An artificial intelligence-based server production line scheduling method, device and medium
By constructing an aging resource domain graph model and generating binding fingerprint values, the problems of low resource utilization and broken binding relationships in the server aging/stress testing production line are solved, and stable scheduling and traceability of the server aging/stress testing segment are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGGUAN RUIZHI HARDWARE TECH CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-09
AI Technical Summary
In server aging/stress testing production lines, traditional scheduling methods struggle to balance power load safety, temperature control stability, batch orchestration flexibility, and data traceability, resulting in low resource utilization. Furthermore, the binding relationship between server serial numbers and rack positions is prone to breakage, making it difficult to achieve continuity and consistency in reorderable testing.
An aging resource domain graph model is constructed, binding fingerprint values and test instance identifier values are generated, and joint constraints are applied through aging basic data to generate candidate batch results. The model is then dynamically reconstructed within the rated power limit and temperature control range of the aging rack to maintain the continuity and traceability of the server-rack binding relationship.
It achieves stable scheduling of aging resources under high load scenarios, avoids repeated testing and aging time drift, ensures that power load and temperature control are within the controlled range, and maintains the continuity and traceability of the server and rack binding relationship.
Smart Images
Figure CN122173345A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and specifically to a server production line scheduling method, equipment, and medium based on artificial intelligence. Background Technology
[0002] In the server manufacturing process, server aging and stress testing are usually important links at the end of the production line. A large number of servers need to run for a long time in the aging area to expose potential defects. In this process, it is necessary to consider the resource boundaries formed by the upper limit of the rated power of the aging rack and the temperature control range of the aging rack, as well as the binding relationship between the server serial number, test recipe version number and rack identification value and the consistency of test duration. As the scale of the production line and the number of server models continue to increase, the traditional scheduling method that relies on manual experience or fixed rules is difficult to balance power load safety, temperature control stability, batch arrangement flexibility and data traceability in large-scale scenarios. Therefore, building a scheduling framework that can perceive the basic data of aging, power curve value and temperature requirement range value based on an artificial intelligence server production line scheduling method has become an important technical direction in server aging / stress testing production lines.
[0003] Currently, in the scheduling process of server aging / stress testing production lines, the upper limit of the rated power of the aging rack and the temperature control range of the aging rack are often only reserved with safety margins based on manual experience or the aging batches are divided according to a single threshold. There is a lack of ability to jointly constrain the batch power timing values and temperature requirement ranges based on aging baseline data and aging resource domain graph models. This easily leads to situations where aging resource utilization is significantly sacrificed to avoid power overruns and temperature range mismatches. Furthermore, when load and environmental fluctuations occur, it is difficult to identify safe combinations that do not exceed the upper limit of the rated power of the aging rack and meet the temperature control range throughout the entire aging cycle. Consequently, it is difficult to establish a quantifiable and verifiable unified scheduling basis for the server aging / stress testing segment regarding power budget, temperature control strategy, and batch division.
[0004] Secondly, in the process of batch division and rack loading management for server aging / stress testing, management is usually carried out only through server serial numbers or simple batch numbers. There is a lack of a unified identity link consisting of binding fingerprint values and test instance identification values between server serial numbers, test recipe version numbers, and rack identification values. When the candidate batch results and rack loading results need to be dynamically rearranged due to adjustments in the rated power limit of the aging rack or changes in the temperature control range of the aging rack, problems such as the breakage of the binding relationship between servers and racks, repeated aging of some servers, uncontrollable deviations in aging time, and inconsistencies between scheduling records and actual execution status are likely to occur. It is difficult to maintain continuous tracking and relocatable consistency of rearrangeable test instance identification values throughout the entire process of batch rearrangement, rack migration, and result archiving, which restricts the practical application of an AI-based server production line scheduling method for multi-constraint joint scheduling. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a server production line scheduling method, equipment, and medium based on artificial intelligence.
[0006] A server production line scheduling method based on artificial intelligence, the method comprising:
[0007] Acquire basic aging data and rack identification values, and generate an aging resource domain map model based on the basic aging data. The basic aging data includes the upper limit of the rated power of the aging rack and the temperature control range of the aging rack.
[0008] Obtain the server serial number, test recipe version number, power curve value, and temperature requirement range value; perform encoding calculations based on the server serial number, test recipe version number, and rack identifier value to generate a binding fingerprint value, and generate a test instance identifier value based on the binding fingerprint value;
[0009] Candidate batch results are generated based on the aging resource domain graph model and test instance identifier values. The batch power time series value is obtained by time alignment and superposition calculation based on the power curve value. The temperature control legal identifier value is obtained by interval intersection calculation based on the temperature requirement interval value and the aging rack temperature control interval value. Candidate batch results with batch power time series value not exceeding the aging rack rated power upper limit value and temperature control legal identifier value being true are retained.
[0010] Based on the candidate batch results and rack position identification values, rack position loading results are generated, and a binding continuity verification chain is generated based on the bound fingerprint values and rack position loading results. When an offset is detected in the upper limit of the rated power of the aging rack or the temperature control range of the aging rack, the test instance identification values that can be rearranged are selected based on the binding continuity verification chain, and the candidate batch results and rack position loading results are regenerated.
[0011] Furthermore, the steps for generating an aging resource domain map model based on the aging baseline data are as follows:
[0012] For each aging rack in the production line, the upper limit of the rated power of the aging rack, the temperature control range of the aging rack, and the rack position identification value corresponding to the aging rack are collected. The above data are collected according to the aging rack as the granularity to form a basic aging data set.
[0013] Based on the aging basic data set, each aging rack is mapped as a resource node in the aging resource domain graph model, and the corresponding aging rack rated power upper limit value, aging rack temperature control range value and rack position identification value are recorded in each resource node to generate an aging resource node set containing multiple resource nodes.
[0014] Based on the physical adjacency, power supply circuit association, and temperature control linkage of the aging rack in the actual production line, an edge connection is established between any two interconnected resource nodes in the aging resource node set in the aging resource domain graph model. The corresponding power coupling attributes and temperature control coupling attributes are recorded in the edge connection to form the aging resource domain graph model.
[0015] Furthermore, the steps for generating the bound fingerprint value and the test instance identifier value are as follows:
[0016] For each server to be aged, the server serial number, test recipe version number and rack position identifier value pre-assigned to the server are read from the production line execution system. The power curve value corresponding to the server under the test recipe version number is called from the power test system. The temperature requirement range value corresponding to the server under the test recipe version number is retrieved from the product specification library to form a test basic parameter quadruple that corresponds one-to-one with the server.
[0017] Based on the test basic parameter quadruple, the server serial number, test recipe version number and rack identifier value are concatenated at the character level according to the preset field concatenation order. The character-level concatenation result is then hashed and a checksum generation operation is performed on the binding encoding function to generate a unique binding fingerprint value corresponding to the combination relationship between the server and the rack.
[0018] Based on the bound fingerprint value and the power curve value and temperature requirement range value that correspond one-to-one with the bound fingerprint value, batch number information and timestamp information are superimposed according to the preset instance numbering rules to generate a unique test instance identifier value in the current aging batch, and a one-to-one mapping relationship is established between the test instance identifier value and the bound fingerprint value.
[0019] Furthermore, the operation logic of the binding encoding function is as follows:
[0020] Based on the server serial number and the test recipe version number, the server serial number is used as the high-order field and the test recipe version number is used as the low-order field for fixed-length encoding. The server serial number encoding and the test recipe version number encoding are then concatenated in a fixed order to form the server identification segment.
[0021] Based on the shelf location identifier value, the shelf location identifier value is formatted and encoded according to a preset length to form a shelf location identifier segment;
[0022] The server identifier segment and the rack identifier segment are concatenated in the order of server identifier segment first and rack identifier segment last. The concatenation result is subjected to multiple rounds of hash operation to extract a fixed length hash result as the main body of the binding fingerprint value. The check bit is calculated based on the main body of the binding fingerprint value and appended to the end of the main body of the binding fingerprint value to obtain the binding fingerprint value used to identify the combination relationship of server, test recipe version number and rack identifier value.
[0023] Furthermore, the steps for retaining candidate batch results where the batch power timing value does not exceed the upper limit of the rated power of the aging rack at any given time and the temperature control valid identification value is true are as follows:
[0024] Based on the rack position identifier value recorded by each resource node in the aging resource domain graph model, for each test instance identifier value, the binding fingerprint value corresponding to the test instance identifier value is called, and the target rack position identifier value is parsed from the binding fingerprint value. The test instance identifier value is assigned to the resource node in the aging resource domain graph model whose rack position identifier value is consistent with the target rack position identifier value. All test instance identifier values corresponding to the resource node are collected in each resource node to form an initial candidate batch set corresponding to each resource node.
[0025] For each candidate batch in the initial candidate batch set, the power curve value and temperature requirement range value corresponding to the test instance identifier value in the candidate batch are called, and the rated power upper limit value and temperature control range value of the aging rack recorded in the resource node where the candidate batch is located are combined to generate the batch power timing value based on the power curve value and the predetermined batch power timing calculation logic.
[0026] Based on the temperature requirement range, the aging rack temperature control range, and the preset temperature control valid identifier calculation logic, a temperature control valid identifier value is generated.
[0027] Furthermore, the batch power timing calculation logic is as follows:
[0028] For each candidate batch, a unified time step and a common time axis covering the aging cycle of all test instances in the candidate batch are determined. The power curve values corresponding to each test instance in the candidate batch are interpolated and time axis aligned according to the time step to generate multiple aligned power curves defined on the common time axis.
[0029] For each time step on the common time axis, perform a point-by-point superposition operation on the power dimension on all aligned power curves corresponding to that time step to obtain the batch superimposed power value corresponding to each time step. Then, combine the batch superimposed power values of each time step in chronological order to generate the batch power time series value used for constraint detection.
[0030] Furthermore, the calculation logic for the temperature control valid identifier is as follows:
[0031] For each candidate batch, the temperature requirement range value corresponding to each test instance in the candidate batch is called, and the interval intersection operation is performed with the temperature control range value of the aging rack corresponding to the aging rack where the candidate batch is located, to obtain the intersection range of the temperature requirement range value of each test instance and the temperature control range value of the aging rack.
[0032] Determine whether all the intersection intervals corresponding to the test instances within the candidate batch are non-empty intervals. If all intersection intervals are non-empty, set the temperature control valid identifier value corresponding to the candidate batch to true. If at least one intersection interval is empty, set the temperature control valid identifier value corresponding to the candidate batch to false. This is used to further filter the candidate batch results based on temperature control conditions, provided that the batch power timing value meets the power constraint.
[0033] Furthermore, the steps for regenerating the candidate batch results and rack loading results are as follows:
[0034] For each candidate batch in the candidate batch results, the bound fingerprint value corresponding to each test instance identifier value in the candidate batch is called, and the rack position identifier value corresponding one-to-one with the test instance identifier value is parsed from the bound fingerprint value. The test instance identifier values in the same candidate batch are grouped and sorted according to the rack position identifier value to generate rack position loading results with rack position identifier value as index and corresponding test instance identifier value sequence as content.
[0035] Based on the test instance identifier value, its corresponding binding fingerprint value, and the batch number information of the candidate batch to which the test instance identifier value belongs, recorded in the rack loading results, the occurrence relationship of the same test instance identifier value in different loading stages or different batches is linked into a binding continuity verification chain node sequence according to the time sequence and loading sequence. A binding continuity verification chain from the initial loading to the current loading state is constructed for each test instance identifier value to indicate the continuity and change record of the test instance identifier value in the rack loading process.
[0036] When the monitoring module detects an offset between the rated power limit or temperature control range value of any aging rack and the original record in the aging baseline data, it calls the rack position identifier value corresponding to that aging rack and, based on the rack position identifier value, retrieves all associated node sequences in the binding continuity verification chain, filters out the test instance identifier values that are still on the aging rack during the offset effective period, marks the above test instance identifier values as reorderable test instance identifier values, and removes the reorderable test instance identifier values from the original rack position loading results.
[0037] An artificial intelligence-based server production line scheduling device includes:
[0038] Memory, used to store computer software programs;
[0039] A processor is used to read and execute the computer software program, thereby implementing any of the aforementioned AI-based server production line scheduling methods.
[0040] A non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the artificial intelligence-based server production line scheduling method according to any one of claims 1 to 8.
[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0042] This invention introduces an aging resource domain graph model, an aging basic data set, and candidate batch results into the server aging test production line to jointly constrain the upper limit of the rated power of the aging rack and the temperature control range of the aging rack. This ensures that the batch power timing value is always limited by the upper limit of the rated power of the aging rack throughout the entire aging cycle, and that the temperature requirement range and the temperature control range of the aging rack are consistent through a temperature control validity identifier. This maintains the power load and temperature control status of each aging rack within a controlled range, provided that the power budget and temperature control range of the aging zone cannot be exceeded. This reduces test interruptions and manual intervention caused by power overruns or temperature range mismatches. Furthermore, during the batch division stage, the aging resource domain graph model performs dual screening of the power and temperature control of candidate batch results, ensuring that subsequent scheduling and rack loading are carried out on the premise of meeting the constraints. This reduces the reliance of the server aging / stress test section on redundant power reservations and lenient temperature control configurations, providing a stable resource constraint foundation for the application of an AI-based server production line scheduling method in high-load aging scenarios.
[0043] Furthermore, this invention generates binding fingerprint values using server serial numbers, test recipe version numbers, rack identifier values, power curve values, and temperature requirement range values, and further generates test instance identifier values. By combining rack loading results with binding continuity verification chains, reorderable test instance identifier values are filtered and reloaded. This allows for dynamic reconstruction of batch power timing values and valid temperature control identifier values at the candidate batch result level when the rated power upper limit or temperature control range value of the aging rack deviates, without repeating the completed test process or introducing disordered drift of aging duration. On this basis, the traceability and portability of the server-rack binding relationship are maintained during multiple rounds of batch reordering and migration. This enables an AI-based server production line scheduling method to balance safety boundary constraints and test continuity when facing multi-constraint joint scheduling requirements, and to complete the dynamic scheduling and stable archiving of server aging / stress test segments.
[0044] In summary, this invention constructs an aging resource domain graph model in the server aging / stress testing production line and introduces a combined identifier structure that binds fingerprint values and test instance identifier values. Under the premise that the upper limit of the rated power of the aging rack and the temperature control range of the aging rack cannot be exceeded and the binding relationship of the rack identifier values remains traceable and portable, the invention performs multi-constraint joint dynamic scheduling on the division of test batches and rack loading, thereby avoiding repeated testing and aging duration drift. Attached Figure Description
[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0046] Figure 1 This is a flowchart of an artificial intelligence-based server production line scheduling method provided in Embodiment 1 of the present invention. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0048] Example 1
[0049] Please see Figure 1As shown in the figure, this embodiment discloses a server production line scheduling method based on artificial intelligence, the method including:
[0050] S11: Obtain aging basic data and rack identification value, and generate an aging resource domain map model based on the aging basic data. The aging basic data includes the upper limit of the rated power of the aging rack and the temperature control range of the aging rack.
[0051] In the server aging test production line, each aging rack serves as an independent power and temperature control unit, and its power budget and temperature control range are insurmountable constraints within the production cycle. To enable subsequent batch scheduling calculations, this step first involves structurally modeling the resource structure of the aging area.
[0052] Specifically, the steps for generating an aging resource domain map model based on aging baseline data are as follows:
[0053] S111: For each aging rack in the production line, collect the upper limit of the rated power of the aging rack, the temperature control range of the aging rack, and the rack position identification value corresponding to the aging rack. Collect the above data according to the aging rack as the granularity to form the aging basic data set.
[0054] In one specific embodiment, the following data acquisition process is performed on each aging rack in the production line:
[0055] First, the rated power limit of the aging rack under the design operating conditions is read through the power supply monitoring system. This value is the maximum allowable load power of the aging rack under the conditions of rated power supply line capacity, switch capacity and safety factor. For example, the rated power limit of an aging rack can be set to 18kW or 22kW, and the specific value is determined by the electrical design parameters.
[0056] Next, the temperature control range value corresponding to the aging rack is read through the temperature control management system. The temperature control range value is in the form of a closed range, such as [20℃, 35℃] or [18℃, 32℃], indicating that the server can operate stably within this temperature control range;
[0057] Then, a unique rack location identifier is assigned to each aging rack. The rack location identifier uses a structured coding format, for example, "R05-A02-03", where:
[0058] R05 indicates the fifth row of aging zones;
[0059] A02 indicates the second power supply circuit;
[0060] 03 indicates the third physical rack position;
[0061] The rack position identifier remains unique throughout the entire production line cycle and is used to locate the actual loading position of the server.
[0062] Finally, the above three types of data are aggregated at the aging rack level to form the following structured data records:
[0063] {Rack location identification value, maximum rated power of aging rack, temperature control range of aging rack}
[0064] A basic aging data set is formed by summarizing the data from all aging racks.
[0065] S112: Based on the aging basic data set, each aging rack is mapped as a resource node in the aging resource domain graph model, and the corresponding aging rack rated power upper limit value, aging rack temperature control range value and rack position identification value are recorded in each resource node to generate an aging resource node set containing multiple resource nodes.
[0066] After obtaining the basic aging data set, a graph structure mapping is performed on each aging rack data record.
[0067] The specific process is as follows:
[0068] First, a resource node is created for each aging rack in the aging resource domain graph model. The node identifier of the resource node adopts the same encoding as the rack position identifier value.
[0069] Then, attribute fields are created within the resource node to store the following information:
[0070] Upper limit of rated power of aging rack;
[0071] Temperature control range for aging rack;
[0072] Shelf space identification value;
[0073] For example:
[0074] Node N(R05-A02-03):
[0075] Maximum power: 22kW;
[0076] Temperature control range: [18℃, 32℃];
[0077] Shelf location identification value: R05-A02-03;
[0078] When there are 20 aging racks in the production line, an aging resource node set containing 20 resource nodes is formed.
[0079] This set of resource nodes forms the node foundation of the aging resource domain graph model and is used for resource matching in subsequent batch scheduling.
[0080] S113: Based on the physical adjacency relationship, power supply circuit association relationship and temperature control linkage relationship of the aging rack in the actual production line, establish an edge connection between any two mutually related resource nodes in the aging resource node set in the aging resource domain graph model, and record the corresponding power coupling attribute and temperature control coupling attribute in the edge connection to form the aging resource domain graph model.
[0081] After the aging resource node set is established, the actual relationship between aging racks needs to be considered, because different aging racks may share power or have temperature control linkage.
[0082] Specifically, the rules for establishing edge connections are as follows:
[0083] Physical adjacency: If two aging racks are physically adjacent and share the same cold air circulation area, then an edge connection is established between the corresponding resource nodes.
[0084] Power supply circuit association: If two aging racks are connected to the same power distribution circuit, an edge connection is established between the two resource nodes, and the power coupling attribute is recorded;
[0085] The power coupling attributes include the shared circuit number, total circuit capacity, and allocation ratio.
[0086] For example, if two aging racks share a circuit with a total capacity of 40kW, then the total capacity of the circuit is recorded as 40kW in the edge connection.
[0087] Temperature control linkage: If multiple aging racks are located in the same temperature control area and are regulated by the same air conditioning control unit, then an edge connection is established between resource nodes and the temperature control coupling attribute is recorded;
[0088] Among them, the temperature control coupling attributes include: the temperature control zone number, the overall set temperature range of the temperature control zone, and the linkage adjustment flag;
[0089] For example, if aging rack A and aging rack B are located in the same temperature control zone T3, then the temperature control zone number is recorded as T3 in the edge connection.
[0090] Through the above steps, an aging resource domain graph model with the following structure is formed:
[0091] In this model, nodes represent independent aging racks, edges represent resource coupling relationships, node attributes represent power upper limits and temperature control ranges, and edge attributes represent power coupling and temperature control coupling.
[0092] This graph model is used for resource constraint verification and linkage constraint analysis during subsequent candidate batch calculations.
[0093] S12: Obtain the server serial number, test recipe version number, power curve value, and temperature requirement range value; perform encoding calculations based on the server serial number, test recipe version number, and rack identifier value to generate a binding fingerprint value, and generate a test instance identifier value based on the binding fingerprint value;
[0094] Before servers enter the aging test phase, each server needs to form a combined identifier structure that simultaneously represents "server physical identity - test recipe version - rack binding relationship". This combined identifier structure remains stable during batch reordering or rack migration to ensure that identity breaks or duplicate tests do not occur during scheduling and refactoring.
[0095] To address this, a two-layer identification system is constructed by generating a binding fingerprint value and a test instance identifier value. The binding fingerprint value is used to solidify the combination relationship between the server and the rack, while the test instance identifier value is used to distinguish instance records in different aging batches.
[0096] Specifically, the steps for generating the bound fingerprint value and the test instance identifier value are as follows:
[0097] S121: For each server to be aged, read the server serial number, test recipe version number and rack position identifier value pre-assigned to the server from the production line execution system, call the power curve value corresponding to the server under the test recipe version number from the power test system, and retrieve the temperature requirement range value corresponding to the server under the test recipe version number from the product specification library to form a test basic parameter quadruple that corresponds one-to-one with the server.
[0098] In one specific embodiment, the server serial number is a factory-unique code, using a fixed-length string format, such as "SN240315000876"; the test formula version number is used to distinguish different test conditions, such as "TFV-03.12", which represents the 12th revision of the third-generation pressure test formula; the rack position identification value has the same encoding structure as the rack position identification value in step S11, such as "R05-A02-03".
[0099] The power curve value is a set of data corresponding to time and power, for example:
[0100] {(t0,380W), (t1,420W), (t2,510W)…(tn,390W)}, where the time unit is minutes and the power unit is watts, covering the complete aging cycle; the temperature requirement range is in closed interval form, for example, [22℃, 30℃].
[0101] The server serial number, test recipe version number, power curve value, and temperature requirement range value are combined to form a four-tuple of basic test parameters:
[0102] {Server serial number, test recipe version number, power curve value, temperature requirement range}.
[0103] This quadruple serves as the basic input data structure for subsequent encoding and scheduling calculations.
[0104] S122: Based on the test basic parameter quadruple, the server serial number, test recipe version number and rack identifier value are concatenated at the character level according to the preset field concatenation order. The character-level concatenation result is then hashed and a checksum generation operation is performed on the binding encoding function to generate a unique binding fingerprint value corresponding to the combination relationship between the server and the rack.
[0105] In practice, the character-level concatenation order remains fixed: server serial number first, test recipe version number second, and rack identifier value last. No delimiters are inserted during the concatenation process to avoid discrepancies in hash results caused by different delimiters.
[0106] For example:
[0107] Server serial number: SN240315000876
[0108] Test recipe version number: TFV0312
[0109] rack position identification value: R05A0203
[0110] The splicing result is:
[0111] SN240315000876TFV0312R05A0203
[0112] The concatenated string serves as input to the binding encoding function.
[0113] Specifically, the operation logic of the binding encoding function is as follows:
[0114] S122.1: Based on the server serial number and the test recipe version number, the server serial number is used as the high-order field and the test recipe version number is used as the low-order field for fixed-length encoding, and the server serial number encoding and the test recipe version number encoding are concatenated in a fixed order to form a server identification segment;
[0115] During implementation, the server serial number uses a fixed-length encoding, padding with zeros at the leading edges if necessary; the test recipe version number also uses a fixed-length encoding, padding with zeros if necessary. The encoding length is set during system deployment and remains unchanged. The concatenation order is server serial number encoding first, followed by test recipe version number encoding.
[0116] S122.2: Based on the shelf location identifier value, the shelf location identifier value is formatted and encoded according to a preset length to form a shelf location identifier segment;
[0117] The rack position identifier value is modified by removing connectors and padding with zeros to ensure it is the correct length. For example, "R05-A02-03" is converted to "R05A0203". If the length is insufficient, zeros are added. This coding rule is kept consistent throughout the production line scheduling system to avoid generating multiple coding forms for the same rack position.
[0118] S122.3: Concatenate the server identifier segment and the rack identifier segment in the order of server identifier segment first and rack identifier segment last. Perform multiple rounds of hash operation on the concatenation result to extract a fixed length hash result as the main body of the binding fingerprint value. Calculate the check bit based on the main body of the binding fingerprint value and append the check bit to the end of the main body of the binding fingerprint value to obtain the binding fingerprint value used to identify the combination relationship between the server, test recipe version number and rack identifier value.
[0119] The hash operation can use a standard cryptographic hash algorithm, such as SHA-256 or an equivalent algorithm, and perform two rounds of hashing on the concatenated result; extract a fixed length of characters from the final hash result as the main body of the bound fingerprint value, such as extracting the first 24 hexadecimal characters;
[0120] The check digit calculation rule is as follows: each character in the main body of the bound fingerprint value is converted into a numerical value and then summed according to a preset weight. The summation result is then moduloed, and the result is used as the check digit. This check digit is appended to the end of the main body of the bound fingerprint value to generate the final bound fingerprint value.
[0121] Through the above processing, different binding fingerprint values are generated when the same server is in different racks; different binding fingerprint values are generated when the same server is in the same rack but has different test recipe version numbers; and the binding fingerprint values generated when the same server, the same test recipe version number, and the same rack are combined are consistent.
[0122] It should be noted that, in order to ensure zero collisions under massive test data, the check bit can adopt a combination of cyclic redundancy check and XOR weighted algorithm. Even if the serial numbers of two servers are extremely similar, or the physical locations of two racks are adjacent, through multiple rounds of shift operations, the generated binding fingerprint value will have a bit difference of more than 50% at the binary level, i.e., the avalanche effect, thereby eliminating the possibility of scheduling identity conflicts at the underlying logic level.
[0123] S123: Based on the bound fingerprint value and the power curve value and temperature requirement range value that correspond one-to-one with the bound fingerprint value, batch number information and timestamp information are superimposed according to the preset instance numbering rules to generate a unique test instance identifier value in the current aging batch, and a one-to-one mapping relationship is established between the test instance identifier value and the bound fingerprint value.
[0124] In one specific embodiment, the current aging batch number is first obtained, for example, "BATCH20240315001"; then the current timestamp information is obtained, for example, "20240315123045".
[0125] Following the preset instance numbering rules, the batch number information is placed in the high-order field, the bound fingerprint value is placed in the middle field, and the timestamp information is placed in the low-order field, and then concatenated sequentially to generate the test instance identifier value, for example:
[0126] BATCH202403150013FA8C91E78BD34A9201F56C3720240315123045
[0127] The test instance identifier value remains unique within the current aging batch.
[0128] Subsequently, a one-to-one mapping relationship between test instance identifiers and bound fingerprint values is established in the database for resolving server-rack combination relationships during subsequent scheduling. When batch reordering occurs, test instance identifiers can be regenerated, but bound fingerprint values remain unchanged, thus ensuring that server-rack combination relationships are not lost due to batch adjustments.
[0129] The above steps enable hierarchical management of server identity identifiers and batch instance identifiers.
[0130] S13: Generate candidate batch results based on the aging resource domain graph model and test instance identifier values. Calculate the batch power time series value by time alignment and superposition based on the power curve value. Calculate the temperature control valid identifier value by interval intersection based on the temperature requirement interval value and the aging rack temperature control interval value. Retain candidate batch results where the batch power time series value does not exceed the upper limit of the aging rack rated power at any time and the temperature control valid identifier value is true.
[0131] After completing step S11 to construct the aging resource domain graph model and step S12 to generate test instance identifier values, each test instance identifier value already has server identity information, test recipe information, and rack binding information.
[0132] Step S13 performs batch formation and constraint screening calculations to ensure that the power budget and temperature control range constraints are simultaneously valid in the time dimension.
[0133] Specifically, the steps for retaining candidate batch results where the batch power timing value does not exceed the upper limit of the rated power of the aging rack at any given time and the temperature control valid identification value is true are as follows:
[0134] S131: Based on the rack position identifier value recorded by each resource node in the aging resource domain graph model, for each test instance identifier value, call the binding fingerprint value that corresponds one-to-one with the test instance identifier value, and parse the target rack position identifier value from the binding fingerprint value. Assign the test instance identifier value to the resource node in the aging resource domain graph model whose rack position identifier value is consistent with the target rack position identifier value. In each resource node, collect all the test instance identifier values corresponding to the resource node to form an initial candidate batch set that corresponds one-to-one with each resource node.
[0135] In one specific embodiment, the system first traverses all test instance identifier values to be scheduled in the current batch; for each test instance identifier value, the bound fingerprint value is obtained through the mapping relationship between the test instance identifier value and the bound fingerprint value; according to the encoding rules of step S122, the rack position identifier segment is parsed from the bound fingerprint value, and the target rack position identifier value is restored, for example, "R05-A02-03" is parsed.
[0136] Then, in the aging resource domain graph model, find the resource node with the rack identifier value "R05-A02-03" and classify the test instance identifier value into the candidate set corresponding to the resource node;
[0137] Once all test instance identifier values have been categorized, an initial candidate batch set is formed within each resource node, for example:
[0138] Resource node R05-A02-03: {Test instance identifier value 1, Test instance identifier value 2, Test instance identifier value 3}
[0139] Resource node R05-A02-04: {Test instance identifier value 4, Test instance identifier value 5}
[0140] This forms an initial candidate batch set that corresponds one-to-one with each resource node.
[0141] S132: For each candidate batch in the initial candidate batch set, call the power curve value and temperature requirement range value corresponding to the test instance identifier value in the candidate batch, and combine the rated power upper limit value and temperature control range value of the aging rack recorded in the resource node where the candidate batch is located, and generate the batch power timing value based on the power curve value and the predetermined batch power timing calculation logic.
[0142] Specifically, the batch power timing calculation logic is as follows:
[0143] S132.1: For each candidate batch, determine a uniform time step and a common time axis that covers the aging cycle of all test instances in the candidate batch. Then, perform interpolation and time axis alignment on the power curve value corresponding to each test instance in the candidate batch according to the time step to generate multiple aligned power curves defined on the common time axis.
[0144] In one embodiment, the uniform time step is set to 1 minute, and the common time axis covers the longest aging period. For example, when the longest aging period in the candidate batch is 120 minutes, the common time axis is defined as t0 to t120.
[0145] If the original power curve sampling interval of a test instance is 5 minutes, then it is linearly interpolated to convert it into power data with a 1-minute interval; if the aging cycle of a test instance is only 90 minutes, then its power value is regarded as 0W after 90 minutes or treated as 0 power after the test ends.
[0146] Through the above processing, each test instance in the candidate batch corresponds to an aligned power curve with the same length and time axis.
[0147] S132.2: For each time step on the common time axis, perform a point-by-point superposition operation on the power dimension on all aligned power curves corresponding to that time step to obtain the batch superposition power value corresponding to each time step, and combine the batch superposition power values of each time step in time order to generate the batch power time series value used for constraint detection.
[0148] For example, at time step t10, the power values of the three servers in the candidate batch are 450W, 520W and 380W respectively, then the batch superposition power value at this time step is 1350W.
[0149] Arrange the superimposed power values of all time steps in chronological order to form batch power time series values, for example:
[0150] {(t0, 920W), (t1, 1050W)…(t120, 480W)}
[0151] Subsequently, the superimposed power value of each time step in the batch power time series value is compared with the upper limit of the rated power of the aging rack recorded by the resource node. If the superimposed power value of any time step exceeds the upper limit of the rated power of the aging rack, the candidate batch is determined to be unsatisfactory in terms of power dimension.
[0152] S133: Generate a temperature control valid identifier value based on the temperature requirement range, the aging rack temperature control range, and the preset temperature control valid identifier calculation logic;
[0153] Specifically, the calculation logic for the temperature control legality identifier is as follows:
[0154] S133.1: For each candidate batch, call the temperature requirement interval value corresponding to each test instance in the candidate batch and perform interval intersection operation with the aging rack temperature control interval value corresponding to the aging rack where the candidate batch is located, to obtain the intersection interval of the temperature requirement interval value of each test instance and the aging rack temperature control interval value.
[0155] For example, if the temperature requirement range for a server is [22℃, 30℃], and the temperature control range for the aging rack is [20℃, 28℃], then the intersection range is [22℃, 28℃].
[0156] If the required temperature range for a server is [30℃, 40℃], and the temperature control range for the aging rack is [20℃, 28℃], then the two ranges have no overlap, and the overlapping range is empty.
[0157] S133.2: Determine whether the intersection intervals corresponding to all test instances in the candidate batch are all non-empty intervals. When all intersection intervals are non-empty, set the temperature control legality flag value corresponding to the candidate batch to true. When there is at least one intersection interval that is empty, set the temperature control legality flag value corresponding to the candidate batch to false. This is used to further filter the candidate batch results based on temperature control conditions under the premise that the batch power timing value meets the power constraint.
[0158] When the batch power timing value does not exceed the upper limit of the rated power of the aging rack at any time, and the temperature control legal identification value is true, the result of the candidate batch is retained;
[0159] The candidate batch result is rejected when the power at any time step exceeds the upper limit of the rated power of the aging rack, or when the temperature control legal identification value is false.
[0160] Through the above steps, it is possible to verify whether the power superposition meets the upper limit constraint of the rated power of the aging rack in the time dimension, and to verify whether the temperature requirement interval of all servers falls within the temperature control interval of the aging rack in the interval dimension. This ensures that the candidate batch results are valid under the dual constraints of power and temperature control, and provides a set of candidate batches that meet multiple constraints for the generation of subsequent rack loading results.
[0161] S14: Generate rack loading results based on candidate batch results and rack identification values, and generate a binding continuity verification chain based on the bound fingerprint value and rack loading results; when an offset is detected in the rated power upper limit value of the aging rack or the temperature control range value of the aging rack, filter the test instance identification values that can be rearranged based on the binding continuity verification chain, and regenerate the candidate batch results and rack loading results.
[0162] After completing step S13 to screen and obtain candidate batch results that meet the dual constraints of power and temperature control, the candidate batches need to be structured and loaded at the physical rack level, and at the same time, a traceable continuous chain record needs to be established to ensure that local rearrangement can be carried out without disrupting the continuity of test instances when resource parameter offsets occur in the future.
[0163] Specifically, the steps for regenerating candidate batch results and rack loading results are as follows:
[0164] S141: For each candidate batch in the candidate batch results, call the binding fingerprint value corresponding to each test instance identifier value in the candidate batch, and parse out the rack position identifier value that corresponds one-to-one with the test instance identifier value from the binding fingerprint value. Group and sort the test instance identifier values in the same candidate batch according to the rack position identifier value to generate rack position loading results with rack position identifier value as index and corresponding test instance identifier value sequence as content.
[0165] In one specific embodiment, the system first reads all test instance identifier values within a candidate batch, for example:
[0166] {T1, T2, T3, T4};
[0167] Then, the corresponding bound fingerprint value is obtained through the mapping relationship between the test instance identifier value and the bound fingerprint value, and the rack position identifier segment is parsed from the bound fingerprint value according to the encoding rules in step S122, for example:
[0168] T1 → R05-A02-03
[0169] T2 → R05-A02-03
[0170] T3 → R05-A02-04
[0171] T4 → R05-A02-04
[0172] They were then grouped according to the rack position identifier value, and sorted according to the generation time or number order of the test instance identifier value, forming the following structure:
[0173] R05-A02-03:{T1, T2}
[0174] R05-A02-04:{T3, T4}
[0175] This structure represents the result of the rack loading.
[0176] The rack loading results are stored in the database in a structured format, with "rack identifier value" as the index field and "test instance identifier value sequence" as the loading content field, for subsequent continuity verification and rearrangement operations.
[0177] S142: Based on the test instance identifier value, its corresponding binding fingerprint value, and the batch number information of the candidate batch to which the test instance identifier value belongs, recorded in the rack loading results, the occurrence relationship of the same test instance identifier value in different loading stages or different batches is linked into a binding continuity verification chain node sequence according to the time sequence and loading sequence. A binding continuity verification chain from the initial loading to the current loading state is constructed for each test instance identifier value to indicate the continuity and change record of the test instance identifier value in the rack loading process.
[0178] In the specific implementation process, the following chain structure is established for the identifier value of each test instance:
[0179] The node content includes:
[0180] {Test instance identifier, bound fingerprint value, batch number information, rack position identifier, loading timestamp};
[0181] When a test instance identifier is reloaded in a subsequent batch, a new node is generated and linked to the previous node, forming a unidirectional chain structure.
[0182] For example, if a test instance identifier value T1 is loaded into R05-A02-03 in the first batch and into R05-A02-04 in the second batch, then its binding continuity check chain is:
[0183] Node 1: {T1, fingerprint A, batch B1, R05-A02-03, timestamp 1}
[0184] Node 2: {T1, fingerprint A, batch B2, R05-A02-04, timestamp 2}
[0185] A continuous chain structure is formed by recording the preceding and following pointers between nodes.
[0186] This structure ensures that: the bound fingerprint value of a test instance identifier remains consistent on the chain, shelf location change records are complete, batch change records are complete, and loading order is traceable.
[0187] S143: When the monitoring module detects that the upper limit of the rated power of any aging rack or the temperature control range of the aging rack is offset from the original record in the aging basic data, it calls the rack position identifier value corresponding to the aging rack, and searches all related node sequences in the binding continuity verification chain based on the rack position identifier value, filters out the test instance identifier value that is still on the aging rack during the offset effective time period, marks the above test instance identifier value as a reorderable test instance identifier value, and removes the reorderable test instance identifier value from the original rack position loading result;
[0188] In one embodiment, if the monitoring module detects that the rated power limit of the aging rack R05-A02-03 has decreased from 22kW to 18kW, the system first determines the effective timestamp of the offset.
[0189] Then, in the binding continuity check chain, search for the node with rack identifier value R05-A02-03 and loading time within the offset effective time period;
[0190] Extract the corresponding test instance identifier value, such as {T1, T2};
[0191] Mark the above test instance identifier values as reorderable test instance identifier values;
[0192] Then delete the corresponding record from the current rack loading results and update the rack loading results.
[0193] It should be noted that the monitoring module collects current sensor data from the power distribution cabinet and air duct pressure data from the temperature control system in real time via the edge computing gateway of the production line. Internally, the module runs an outlier detection model, which pre-learns the parameter fluctuation curves of the aging rack under normal load. When the actual measured upper limit of rated power or temperature control range deviates from the predicted baseline by more than 5% due to hardware aging, fan failure, or sudden changes in ambient temperature, the system automatically determines this as a "parameter offset" and triggers subsequent binding continuity check chain retrieval and rearrangement logic.
[0194] This process does not change the bound fingerprint value or the test instance identifier value; it only changes the rack loading status.
[0195] S144: Based on the updated rated power upper limit and temperature control range of the aging rack, the reorderable test instance identifier value is used as the set to be reordered and re-input into the candidate batch generation logic and the batch power timing value and temperature control legal identifier value calculation logic to generate new candidate batch results. Based on the new candidate batch results, the rack loading result generation step is executed again to obtain updated rack loading results that match the latest rated power upper limit and temperature control range of the aging rack.
[0196] In the specific implementation process, the set of reorderable test instance identifier values is input into the candidate batch generation logic of step S13, and the time alignment superposition calculation and interval intersection calculation are re-executed.
[0197] After generating new candidate batch results, proceed to step S141 to load the racks.
[0198] Throughout the process, the test instance identifier value and the bound fingerprint value remain unchanged; only the batch number information and loading record are updated.
[0199] Through the above processing, when the aging rack resource parameters change, only the affected test instances are partially rearranged. Secondly, the test instances are not renumbered and the test timeline is not reset. At the same time, the binding continuity verification chain is kept intact, so that the updated rack loading results match the latest aging rack rated power upper limit and aging rack temperature control range value.
[0200] It should be noted that the weights or coefficients in this embodiment are all set after statistical analysis of historical experimental data.
[0201] Example 2
[0202] This embodiment discloses an artificial intelligence-based server production line scheduling device, specifically an electronic device, comprising:
[0203] Memory, used to store computer software programs;
[0204] A processor is used to read and execute the computer software program, thereby implementing any of the aforementioned AI-based server production line scheduling methods.
[0205] Example 3
[0206] This embodiment discloses a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements any one of the artificial intelligence-based server production line scheduling methods described above.
[0207] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. An artificial intelligence-based server production line scheduling method, characterized by, The method includes: Acquire basic aging data and rack identification values, and generate an aging resource domain map model based on the basic aging data. The basic aging data includes the upper limit of the rated power of the aging rack and the temperature control range of the aging rack. Obtain the server serial number, test recipe version number, power curve value, and temperature requirement range value; perform encoding calculations based on the server serial number, test recipe version number, and rack identifier value to generate a binding fingerprint value, and generate a test instance identifier value based on the binding fingerprint value; Candidate batch results are generated based on the aging resource domain graph model and test instance identifier values. The batch power time series value is obtained by time alignment and superposition calculation based on the power curve value. The temperature control legal identifier value is obtained by interval intersection calculation based on the temperature requirement interval value and the aging rack temperature control interval value. Candidate batch results with batch power time series value not exceeding the aging rack rated power upper limit value and temperature control legal identifier value being true are retained. Based on the candidate batch results and rack position identification values, rack position loading results are generated, and a binding continuity verification chain is generated based on the bound fingerprint values and rack position loading results. When an offset is detected in the upper limit of the rated power of the aging rack or the temperature control range of the aging rack, the test instance identification values that can be rearranged are filtered based on the binding continuity verification chain, and the candidate batch results and rack position loading results are regenerated.
2. The server production line scheduling method based on artificial intelligence according to claim 1, characterized in that, The steps for generating an aging resource domain map model based on aging baseline data are as follows: For each aging rack in the production line, the upper limit of the rated power of the aging rack, the temperature control range of the aging rack, and the rack position identification value corresponding to the aging rack are collected. The above data are collected according to the aging rack as the granularity to form a basic aging data set. Based on the aging basic data set, each aging rack is mapped as a resource node in the aging resource domain graph model, and the corresponding aging rack rated power upper limit value, aging rack temperature control range value and rack position identification value are recorded in each resource node to generate an aging resource node set containing multiple resource nodes. Based on the physical adjacency, power supply circuit association, and temperature control linkage of the aging rack in the actual production line, an edge connection is established between any two interconnected resource nodes in the aging resource node set in the aging resource domain graph model. The corresponding power coupling attributes and temperature control coupling attributes are recorded in the edge connection to form the aging resource domain graph model.
3. The server production line scheduling method based on artificial intelligence according to claim 2, characterized in that, The steps to generate the bound fingerprint value and test instance identifier value are as follows: For each server to be aged, the server serial number, test recipe version number and rack position identifier value pre-assigned to the server are read from the production line execution system. The power curve value corresponding to the server under the test recipe version number is called from the power test system. The temperature requirement range value corresponding to the server under the test recipe version number is retrieved from the product specification library to form a test basic parameter quadruple that corresponds one-to-one with the server. Based on the test basic parameter quadruple, the server serial number, test recipe version number and rack identifier value are concatenated at the character level according to the preset field concatenation order. The character-level concatenation result is then hashed and a checksum generation operation is performed on the binding encoding function to generate a unique binding fingerprint value corresponding to the combination relationship between the server and the rack. Based on the bound fingerprint value and the power curve value and temperature requirement range value that correspond one-to-one with the bound fingerprint value, batch number information and timestamp information are superimposed according to the preset instance numbering rules to generate a unique test instance identifier value in the current aging batch, and a one-to-one mapping relationship is established between the test instance identifier value and the bound fingerprint value.
4. The server production line scheduling method based on artificial intelligence according to claim 3, characterized in that, The operation logic of the binding encoding function is as follows: Based on the server serial number and the test recipe version number, the server serial number is used as the high-order field and the test recipe version number is used as the low-order field for fixed-length encoding. The server serial number encoding and the test recipe version number encoding are then concatenated in a fixed order to form the server identification segment. Based on the shelf location identifier value, the shelf location identifier value is formatted and encoded according to a preset length to form a shelf location identifier segment; The server identifier segment and the rack identifier segment are concatenated in the order of server identifier segment first and rack identifier segment last. The concatenation result is subjected to multiple rounds of hash operation to extract a fixed length hash result as the main body of the binding fingerprint value. The check bit is calculated based on the main body of the binding fingerprint value and appended to the end of the main body of the binding fingerprint value to obtain the binding fingerprint value used to identify the combination relationship of server, test recipe version number and rack identifier value.
5. The server production line scheduling method based on artificial intelligence according to claim 4, characterized in that, The steps to retain candidate batches whose batch power timing values do not exceed the upper limit of the rated power of the aging rack at any given moment and whose temperature control valid flag value is true are as follows: Based on the rack position identifier value recorded by each resource node in the aging resource domain graph model, for each test instance identifier value, the binding fingerprint value corresponding to the test instance identifier value is called, and the target rack position identifier value is parsed from the binding fingerprint value. The test instance identifier value is assigned to the resource node in the aging resource domain graph model whose rack position identifier value is consistent with the target rack position identifier value. All test instance identifier values corresponding to the resource node are collected in each resource node to form an initial candidate batch set corresponding to each resource node. For each candidate batch in the initial candidate batch set, the power curve value and temperature requirement range value corresponding to the test instance identifier value in the candidate batch are called, and the rated power upper limit value and temperature control range value of the aging rack recorded in the resource node where the candidate batch is located are combined to generate the batch power timing value based on the power curve value and the predetermined batch power timing calculation logic. Based on the temperature requirement range, the aging rack temperature control range, and the preset temperature control valid identifier calculation logic, a temperature control valid identifier value is generated.
6. The server production line scheduling method based on artificial intelligence according to claim 5, characterized in that, The batch power timing calculation logic is as follows: For each candidate batch, a unified time step and a common time axis covering the aging cycle of all test instances in the candidate batch are determined. The power curve values corresponding to each test instance in the candidate batch are interpolated and time axis aligned according to the time step to generate multiple aligned power curves defined on the common time axis. For each time step on the common time axis, perform a point-by-point superposition operation on the power dimension on all aligned power curves corresponding to that time step to obtain the batch superimposed power value corresponding to each time step. Then, combine the batch superimposed power values of each time step in chronological order to generate the batch power time series value used for constraint detection.
7. The server production line scheduling method based on artificial intelligence according to claim 6, characterized in that, The calculation logic for the valid temperature control identifier is as follows: For each candidate batch, the temperature requirement range value corresponding to each test instance in the candidate batch is called, and the interval intersection operation is performed with the temperature control range value of the aging rack corresponding to the aging rack where the candidate batch is located, to obtain the intersection range of the temperature requirement range value of each test instance and the temperature control range value of the aging rack. Determine whether all the intersection intervals corresponding to the test instances within the candidate batch are non-empty intervals. If all intersection intervals are non-empty, set the temperature control valid identifier value corresponding to the candidate batch to true. If at least one intersection interval is empty, set the temperature control valid identifier value corresponding to the candidate batch to false. This is used to further filter the candidate batch results based on temperature control conditions, provided that the batch power timing value meets the power constraint.
8. The server production line scheduling method based on artificial intelligence according to claim 7, characterized in that, The steps to regenerate the candidate batch results and rack loading results are as follows: For each candidate batch in the candidate batch results, the bound fingerprint value corresponding to each test instance identifier value in the candidate batch is called, and the rack position identifier value corresponding one-to-one with the test instance identifier value is parsed from the bound fingerprint value. The test instance identifier values in the same candidate batch are grouped and sorted according to the rack position identifier value to generate rack position loading results with rack position identifier value as index and corresponding test instance identifier value sequence as content. Based on the test instance identifier value, its corresponding binding fingerprint value, and the batch number information of the candidate batch to which the test instance identifier value belongs, recorded in the rack loading results, the occurrence relationship of the same test instance identifier value in different loading stages or different batches is linked into a binding continuity verification chain node sequence according to the time sequence and loading sequence. A binding continuity verification chain from the initial loading to the current loading state is constructed for each test instance identifier value to indicate the continuity and change record of the test instance identifier value in the rack loading process. When the monitoring module detects an offset between the rated power limit or temperature control range value of any aging rack and the original record in the aging baseline data, it calls the rack position identifier value corresponding to that aging rack and, based on the rack position identifier value, retrieves all associated node sequences in the binding continuity verification chain, filters out the test instance identifier values that are still on the aging rack during the offset effective period, marks the above test instance identifier values as reorderable test instance identifier values, and removes the reorderable test instance identifier values from the original rack position loading results.
9. A server production line scheduling device based on artificial intelligence, characterized in that, include: Memory, used to store computer software programs; A processor is used to read and execute the computer software program, thereby implementing the AI-based server production line scheduling method according to any one of claims 1 to 8.
10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements the artificial intelligence-based server production line scheduling method according to any one of claims 1 to 8.