields:

[0178] Match field: source IP network segment ( / 24 subnet mask), destination port number;

[0179] Action field:

[0180] Output port (output_port) set to target area switch port;

[0181] Queue identifier (queue_id) corresponds to QoS priority;

[0182] Timeout field: hard_timeout=120s (temporary migration)

[0183] Batch migration uses a token bucket algorithm to control the rate:

[0184]

[0185] Where is the allowed migration rate at time , unit Mbps, is the time after the start of migration, unit seconds, to ensure initial low-speed migration (50Mbps), is the rate adjustment midpoint time (default set to 10 seconds), this algorithm ensures that the initial migration rate is about 50Mbps ( =0), and as time increases, it approaches the upper limit of 350Mbps, avoiding instantaneous congestion.

[0186] It should be noted that when the software defined network migration fails, three levels of degradation are started: Level 1 reduces the CPU frequency of the source area server to 80% of the base frequency; Level 2 migrates non-critical virtual machines through vMotion; Level 3 closes the backup node. The event log contains 12 fields such as timestamp, migration traffic, and temperature change rate, stored in Apache Parquet format, with a retention period of 30 days. Log analysis uses Spark Streaming to detect abnormal patterns in real time.

[0187] In one possible implementation, as shown in the policy execution logic, Figure 3 the trigger conditions for each level of degradation policy are:

[0188] Level 1 trigger condition: when and the number of SDN migration failures CPU frequency is set through IPMI raw 0x080x05 instruction:

[0189]

[0190] where is the adjusted CPU frequency, unit GHz, is the CPU base frequency (such as 2.4 GHz), is the CPU maximum turbo frequency (such as 3.8 GHz), is the current network flow balance index is the frequency reduction coefficient (default value 0.2).

[0191] Level 2 trigger condition: after Level 1 executes for 10 minutes still <0.65:

[0192] Virtual machine migration selection criteria:

[0193] if memory usage <40% && CPU utilization <50% && service type ≠ database

[0194] then migration priority = high

[0195] Migrate selected virtual machines through vMotion API, with a migration bandwidth limit of 1Gbps.

[0196] Level 3 trigger condition: after Level 2 executes still <0.6:

[0197] Shutdown sequence: backup node → test environment → non-core business container (Kubernetes Pod)

[0198] Close instruction: systemctl stop service_name or kubectl scale deployment --replicas=0

[0199] It can be understood that the above three-level degradation strategy constitutes a security system of the system. When the master control strategy fails, the system ensures the safety boundary of the system by gradually reducing the service load, and at the same time, gains valuable time for fault diagnosis and system recovery. The hierarchical and progressive fault-tolerant mechanism and the core control algorithm together constitute a complete technical solution, ensuring the reliability and robustness of the data center thermal environment management.

[0200] The technical scheme provided by the embodiment of the application has at least the following beneficial effects:

[0201] (1) In the application, through the coupling model of the dynamic thermal load coefficient and the network flow balance index, the network flow characteristics and the air conditioner refrigeration parameters are first realized in real time linkage, so that the air duct opening degree can be dynamically optimized according to the network state such as transmission control protocol throughput, user datagram protocol packet loss rate, etc., so as to fundamentally eliminate local hot spots and significantly reduce refrigeration energy consumption;

[0202] (2) In the application, the gradient descent algorithm is used to cooperatively optimize the multi-dimensional air duct opening matrix, and the adaptive mechanism of proportional-integral-derivative control is combined to effectively solve the response lag problem of the traditional control strategy, quickly complete accurate air supply adjustment under sudden scenes such as border gateway protocol routing fluctuations, and greatly reduce the risk of equipment overheating;

[0203] (3) In the application, the software-defined network rerouting triggered by the network flow balance index and the air conditioner control cross-system linkage build a network refrigeration integrated closed-loop control: when the air duct regulation reaches the limit, automatically migrate the overload area flow to the low-temperature area, break through the single refrigeration bottleneck, and overall improve the energy efficiency level of the data center.

[0204] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the application, which should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

[0205] In the case of no conflict, the embodiments of the application and the features in the embodiments can be combined with each other to obtain new embodiments.

[0206] The above is only a specific embodiment of the application, but the protection scope of the application is not limited thereto, and the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. An intelligent shunt control method for an air conditioner shunt based on dynamic adjustment of flow data, characterized in that, The method comprises the following steps: S1, real-time monitoring of network traffic data and environmental temperature data of each sub-area in the local area network, wherein the network traffic data comprises transmission control protocol throughput, user datagram protocol packet loss rate and border gateway protocol route fluctuation frequency; S2, calculating a dynamic thermal load coefficient based on the network traffic data, which represents the influence of network traffic changes on the dynamic thermal load of the refrigeration system; The S2 specifically comprises: S201, collecting network traffic peak value and traffic arithmetic mean value within a preset time window; S202, calculating theoretical traffic entropy value based on Shannon channel capacity formula: wherein, is a theoretical flow entropy value, is a physical channel bandwidth, in Hz, is a network flow peak value, in bps, is a network noise floor power, in W; S203, calculating traffic change gradient modulus value: wherein, is the flow change gradient module, is the sampling time window length, in s, is the time window starting time, is the network traffic, is the time variable; S204, generating dynamic thermal load coefficient: wherein, is the arithmetic mean of the flow, in bps, is the dimensionless dynamic thermal load factor; S3, generating a network flow balance index based on the environmental temperature data and the dynamic thermal load coefficient, which comprehensively reflects the matching degree of network state and thermodynamic environment; S4, constructing a multi-dimensional air duct opening matrix of the air conditioner flow divider, and the matrix elements represent the opening proportion of each air duct to the sub-area; S5, taking the network flow balance index as the optimization target, and dynamically adjusting all element values of the air duct opening matrix by using gradient descent algorithm; S6, generating control instructions based on the optimized air duct opening matrix to drive the physical actuator of the air conditioner flow divider to complete air duct opening adjustment; S7, when the network flow balance index is lower than the alarm threshold, activating a network traffic rerouting mechanism to migrate traffic data of overloaded sub-areas through a software-defined network controller.

2. The intelligent split control method of the air conditioner splitter based on dynamic adjustment of traffic data according to claim 1, wherein, The calculation of the traffic change gradient modulus value in the S203 further comprises: S2031、constructing a cubic spline interpolation function for the discrete flow data within the time window ; S2032, solving the time domain integral of the absolute value of the second derivative by numerical integration: wherein, is the time window end time, is the time window start time, is the sample point index, is a cubic spline interpolation function, is the th sample time point, is the total number of sample points, the interpolation function satisfies the natural boundary condition .

3. The intelligent split control method for air conditioner splitter based on dynamic adjustment of flow data according to claim 1, characterized in that, The calculation of the network flow balance index in the S3 specifically comprises: S301, calculating temperature variance between sub-areas: wherein, is the temperature variance between sub-regions, is the sub-region index, is the number of sub-regions, is the first sub-region temperature in °C, is the average temperature in °C; S302, generating network flow balance index: wherein, is the network flow balance index, is the inter-subregion temperature variance, is the dimensionless dynamic thermal load factor, is the preset optimal temperature, is the temperature vector, is the preset optimal temperature, in °C, is the maximum temperature difference allowed by the system, in °C, is the all-ones vector, is the Euclidean norm, i.e., the L2 norm, is the inter-subregion temperature variance, is the dynamic thermal load factor, is the dimensionless network flow balance index.

4. The intelligent split control method of claim 3, wherein, The temperature variance calculation in the S301 adopts a dynamic sliding window mechanism, specifically comprising: The window size is determined by the following formula: wherein, is a dimensionless dynamic thermal load coefficient, is a basic sampling period, in s; is a thermal load coefficient lower threshold value; is an adaptive window size, in number of samples.

5. The intelligent split control method for air conditioner splitter based on flow data dynamic adjustment according to claim 1, characterized in that, The initialization of the air duct opening matrix in the S4 satisfies: in, For air duct indexing, For sub-region indexing, For the air passage Pair of sub-regions The initial opening value, For the air passage Pair of sub-regions The opening ratio, For the first The physical area of ​​the sub-region, in units of , for the first The physical area of ​​the sub-region, in units of ; The total number of air ducts, This represents the total number of sub-regions.

6. The intelligent split control method for air conditioner splitter based on dynamic adjustment of flow data according to claim 3, characterized in that, The gradient descent algorithm in the S5 adopts a proportional-integral-derivative control mechanism, specifically comprising: Bias input quantity is defined as: wherein, is a time of day offset input, is the current time, is a network flow equalization index that varies with time t, is a preset target threshold, integration time constant self-adaptive adjustment with ambient temperature wherein, is an integration time constant, and is a calibration factor, is a current ambient temperature, is a preset optimal temperature.

7. The intelligent split control method of an air conditioner splitter based on dynamic adjustment of flow data according to claim 6, wherein, The network traffic rerouting mechanism in the S7 specifically comprises: S701、identify a sub-area satisfying as a migration source, an equalization index alarm threshold for the network flow; S702, select a sub-region satisfying as a migration target; S703, modifying switch forwarding table entries through OpenFlow protocol to migrate traffic formula: wherein is the real-time traffic load of the source sub-region for migration, in bps, is the real-time traffic load of the target sub-region for migration, in bps, is a preset traffic balancing threshold, in bps, is the actual migration traffic value, in bps, which is ensured by the migration traffic formula that the traffic of the source region after migration is not lower than the threshold and the traffic of the target region does not exceed the threshold.

8. The intelligent split control method for air conditioner splitter based on flow data dynamic adjustment according to claim 1, characterized in that, The collection of environmental temperature data in the S1 specifically comprises: S101, deploying at least three temperature sensors in each sub-area to form a redundant array; S102, processing original temperature data by using weighted median filtering: wherein, is the filtered sub-region temperature, is the measurement value of the sensor , is the sensor accuracy weight, is the filtered temperature variable to be solved, is the sensor index; S103, performing time synchronization compensation on the filtered data, the compensation amount , is a synchronization coefficient, is a first-order derivative of the flow with respect to time, is a small change amount of the network flow, is a small change amount of time.

9. The intelligent split control method for air conditioner splitter based on flow data dynamic adjustment according to claim 2, characterized in that, The cubic spline interpolation function further comprises: The cubic spline interpolation function satisfies the continuity condition at the nodes ​ wherein ; wherein, denotes the function's derivative of order, this constraint ensures the smoothness of the flow variation gradient calculation, is the left limit time instant of the sampling time point, is the right limit time instant of the sampling time point, is the derivative order.

Images