Unmanned workshop equipment collaborative intelligent optimization method

Abstract

The invention discloses an unmanned workshop equipment collaborative intelligent optimization method, which comprises the following steps of: acquiring and processing multi-dimensional information such as equipment health degree, load rate and operation state in real time through an equipment state sensing module, and acquiring a high-credibility state feature vector by using a time-frequency analysis and normalization technology; a multi-target dynamic priority evaluation function is constructed in combination with task remaining periods and original priorities, and priority self-adaptive sorting of tasks with resource conflicts is achieved; in the scheduling process, the equipment and task states are continuously monitored, reordering is automatically triggered when abnormal changes are detected, dynamic optimization of the scheduling scheme is achieved, the utilization rate of production equipment and the task completion efficiency can be improved, and the self-adaptability and robustness of an unmanned workshop scheduling system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned workshop production scheduling, and particularly relates to a device collaborative intelligent optimization method for an unmanned workshop. BACKGROUND

[0002] Currently, the task scheduling optimization technology in the field of intelligent manufacturing and unmanned workshops is rapidly developing, especially in multi-device collaboration, dynamic resource allocation, and task priority management. The mainstream approach in the industry usually focuses on static task scheduling strategies, which pre-set fixed task priority weights, combine historical operation data of the production process and device capacity parameters, and generate global or batch task execution plans. This type of solution mainly focuses on static indicators such as task arrival order, duration estimation, and processing complexity, emphasizing the rationality of task queues and the balance of resource allocation.

[0003] In recent years, with the improvement of the intelligent level of workshops, some scheduling systems have begun to incorporate limited real-time information, such as task execution progress or device start and stop status. However, these information still play a shallow auxiliary role in the scheduling model, only used for simple exception correction or redistribution, and unable to deeply integrate device operating status and resource dynamic characteristics. In addition, most existing technologies use time-sharing polling or batch scheduling by time period, lacking a fine-grained task / device adaptation mechanism. They can only passively handle sudden device failures, load fluctuations, and resource conflicts, and cannot achieve intelligent real-time reordering of task priorities.

[0004] In actual application scenarios such as SMT patch production lines, electronic component assembly lines, and automated workshops, there are numerous devices, complex task types, and frequent concurrent task competition for the same device, leading to resource conflicts becoming a key bottleneck affecting production efficiency. Existing technologies often use classic scheduling algorithms such as first-come-first-served (FCFS), static weighted priority, or time-slice round-robin. When tasks compete for the same device, they cannot dynamically adjust the task execution order based on the current "health" of the device (such as reliability assessment, fault probability), "load rate" (such as current energy consumption, processing intensity), and other state parameters. Furthermore, the state changes of devices in the production process (such as temperature abnormalities, severe vibrations, or load overruns) often cannot be sensed by the scheduling system in a timely manner and fed back to the task priority evaluation algorithm, resulting in high-health idle devices being left idle for a long time, while high-load, potentially faulty devices are overloaded with tasks, causing resource waste, task delays, and even shutdown failures.

[0005] Although some advanced scheduling optimization systems try to integrate limited device state information, such as device occupancy / idle flag, estimated release time label, or use simple device health diagnosis model, they have not formed a standardized and dynamic priority reordering mechanism. The scheduling system still lacks intelligent feedback and scheduling weight adjustment capability for complex device states in multi-target conflict scenarios. SUMMARY

[0006] The present application provides an unmanned workshop device collaborative intelligent optimization method to solve the above technical problems.

[0007] The technical solution of the present application is as follows: an unmanned workshop device collaborative intelligent optimization method, comprising:

[0008] S1: deploying a device state perception module in the unmanned workshop, collecting health degree, load rate and running state data of each production device in real time, and recording task execution progress information;

[0009] S2: filtering and normalizing the collected device state data to eliminate abnormal fluctuations and dimension differences, and generating a standardized device state vector;

[0010] S3: based on the current task queue and resource allocation state, constructing a task-resource dependency graph, and detecting whether there is a conflict situation of multiple tasks competing for the same device or resource;

[0011] S4: if resource conflict is detected, constructing a multi-objective dynamic priority evaluation function according to the standardized device state vector, task remaining duration and priority original weight;

[0012] S5: inputting the device health degree and load rate as feedback factors into the dynamic priority evaluation function to calculate the real-time priority score of each conflict task;

[0013] S6: based on the priority score, sorting the conflict tasks to generate an optimized task execution sequence, and issuing the updated task scheduling scheme to the collaborative control execution module;

[0014] S7: continuously monitoring the changes of device state and task progress during task execution, and when detecting that the device state mutation or task progress deviation exceeds the set threshold, retriggering the task priority reordering mechanism;

[0015] S8: recording the execution time, scheduling result and execution feedback of each priority reordering for subsequent online learning and parameter optimization of the scheduling model.

[0016] The unmanned workshop device collaborative intelligent optimization method provided by the present application has the following beneficial effects:

[0017] (1) Traditional workshop task scheduling is mostly based on preset priority, task duration or static process flow, lacking dynamic integration of actual running state of equipment, which is easy to cause equipment idling or overload, task delay and other phenomena. The present application realizes real-time response of task priority evaluation to equipment availability through whole-process perception and standardized processing of equipment health degree, load rate and real-time running state;

[0018] (2) The present application innovatively deeply integrates multi-modal equipment state data and task execution progress, and adopts dynamic conflict detection and priority reordering mechanism. In the scene of multiple tasks competing for the same equipment or resource, by evaluating equipment load sensitivity and health degree attenuation factor, repeated allocation of equipment with high load or critical health state is effectively avoided, and equipment with good state and low load is preferentially utilized, improving the overall utilization rate and life of equipment resources;

[0019] (3) Compared with the prior art, in a dynamic environment, traditional static task priority sorting is easy to cause mismatch between task execution and equipment capacity due to failure to reflect actual equipment load and health condition. The multi-objective dynamic priority function of the present application integrates equipment state, task remaining duration, original business priority and other multi-dimensional factors, and comprehensively evaluates and dynamically corrects task execution order. In the production peak, equipment state fluctuation or sudden failure, the system can quickly adaptively adjust through real-time reordering, and the average task waiting and delay is significantly reduced, greatly improving the production efficiency and the certainty of task completion;

[0020] (4) The present application not only integrates equipment state and task information in real time, but also supports online incremental learning and dynamic optimization of parameters through whole-process data backtracking and feedback mechanism (including scheduling execution result, equipment state data and abnormal event), so that the scheduling strategy can continuously self-improve according to the actual production. The scheduling model optimizes parameters in real time through historical priority reordering effect and execution feedback, gradually adapts to the changing production environment, and continuously converges to a better scheduling strategy in long-term use, avoiding performance degradation caused by traditional rule rigidity. BRIEF DESCRIPTION OF DRAWINGS

[0021] Fig. 1 is a flow chart of a kind of unmanned workshop equipment collaborative intelligent optimization method of the present application;

[0022] Fig. 2 is a sub-flow chart of a kind of unmanned workshop equipment collaborative intelligent optimization method of the present application;

[0023] Fig. 3 is another sub-flow chart of a kind of unmanned workshop equipment collaborative intelligent optimization method of the present application. DETAILED DESCRIPTION

[0024] In order to make the objects and advantages of the present application more clear, the present application will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0025] The preferred implementation methods of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these implementation methods are only used to explain the technical principles of the present application and not intended to limit the protection scope of the present application.

[0026] As used herein, the singular forms "a", "an" and "the" also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "comprising" or "including" or "having" or the like specifies the presence of stated features, integers, steps, operations, components, parts, or combinations thereof, but does not exclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof. Meanwhile, the term "and / or" used in the specification includes any and all combinations of the associated listed items.

[0027] Please refer to Figs. 1-3 As shown in the figure, an intelligent optimization method for collaborative unmanned workshop equipment, comprising:

[0028] S1: deploying a device state perception module in the unmanned workshop, collecting health degree, load rate and running state data of each production equipment in real time, and recording task execution progress information;

[0029] S2: filtering and normalizing the collected device state data to eliminate abnormal fluctuations and dimension differences, and generating a standardized device state vector;

[0030] S3: based on the current task queue and resource allocation state, constructing a task-resource dependency graph, and detecting whether there is a conflict situation in which multiple tasks compete for the same equipment or resource;

[0031] S4: if resource conflict is detected, constructing a multi-objective dynamic priority evaluation function according to the standardized device state vector, task remaining duration and priority original weight;

[0032] S5: inputting the equipment health degree and load rate as feedback factors into the dynamic priority evaluation function to calculate the real-time priority score of each conflict task;

[0033] S6: based on the priority score, sorting the conflict tasks to generate an optimized task execution sequence, and issuing the updated task scheduling scheme to the collaborative control execution module;

[0034] S7: Continuously monitor the device state and task progress changes during task execution, and when detecting device state mutations or task progress deviations exceeding the set threshold, retrigger the task priority reordering mechanism;

[0035] S8: Record the execution time, scheduling results and execution feedback of each priority reordering, for online learning and parameter optimization of subsequent scheduling models.

[0036] The step S1: Deploy a device state perception module inside the unmanned workshop, real-time collect the health degree, load rate and running state data of each production device, and record the task execution progress information. Specifically, it includes:

[0037] S1.1: Deploy a multi-modal sensor array at the key control node of the electronic component production equipment to synchronously collect the device running voltage, current, temperature, vibration signals, and obtain the basic physical parameters of the device running state;

[0038] Select high information density areas at the key control nodes of the electronic component production equipment as the sensor deployment location to ensure that the collected signals can fully reflect the running state of the device. A multi-modal sensor array combination is adopted, including high-precision voltage sensors (range 0-480V, accuracy ±0.1%), current sensors (range 0-200A, accuracy ±0.5%), thermocouple temperature sensors (detection range -50℃ to +250℃, resolution 0.1℃) and three-axis MEMS vibration sensors (frequency response range 0.5Hz-2kHz, sensitivity ±0.03g), to realize synchronous collection of voltage, current, temperature and vibration signals;

[0039] According to the sampling theorem, the highest frequency of the 10-20 harmonics of each modal sensor is set, among which the sampling frequency of voltage and current signals is set to 20x50=1kHz to completely retain harmonic information, the sampling frequency of temperature signals is 1Hz for trend tracking, and the sampling frequency of vibration signals is 5kHz to capture high-frequency mechanical impact characteristics;

[0040] A hardware clock synchronization mechanism (accuracy ±1μs) is used to align the timestamps of each modal sampling, avoiding cross-modal feature mismatch caused by collection delay;

[0041] Further, through a multi-channel synchronous acquisition card, the output signals of each sensor are digitized, the resolution of voltage and current signals is set to 16bit to improve the dynamic range, and the resolution of temperature and vibration signals is set to 24bit to enhance the detection sensitivity of small changes, and the digitized signals are buffered in a local ring buffer to resist instantaneous network delay;

[0042] Further, the original signals collected synchronously are pre-processed, including band-pass filtering (voltage / current: 20 Hz-2 kHz, vibration: 0.5 Hz-1.5 kHz) to suppress power frequency interference and high frequency noise, and low-pass filtering (cut-off frequency 0.5 Hz) for temperature signals to remove environmental transient disturbances, so as to obtain clean and high-fidelity basic physical quantity sequences;

[0043] Through the above multi-modal sensor array arrangement and synchronous signal acquisition and processing, the dynamic changes of the equipment operating voltage, current, temperature and vibration are converted into accurately calibrated digitized signal streams, providing reliable basic data input for subsequent power spectrum feature extraction and equipment health modeling, and achieving the expected technical effect of full-dimensional perception of the equipment operating state;

[0044] For example, on a chip mounter production line, a voltage sensor (range 0-480V, accuracy ±0.1%) of model HVS-480 is deployed at the input end of the main control power supply, a Hall current sensor (range 0-200A, accuracy ±0.5%) of model HCS-200 is installed in series in the main motor loop, a K-type thermocouple (-50℃-+250℃, resolution 0.1℃) is used for fixed-point temperature measurement on the periphery of the motor bearing seat, and an ADXL355 three-axis MEMS vibration sensor (frequency response 0.5 Hz-2 kHz, sensitivity ±0.03g) is embedded in the vibration-sensitive position of the rack. The voltage and current signal sampling frequency is set to 20x50Hz=1kHz to cover up to the 25th harmonic, the temperature signal is sampled at 1Hz, and the vibration signal is sampled at 5kHz to ensure complete collection of impact events. Through a PXI-6259 multi-channel synchronous acquisition card, A / D conversion is performed, the voltage / current channel is set to 16bit, the temperature / vibration channel is set to 24bit, and the hardware common clock is synchronized. The original signals are subjected to FIR band-pass filtering (voltage / current: 20 Hz-2 kHz, vibration: 0.5 Hz-1.5 kHz) and IIR low-pass filtering (temperature: 0.5 Hz), and a stable multi-modal data stream that meets the quantization accuracy requirements is output. Experimental verification shows that the basic physical quantity acquisition system configured in this way can achieve a state fluctuation capture rate of more than 90% under typical production rhythms, and the signal-to-noise ratio is improved by about 15dB compared to before processing, providing high-quality input for subsequent device load rate calculation and health assessment;

[0045] S1.2: Based on the collected voltage and current signals, the fast Fourier transform algorithm is used to extract the power spectrum features of the equipment operation, and a device load rate feature vector is generated;

[0046] S1.3: The wavelet packet transform algorithm is applied to the temperature and vibration signals for time-frequency domain decomposition, energy entropy and kurtosis indicators are extracted, a device health assessment model is established combined with historical fault data, and a device health index is calculated;

[0047] S1.4: Transmit the device load rate feature vector and the device health index to the edge computing node through industrial Ethernet, and timestamp align with the current task execution state information of the device to generate a structured device state data packet;

[0048] S1.5: Upload the structured device state data packet to the shared memory area of the task scheduling optimization module for subsequent task priority reordering mechanism to call as the key feedback factor input of the dynamic priority evaluation function.

[0049] The step S2: filtering and normalizing the collected device state data to eliminate abnormal fluctuations and dimension differences, and generating a standardized device state vector. Specifically, it includes:

[0050] S2.1: Perform sliding window filtering on the original device health data, load rate data and running state data collected by the device state perception module to eliminate noise interference caused by transient abnormal fluctuations of the sensor and obtain the filtered device state sequence;

[0051] The original device health data, load rate data and running state data from the device state perception module are filtered using a sliding window mean filtering algorithm (the window length w is set according to the sampling frequency and the device state change period), which realizes the smoothing processing of time series data to reduce the influence of transient abnormal fluctuations;

[0052] Further, by using the sliding window median filtering method (the window length is the same as before, and the boundary compensation uses mirror extension), the suppression of high amplitude pulse noise is realized, and the de-pulsed device state sequence is obtained;

[0053] Further, for the data sequence filtered by mean and median, a sliding window variance detection algorithm is used to calculate the variance σ win of each window, and according to the set noise threshold θ var , it is determined whether there is a non-stationary fluctuation region in the current window. If it is determined that it is noise interference, a weighted smoothing operation is automatically triggered for secondary correction;

[0054] Further, an exponential weighted moving average (EWMA) algorithm (the weighting coefficient α is adaptively adjusted according to the data stationarity) is used to smooth the corrected sequence to generate a filtered device state sequence with strong continuity and key information change trend;

[0055] Through the above-mentioned cascade filtering chain, different types of noise (high-frequency random noise, pulse noise, low-frequency drift) are suppressed at the same time, and the key device state change characteristics are maintained, realizing stable time series output;

[0056] Exemplarily, on a surface mount production line, the equipment state perception module outputs health degree data at a frequency of 1 Hz, outputs load rate data at a frequency of 1 Hz, and outputs running state data at a frequency of 2 Hz. After time alignment of the three types of data, a sliding window length w = 5 is set, and mean filtering is performed to obtain preliminary smoothing results. In the median filtering stage, about 2% of the sampling points of the running state data are identified and replaced with adjacent median values to eliminate the peak distortion caused by poor sensor contact. In the variance detection, θ var = 0.05 is set, and about 5% of the windows exceed the threshold, and the fluctuation amplitude is reduced by about 40% through weighted smoothing. In the EWMA smoothing stage, the initial weighting coefficient is set to a = 0.3, and when the state change slows down, it is automatically reduced to a = 0.1. The finally generated filtered equipment state sequence is about 12 dB higher than the original sequence in the signal-to-noise ratio (SNR) index, and retains the characteristics of slow decay of equipment health degree and periodic fluctuation of load rate, providing high credibility input for the standardized normalization calculation of S2.2 step;

[0057] S2.2: Based on the sliding window filtered equipment state sequence, the Z-score standardization method is used to normalize the data in each dimension to eliminate the dimensional differences between different equipment state parameters, and a normalized equipment state feature vector is generated;

[0058] S2.3: The normalized equipment state feature vector is input into the equipment state stability detection module, and the standard deviation and change rate in the time window are calculated to determine whether the equipment state is in a stable running interval. If it exceeds the set threshold, it is marked as state abnormality and triggers the data compensation mechanism;

[0059] S2.4: The linear interpolation compensation algorithm based on historical state data is used to correct the state abnormality of the equipment data to maintain the continuity and credibility of the equipment state vector, and the compensated equipment state feature vector is output;

[0060] S2.5: The compensated equipment state feature vector and the task execution progress information are time stamped and aligned and fused to generate a standardized equipment state-task progress joint vector, which is used as an input parameter of the subsequent multi-objective dynamic priority evaluation function.

[0061] The step S3: based on the current task queue and resource allocation state, a task-resource dependency graph is constructed, and whether there is a conflict situation in which multiple tasks compete for the same equipment or resource is detected. As Fig. 2 shown, specifically comprising:

[0062] S3.1: Obtain the task list in the current task queue and its required resource type information, structurally model the resource demand of the task, and establish a preliminary mapping relationship between the task and the resource;

[0063] S3.2: Based on the real-time device availability data provided by the device state perception module, generate a device resource state matrix to record the current occupation state and the predicted release time of each device as the basic input for resource conflict detection;

[0064] Based on the real-time device availability data provided by the device state perception module, a timestamp matching algorithm is used to synchronize and align the device occupation state data and the task execution progress data (parameters: sampling period matching current device state update frequency);

[0065] Further, by a state classification mapping method (parameters: device running state code mapping table), the real-time running state of the device is discretized into "idle", "occupied", "maintenance" and other category labels to form a structured device state identifier;

[0066] Further, a predicted release time calculation model (parameters: task remaining duration T rem and device operation rate R dev ) is used to calculate the predicted release time of each device according to the formula and bind the time label to the device state record;

[0067] Further, the device resource state matrix is initialized by a two-dimensional array structure (matrix dimension: number of devices x number of state parameters), and the category label, occupation identifier and predicted release time of each device are sequentially filled into the column vector of the corresponding row of the matrix to ensure that the index order of the matrix data corresponds to the device identifier one by one;

[0068] Further, a matrix sparse storage optimization algorithm (parameters: sparsity threshold θ) is applied to store and compress the device resource state matrix, reduce the storage units occupied by idle devices, and improve the memory access efficiency of subsequent conflict detection;

[0069] Through the above processing method, the real-time device availability information of the previous step is converted into a device resource state matrix, realizing the unified modeling and structured output of the current occupation state and the predicted release time of the device, and providing efficient and accurate basic data support for resource conflict detection;

[0070] Exemplarily, in a certain electronic component mounting production workshop, the device state perception module collects the running states of 10 chip mounters at a frequency of 1 Hz, where the state code 0 represents idle, 1 represents task execution, and 2 represents maintenance, the remaining task duration of an occupied task is in seconds, and the work rate is 1 piece per second. After time stamp matching, device #3 shows a state code of 1, a remaining duration of 120 seconds, and a work rate of 1 piece per second, and the predicted release time is calculated as The category label “occupied”, the occupation identifier 1, and the predicted release time 120 are filled into the third row of the matrix, and the columns correspond to [state category, occupation mark, release time] in turn. After sparse storage processing of the matrix, the memory occupied by the idle device row is reduced by 60%, providing high-density effective data input for the subsequent conflict detection acceleration module;

[0071] S3.3: Using the directed graph modeling method in graph theory, the tasks are regarded as nodes in the graph, and the resource request relationship between the tasks and the devices is regarded as edges, a task-resource dependency graph is constructed, and the interaction dependency between the tasks and the resources is presented in a visual manner;

[0072] After receiving the preliminary mapping relationship between the tasks and the resources established by S3.1 and the device resource state matrix generated by S3.2, each task in the task list is defined as a task node in the directed graph, and each device in the device list is defined as a resource node in the directed graph, to form a double-type node set of the tasks and the devices;

[0073] Using the directed graph modeling method (parameters: node set V = task node set ∪ resource node set, edge set E is defined according to task resource demand), bidirectional modeling of the request relationship from the tasks to the resources and the occupation relationship from the resources to the tasks is realized, the directed edge from the task node to the required resource node is marked as a request edge, and the directed edge from the resource node to the task node being occupied by the resource node is marked as an allocation edge;

[0074] Further, through a node attribute appending algorithm (parameters: task attributes include task ID, remaining duration, and priority; resource attributes include device ID, current state code, and predicted release time), key attributes from the task queue and the device resource state matrix are bound in the node structure of the graph, to support the attribute filtering function in the subsequent conflict detection;

[0075] Further, an adjacency matrix construction method (parameters: the matrix dimension is |V|×|V|, and the element a i,j The value 1 indicates the existence of a directed edge, and 0 indicates no edge), an adjacency matrix expression form of the task-resource dependency is generated, and efficient operation support for the graph theory algorithm is realized;

[0076] Further, the task nodes and resource nodes are positioned in a two-dimensional plane or a three-dimensional space by a graph visualization layout algorithm (parameters: force-directed layout iteration number, node repulsion coefficient, edge length constraint), to generate a visual dependency topology graph, which helps to intuitively present the interactive dependency structure between multiple tasks and multiple devices;

[0077] By the above modeling and visualization processing, the real-time dependency relationship between tasks and resources is converted into a task-resource dependency graph with computability and observability, which provides a structured input and topological constraint condition for the deadlock detection of S3.4 and the resource conflict detection of S3.5;

[0078] For example, in an electronic component production scenario including 5 tasks (T1 to T5) and 3 devices (D1 to D3), tasks T1 and T2 request device D1, task T3 requests device D2, task T4 requests device D3 which is currently occupied by D2, and task T5 requests devices D1 and D3. The node set V includes 8 nodes, the task node set = {T1, T2, T3, T4, T5}, and the device node set = {D1, D2, D3}. The request edges are, for example, T1→D1, T2→D1, T5→D1, and T5→D3, and the allocation edge is, for example, D2→T4. In the adjacency matrix, a 1,6 =1 indicates that the request relationship from T1 to D1 is established. The force-directed layout algorithm sets the iteration number to 500 steps, the node repulsion coefficient to 200, and the edge length constraint to 100 units of length. In the final graph layout, D1 and T1, T2, and T5 form a high-density star structure, which intuitively indicates the potential conflict risk of multiple tasks competing for the same device, providing a clear structural basis for subsequent conflict detection. Since the data input of this process is bound to the task remaining duration and the device expected release time, the system can prepare attribute filtering conditions when the graph is constructed, so that the loop detection of S3.4 and the conflict statistics of S3.5 are completed on a shared structural model, effectively improving the calculation efficiency;

[0079] S3.4: Based on the task-resource dependency graph, a topological sorting algorithm is used to detect whether there is a loop structure in the graph. If there is a loop, it is determined that there is a task scheduling deadlock risk, and a deadlock resolution mechanism is triggered and a high-priority conflict task is marked;

[0080] Based on the task-resource dependency graph and its adjacency matrix expression constructed by S3.3, the loop risk analysis and detection of task scheduling deadlock are realized;

[0081] A topological sorting algorithm based on depth-first search (DFS) (parameters: use visited set and recursion stack set recStack) is used to realize the ordered traversal of the dependency graph and the generation of the topological sequence of the nodes with in-degree zero, to preliminarily determine whether there is a feasible scheduling sequence.

[0082] Further, the iterative updating process of the topological sorting is implemented by the in-degree statistics and the node dequeuing operation (parameters: the initial queue structure Q contains all nodes with zero in-degree, and the in-degree array inDegree is derived from the column-by-column cumulative calculation of the adjacency matrix), and the sorting result length lenTopo is obtained;

[0083] Further, the length determination method is used to compare lenTopo with the total number of nodes |V|, when |V|-lenTopo is greater than 0, it is judged that there is a loop structure, and it is marked as a task scheduling deadlock risk; when lenTopo is equal to |V|, it means that there is no loop;

[0084] Further, for the detected loop node, the loop reverse tracking algorithm (parameters: DFS recursive stack path information is used) is adopted to backtrack the task node sequence constituting the loop, generate the deadlock-involved task set, and add a high-priority conflict label in the task attribute;

[0085] Further, the deadlock resolution mechanism is triggered (parameters: conflict resolution strategy based on task priority and device release time), and the resource occupation resolution or priority temporary reduction processing is performed on the deadlock-involved task set, and a feasible task dependency topology is generated, which is ready for S3.5 to perform resource conflict event statistics;

[0086] Through the above topological sorting and loop detection processing, the unschedulable structure in the task-resource dependency graph is converted into identifiable deadlock risk data, the high-priority label of the deadlock task is realized, and explicit risk input is provided for subsequent dynamic priority reordering;

[0087] For example, in a dependency graph containing task nodes T1 to T4 and device nodes D1 and D2, there are directed edge sets T1→D1, D1→T2, T2→D2, D2→T1, which form a loop T1-D1-T2-D2-T1. The in-degree statistics of the adjacency matrix obtains the initial in-degree array [1, 1, 0, 1, 0, 1], and the initial queue Q only contains the node D2 with zero in-degree. During the topological sorting process, lenTopo=4, the total number of nodes |V|=6, and |V|-lenTopo=2>0, it is determined that there is a deadlock. The loop task set {T1, T2} is obtained by using DFS backtracking, a high-priority conflict label is added to it, and the deadlock resolution mechanism is called to delay the execution of the resource request of T2, so that the topological sorting generates a loop-free sequence [T3, D2, T2, D1, T1, D3], and the deadlock risk is resolved and the feasibility of subsequent scheduling optimization is ensured; Topo

[0088] ​S3.5: Perform resource conflict detection algorithm on the task-resource dependency graph, traverse all device nodes, count the number of times the same device is requested by multiple tasks within the same time window, if it exceeds the device concurrent processing capacity, it is determined as a resource conflict event and the conflict task set is recorded.

[0089] The step S4: if resource conflict is detected, a multi-objective dynamic priority evaluation function is constructed according to the standardized device state vector, the task remaining duration and the priority original weight. As shown in the formula (1), it specifically includes: Fig. 3

[0090] S4.1: Based on the standardized device state vector output by the device state perception module, the current device health degree and load rate are extracted as the key influence factors of dynamic priority evaluation, to reflect the real-time availability of the device;

[0091] Based on the conflict task set passed from the task scheduling optimization module in step S3 and the standardized device state vector output by the device state perception module, the key parameters reflecting the real-time availability of the device are selected as the core input of dynamic priority evaluation;

[0092] The parameter screening algorithm (parameters: feature importance threshold τ h and τ l ) is adopted to calculate the correlation coefficient of each feature to the conflict task execution efficiency and delay risk in all feature dimensions of the standardized device state vector, and the health degree feature and the load rate feature are selected as the main influence factors for determining the availability of the device;

[0093] Further, through the feature reverse normalization algorithm (parameters: normalization center μ, standard deviation σ), the standardized health degree and load rate are mapped back to the physical quantity interval, which is used to combine with the device running safety threshold for state validity judgment, to ensure the authenticity of the subsequent priority evaluation input;

[0094] Further, the outlier correction method (parameters: health degree threshold H min , load rate threshold L max ) is adopted to perform interval clipping processing on the mapping result, when the health degree is less than H min , it is set to H min , and when the load rate is greater than L max , it is set to L max , to eliminate the disturbance of extreme values on the evaluation model;

[0095] Further, a device availability comprehensive factor calculation model (parameters: weight coefficients α, β) is constructed, and the device availability comprehensive index is calculated by the following formula:

[0096]

[0097] Wherein, H​norm and L norm are the normalized health and normalized load rate, respectively;

[0098] Further, the equipment availability comprehensive index I avail is stored in association with the resource requirement matching degree parameter of the conflict task, forming an equipment availability input vector for the S4.4 multi-objective dynamic priority evaluation function;

[0099] Through the above method, the redundant features in the normalized equipment state vector are reduced to two core indicators of health and load rate, and the availability input vector that comprehensively reflects the real-time availability of the equipment is calculated, realizing the accurate response of priority evaluation to the real-time running state of the equipment;

[0100] For example, in an electronic component mounting workshop, in the normalized equipment state vector, the health parameter of equipment #5 is 0.82, and the load rate parameter is 0.65. After inverse normalization (μ H = 75, σ H = 10, μ L = 50%, σ L = 20%), the health is 83 points and the load rate is 63%. The threshold clipping (H min = 60 points, L max = 85%) does not trigger the correction. Set the weights α = 0.6 and β = 0.4, and input the formula to calculate the availability comprehensive index:

[0101] , which means that the equipment is in a higher availability level. In the S4.4 evaluation function, this index will be used as a key factor of equipment availability, increasing the probability of equipment with high health and low load to undertake priority tasks, and ultimately reducing the task delay rate by about 15%;

[0102] S4.2: Normalize the remaining duration of each task in the task queue to eliminate the dimensional influence brought by the differences in task type and process complexity, and generate a normalized task remaining duration parameter;

[0103] Based on the equipment availability input vector from S4.1 and the remaining duration raw data of the conflict task set, the minimum-maximum normalization method (parameters: global minimum remaining duration T min and global maximum remaining duration T max ) is used to realize the scale unification of the task remaining duration in the [0, 1] interval;

[0104] Further, through the normalization calculation formula

[0105]

[0106] to realize the dimensionless of each task remaining duration, where Trem remaining duration for the current task;

[0107] Further, the impact of different process complexity or task category on the remaining duration is taken into account by the task type weight correction algorithm (parameter: weight coefficient vector W type ), to generate the corrected normalized remaining duration T norm' ;

[0108] Further, the robust scale adjustment method (parameters: median M and absolute median deviation MAD t ) is used to suppress outliers in the corrected normalized remaining duration, to reduce the disturbance of extremely short or long duration tasks on the evaluation model;

[0109] Further, the processed normalized task remaining duration parameters are indexed and corresponded to the equipment availability input vector to form a multi-dimensional joint data structure containing task ID, normalized remaining duration and associated equipment state, which is used as the task urgency input of the S4.4 multi-objective dynamic priority evaluation function;

[0110] Through the above normalization and correction processing, the original task remaining duration data is converted into normalized task remaining duration parameters with comparability, stability and process independence, realizing the precise quantification of task time urgency for subsequent priority evaluation;

[0111] For example, in an electronic component mounting workshop, the conflict task set includes T7, T8, T9, with corresponding remaining durations of 180 seconds, 420 seconds and 60 seconds respectively, and T min = 60 seconds, T max = 420 seconds. The minimum-maximum normalization calculation gives T7:

[0112] T8: T9: The task category weight coefficients are T7 = 1.0, T8 = 0.8, T9 = 1.2, and after correction, T7 = 0.333 x 1.0 = 0.333, T8 = 1.000 x 0.8 = 0.800, T9 = 0.000 x 1.2 = 0.000. In the outlier suppression step, the median M = 0.333, the absolute median deviation 0.333, no outliers exceeding 3 x MAD are found. The final generated normalized task remaining duration parameter vector is [0.333, 0.800, 0.000], and it is bound with the availability input vector of equipment #2, equipment #5 and equipment #7, to provide high-precision task time urgency input for multi-objective priority evaluation;

[0113] S4.3: Based on the task original priority weight table, the basic priority value of each task is obtained as the initial reference value for dynamic priority evaluation, to maintain the business logic consistency of the task scheduling strategy;

[0114] Based on the standardized task remaining duration parameters and task metadata table output from the task scheduling optimization module in step S4.2, the task original priority weight table index algorithm is called (parameters: task ID mapping table, weight lookup hash table), which realizes high-speed retrieval from task ID to priority weight;

[0115] Further, through the weight validity verification algorithm (parameters: weight value domain [W min ,W max ]), interval checking is performed on the retrieved priority weight, and the weight value exceeding the set interval is clipped to the boundary value, and the clipping event is recorded to the weight correction log, to ensure that the basic priority data input into the evaluation model does not occur logical exception;

[0116] Further, the data type standardization method (parameters: target type float64, precision 6 decimal places) is used to realize the data type unification and precision consistency control of all basic priority values, to prevent type mismatch or precision loss in weight operation of subsequent multi-objective evaluation functions;

[0117] Further, the priority normalization algorithm (parameters: global minimum weight W min and global maximum weight W max ) is used to map the basic priority to the interval [0,1], and the normalization formula is:

[0118]

[0119] Wherein, W is the task original priority weight;

[0120] Further, a bidirectional mapping relationship table of task ID and normalized priority is established, which is used as the business priority input vector of the S4.4 multi-objective dynamic priority evaluation function;

[0121] Through the above mapping, verification, standardization and normalization processing methods, the original priority weight of the task is converted into a basic priority reference value that can be directly called by the evaluation function, which is logically consistent and numerically stable, realizing the business logic consistency while ensuring the computability of the evaluation input;

[0122] For example, in a conflict task set of an electronic component mounting workshop, the original priority weights of tasks T12, T13 and T14 are 85, 120 and 95 respectively, and the allowed interval [W min ,W max ] of the weight table is set as [60, 150].] for [50, 100], in the weight validity verification link, the weight 120 of T13 is clipped to 100, and the clipping log is recorded. Set the global minimum weight W min = 50, the global maximum weight W max = 100, the normalization calculation obtains: the T13's T14's After type unification and precision control processing, the generated priority input vector is [0.700000, 1.000000, 0.900000], and a bidirectional mapping table is formed with the task ID, which provides stable service priority input for the multi-objective priority evaluation function of S4.4, ensuring that the scheduling strategy can maintain the consistency of the business logic sequence of task processing in the dynamic adjustment process;

[0123] S4.4: Construct a multi-objective dynamic priority evaluation function, which includes a standardized device state vector, a standardized task remaining duration parameter and a task original priority weight, and the function form adopts a combination of weighted linear combination and nonlinear decay factor to balance the relationship between device health, task urgency and business priority;

[0124] S4.5: Parameter initialization and weight distribution are performed on the multi-objective dynamic priority evaluation function, and initial weight coefficients are set based on historical scheduling data and expert experience to provide adjustable mathematical expressions for subsequent task priority score calculation.

[0125] The step S5: input the device health and load rate as feedback factors into the dynamic priority evaluation function to calculate the real-time priority score of each conflicting task. Specifically, it includes:

[0126] S5.1: Obtain the device health parameter in the standardized device state vector, and perform nonlinear mapping on the health degree based on the exponential decay function to generate a device availability decay factor, which is used to quantify the influence of device failure risk on task scheduling;

[0127] S5.2: Obtain the device load rate parameter in the standardized device state vector, and perform segmented weighting processing on the load rate using a load sensitivity function to generate a device load sensitivity factor, which is used to reflect the degree of restriction of the current load state of the device on the task execution efficiency;

[0128] S5.3: Based on the device availability decay factor and the device load sensitivity factor, a task priority feedback adjustment model is constructed by combining the task original priority weight to generate a task priority dynamic adjustment coefficient, which is used to quantify the comprehensive influence of device state on task priority;

[0129] Based on the device availability attenuation factor and the device load sensitivity factor, combined with the task original priority weight input, a priority feedback adjustment coefficient construction method (parameters: attenuation factor F avail , sensitivity factor F load , normalized priority W norm ) is adopted to realize multi-dimensional state comprehensive correction of the priority of each conflict task.

[0130] Further, a multi-factor weighted combination algorithm (parameters: weight coefficients γ1, γ2, γ3) is used to construct a task priority feedback adjustment formula:

[0131] K adj = γ1F avail + γ2(1-F load )+ γ3W norm

[0132] Wherein, K adj is a dynamic adjustment coefficient for correcting the task priority score.

[0133] Further, a normalization mapping algorithm (parameters: global minimum K min , global maximum K max ) is used to map the dynamic adjustment coefficient to the interval [0, 1] to ensure that the adjustment result can be directly input into the priority score calculation model under a unified scale. The normalization formula is:

[0134]

[0135] Further, a smoothing filtering algorithm (parameters: smoothing window w, smoothing factor μ) is used to smooth the time series of the continuously calculated adjustment coefficient, eliminate the sudden effect caused by the short-term fluctuation of the device state, and generate a stable task priority dynamic adjustment coefficient matrix.

[0136] Further, the stable adjustment coefficient is bound one by one with each conflict task ID, and is sorted according to the time stamp to form a task priority adjustment factor library for subsequent S5.4 weighted Euclidean distance priority score calculation direct call.

[0137] Through multi-factor weighted combination, normalization mapping and time series smoothing processing, the device real-time state and business priority data are converted into dynamic adjustment coefficients under a unified scale, realizing fine and adaptive correction of the task priority.

[0138] Exemplary, in an electronic component mounting workshop, the conflict task set {T21, T22, T23} corresponds to the equipment health degradation factor respectively 0.88, 0.65, 0.92, the equipment load sensitivity factor respectively 0.40, 0.75, 0.55, and the normalized business priority is 0.70, 0.85, 0.60. Set the weight coefficients γ1=0.5, γ2=0.3, γ3=0.2, substitute into the formula to calculate the dynamic adjustment coefficient of T21 as K adj , set the global K min =0.60, K max =0.85, normalize to obtain K adj' , after sliding smoothing processing of w=3, μ=0.5, the stable adjustment coefficient is 0.642, and finally T21 is bound as the priority weight correction input. In the execution verification, the coefficient makes the scheduling probability of T21 on the equipment with high health and low load increase by about 14%, and the overall conflict task delay rate decreases by 9.8%;

[0139] S5.4: Obtain the task remaining duration information, combine the task priority dynamic adjustment coefficient, calculate the task priority score using the weighted Euclidean distance algorithm, generate a task priority score vector, and use it to represent the priority execution degree of each conflict task at the current scheduling time;

[0140] S5.5: Normalize the task priority score vector, convert it to a task execution probability distribution using the Softmax function, generate a task execution probability vector, and use it as the input basis for the subsequent task sorting and scheduling decision module.

[0141] The step S6: based on the priority score, sort the conflict tasks, generate an optimized task execution sequence, and issue the updated task scheduling scheme to the collaborative control execution module. Specifically, it includes:

[0142] S6.1: Normalize the real-time priority score of the conflict task to eliminate the priority score deviation between different task types, obtain a normalized priority score vector, and use it for subsequent sorting calculation;

[0143] S6.2: Based on the normalized priority score vector, use a multi-objective sorting algorithm to sort the priority of the conflict task, combine the task remaining duration and the equipment load rate, and generate a task priority sorting sequence;

[0144] Based on the normalized priority score vector, construct the sorting calculation input of the conflict task, use a multi-objective sorting algorithm (parameters: task priority score vector V norm , standardized task remaining duration vector T norm , standardized equipment load rate vector L norm ), realize the comprehensive sorting score generation of multi-dimensional task attributes;

[0145] Further, the comprehensive ranking score value of each task is calculated according to the following formula by a multi-criteria decision weighting model (parameters: target weight coefficients α1, α2, α3):

[0146] S total = α1V norm + α2(1-T norm ) + α3(1-L norm )

[0147] Wherein, 1-T norm is used to represent that the shorter the duration, the higher the score, and 1-L norm is used to represent that the lower the load rate, the higher the score;

[0148] Further, the weight parameter normalization processing algorithm is adopted to normalize α1, α2, and α3 to ensure α1+α2+α3=1, so that each target participates in the comprehensive score calculation under the unified dimension;

[0149] Further, based on the comprehensive ranking score value, a stable sorting algorithm (parameter: sorting stability flag β) is adopted to arrange the task set in descending order, so as to ensure the relative order of the original task queue when the scores are equal, and generate a task priority sorting sequence;

[0150] By the multi-target weighting and stable sorting algorithm, the priority score, remaining duration, and equipment load rate are mapped into an executable task priority order, so as to realize the optimization sorting of conflict tasks;

[0151] For example, in the electronic component mounting area, the input data of the conflict task set {T31, T32, T33} is: V norm ={0.82, 0.76, 0.65}, T norm ={0.40, 0.55, 0.30}, L norm ={0.65, 0.80, 0.50}, and α1=0.5, α2=0.3, α3=0.2, and the normalization keeps the total sum as 1;

[0152] Substituting the formula, the comprehensive score of T31 is

[0153] S total =0.5×0.82+0.3×(1-0.40)+0.2×(1-0.65)=0.41+0.18+0.07=0.66, the comprehensive score of T32 is

[0154] =0.5×0.76+0.3×0.45+0.2×0.20=0.38+0.135+0.04=0.555, and the comprehensive score of T33 is

[0155] = 0.5 x 0.65 + 0.3 x 0.70 + 0.2 x 0.50 = 0.325 + 0.21 + 0.10 = 0.635. The ranking result is T31(0.66) > T33(0.635) > T32(0.555). Through the ranking, the probability of T31 executing on a high-priority available device is increased by 15.4%, and the average waiting time of task conflict is reduced by 11.2%;

[0156] T33(0.635) > T32(0.555). Through the ranking, the probability of T31 executing on a high-priority available device is increased by 15.4%, and the average waiting time of task conflict is reduced by 11.2%;

[0157] S6.3: Match the task priority ranking sequence with the current device availability state, build a task-device mapping relationship table, and ensure that the task scheduling scheme matches the actual running capability of the device;

[0158] S6.4: Based on the task-device mapping relationship table, generate an optimized task execution sequence, and combine the device startup delay and task switching cost to dynamically adjust the scheduling time window to improve scheduling stability;

[0159] S6.5: Encapsulate the optimized task execution sequence into a structured scheduling instruction, and issue it to the collaborative control execution module through an industrial communication protocol to drive the device to execute the task according to the new scheduling sequence.

[0160] The step S7: continuously monitor the device state and task progress changes during task execution, and re-trigger the task priority reordering mechanism when the device state mutation or task progress deviation exceeds the set threshold. Specifically, it includes:

[0161] S7.1: Perform sliding window filtering processing on the health degree, load rate and running state data collected by the device state perception module in real time to eliminate transient noise interference, and generate a smooth device state sequence as the input basis for device state change detection;

[0162] S7.2: Based on the smooth device state sequence, calculate the device health degree change rate and load rate change slope, and use the statistical process control (SPC) method to identify the device state mutation point. When the change rate exceeds the set control limit threshold, a device state mutation event signal is generated;

[0163] S7.3: Compare the task expected completion time and actual progress timestamp of the task execution progress information, calculate the task progress deviation, and use the moving average method to evaluate the task execution delay trend. When the deviation exceeds the preset deviation threshold, a task progress abnormal signal is generated;

[0164] The collected task execution progress information is matched by comparing the task timestamps (parameters: task expected completion time T plan , task actual execution timestamp T act ), to realize the corresponding matching of the expected progress and the actual progress of the task;

[0165] Further, the task progress offset Δ is calculated by a time difference calculation model (formula as follows) prog , and the basic data reflecting the progress deviation degree is generated:

[0166] Δ prog = T act - T plan

[0167] Wherein, T act is the actual time of task execution to the current time, T plan is the expected completion time of the task in the scheduling plan;

[0168] Further, the sliding moving average method (parameters: time window length W, sliding step s) is used to smooth the task progress offset sequence continuously collected, to realize the denoising and stable capture of the delay trend, and generate the smooth progress offset sequence Δ prog,avg :

[0169]

[0170] Wherein, Δ prog(t-i) is the task progress offset value in the ith sliding window;

[0171] Further, the delay trend analysis algorithm (parameters: delay change rate threshold θ delay ) is used to calculate the change slope k delay of the smooth progress offset sequence on the time axis, to realize the quantitative description of the delay trend;

[0172] Further, the threshold comparison strategy is used to compare the smooth offset value Δ prog,avg with the preset task progress abnormal threshold θ prog , when Δ prog,avg ≥ θ prog , the task progress abnormal signal is output;

[0173] Through the above method, the task time offset is converted into stable delay trend and abnormal signal, to realize the early identification and quantitative triggering of the production task delay risk;

[0174] For example, in the electronic component mounting station, the expected completion time of task number T45 is 120 seconds, and the collected actual time stamp is 135 seconds, the task progress offset Δ prog = 15 seconds is obtained by substituting the formula. In the moving average calculation, the window length W = 5, the step s = 1 second, the offset sequence {14, 15, 16, 14, 17} is smoothed, and the smooth progress offset sequence Δ is obtained. When the preset abnormal threshold θprog = 10 seconds comparison, greater than threshold, trigger task progress abnormal signal. In this scenario, the scheduling system will be based on abnormal signal instantaneously evaluate task priority and adjust the execution order, T45 ahead of schedule to low load device execution, effectively reduce the risk of delay, and in subsequent statistics to verify the average delay shortened 18%;

[0175] S7.4: Based on the device state mutation event signal and the task progress abnormal signal, a dynamic rescheduling trigger logic function is constructed, when any signal triggers, output the reordering trigger instruction to the task scheduling optimization module, to start the task priority reordering process;

[0176] S7.5: After the reordering mechanism is triggered, the current device state vector, task progress offset and historical scheduling feedback data are input into the scheduling model optimization module, which is used to update the feedback factor weight in the multi-objective dynamic priority evaluation function, to realize the online adaptive adjustment of the scheduling strategy.

[0177] The step S8: record the execution time of each priority reordering, scheduling result and execution feedback, for subsequent online learning and parameter optimization of scheduling model. Specifically includes:

[0178] S8.1: Time stamp record is performed on the trigger time point of each priority reordering event, to generate time sequence identification of scheduling event, for subsequent time sequence analysis and model training of scheduling behavior;

[0179] S8.2: Based on the task execution sequence and resource allocation scheme output by the scheduling optimization module, a scheduling result data set is generated, the scheduling result including task priority score, execution order, device allocation and expected completion time, to build a complete scheduling decision sample;

[0180] S8.3: In the task execution process, the actual execution data fed back by the cooperative control execution module is collected, including task start time, completion time, device state change and abnormal event, to generate a scheduling execution feedback data set;

[0181] In the task execution process, the real-time data stream of the cooperative control execution module is taken as the input condition, which includes task execution state reporting, device running parameter and abnormal event monitoring information;

[0182] The high-precision time synchronization protocol (parameters: accuracy ≤ 1 millisecond) is used to align the task start time and completion time output by the cooperative control execution module with the time stamp, to realize the unified time reference of different device feedback data;

[0183] Further, real-time capture task state change events through event-driven data acquisition mechanism (parameters: event trigger frequency 10Hz, cache depth 1024 records) to extract task number, task state identifier and corresponding timestamp, and generate original task time series data;

[0184] Further, parse key state indicators of the device using a running parameter parsing algorithm (parameters: sampling period 100ms), including spindle speed, current, voltage, temperature, etc., associate the instantaneous state vector with the task execution phase, and form a task-device state mapping table;

[0185] Further, compare running parameters with preset normal intervals through an abnormal event detection module (algorithm: based on statistical threshold and rule matching), and when a state point exceeding the control limit value or an abnormal pattern code is matched, generate an abnormal event record, and attach abnormal type, affected device and time information;

[0186] Through a data fusion module, associate and aggregate task time series data, task-device state mapping table and abnormal event records according to task number, construct a scheduling execution feedback dataset containing task execution information throughout the process, and realize the structuring and retrievability of feedback data;

[0187] For example, on an electronic component mounting station, the task start time of task number T58 is collected as 10:23:15.005, the completion time is 10:23:45.215, the timestamp synchronization accuracy is set to 1ms, and the normalized execution length is 30.210s. The spindle speed vector parsed by the running parameter parsing acquisition has a mean value of 1500rpm and a standard deviation of 5.6rpm during execution, and the temperature parameter has a peak value of 80.5℃ at the 18th second, exceeding the preset threshold of 80℃ to trigger a high temperature alarm event. The abnormal event detection module records the event type as "high temperature", marks the affected device as D12, and the occurrence time is 10:23:33.105. The data fusion module aggregates the task time series data, running parameter data and high temperature abnormal event record to generate a scheduling execution feedback dataset containing task execution information throughout the process and abnormal markers, providing comprehensive basic data for subsequent scheduling model performance evaluation and weight dynamic optimization;

[0188] S8.4: Structured storage of scheduling event timestamp, scheduling result dataset and execution feedback dataset to generate a scheduling process holographic log record to support the construction of scheduling model training data and the calculation of scheduling performance evaluation indicators;

[0189] S8.5: Based on the scheduling process holographic logging, the online incremental learning algorithm of the scheduling model is executed to update and optimize the parameter weights in the dynamic priority evaluation function in real time, so as to improve the adaptability and stability of the scheduling strategy in the complex dynamic environment.

[0190] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

[0191] The above description is only the preferred embodiments of the present application and is not used to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and rules of the present application shall be included in the protection scope of the present application.

Claims

1. An unmanned plant equipment collaborative intelligent optimization method, characterized in that, The method comprises the following steps: S1: deploying a device state perception module in the unmanned workshop, collecting health degree, load rate and running state data of each production device in real time, and recording task execution progress information; S2: filtering and normalizing the collected device state data to generate a standardized device state vector; S3: based on the current task queue and resource allocation state, a task-resource dependency graph is constructed, and whether there is a conflict situation in which multiple tasks compete for the same device or resource is detected; S4: if a resource conflict is detected, a multi-objective dynamic priority evaluation function is constructed according to the standardized device state vector, the task remaining duration and the priority original weight; S5: inputting the device health degree and load rate as feedback factors into the multi-objective dynamic priority evaluation function to calculate the real-time priority score of each conflict task; S6: based on the priority score, the conflict tasks are sorted to generate an optimized task execution sequence, and the updated task scheduling scheme is issued to the collaborative control execution module; S7: continuously monitor the changes of device state and task progress during task execution, and when detecting that the device state mutation or task progress deviation exceeds the set threshold, the task priority reordering mechanism is triggered again; S8: record the execution time, scheduling result and execution feedback of each priority reordering for online learning and parameter optimization of subsequent scheduling model.

2. The method of claim 1, wherein, The step S1 specifically comprises: Deploy a multi-modal sensor array at the key control node of the electronic component production device, synchronously collect device running voltage, current, temperature and vibration signals, and obtain the basic physical parameters of the device running state; Based on the collected voltage and current signals, the fast Fourier transform algorithm is used to extract the device running power spectrum features to generate a device load rate feature vector; The wavelet packet transform algorithm is applied to time-frequency domain decomposition of temperature and vibration signals to extract energy entropy and kurtosis indicators, and a device health degree evaluation model is established combined with historical fault data to calculate the device health degree index; The device load rate feature vector and the device health degree index are transmitted to the edge computing node through the industrial Ethernet, and are timestamped with the current task execution state information of the device to generate a structured device state data packet.

3. The method of claim 2, wherein, In the step S1, the device state perception module comprises a multi-modal sensor array, the sensor array comprises a high-precision voltage sensor, a current sensor, a thermocouple temperature sensor and a three-axis MEMS vibration sensor, each modal signal is synchronously collected and synchronized through a hardware clock, the sampling frequency is set to 10-20 times the highest frequency of the harmonic frequency, the voltage and current signals are 1kHz, the temperature signal is 1Hz, and the vibration signal is 5kHz.

4. The method of claim 1, wherein, The step S2 specifically comprises: Performing sliding window filtering processing on the original device health degree data, load rate data and running state data collected by the device state perception module to obtain a filtered device state sequence; Based on the filtered device state sequence, the Z-score standardization method is used to normalize each dimension of data to generate a normalized device state feature vector; The normalized equipment state feature vector is input into an equipment state stability detection module, the standard deviation and the change rate thereof in a time window are calculated, whether the equipment state is in a stable operation interval is judged, and if the set threshold is exceeded, the state is marked as abnormal and a data compensation mechanism is triggered; The equipment data with an abnormal state is corrected by using a linear interpolation compensation algorithm based on historical state data, and a compensated equipment state feature vector is output; The compensated equipment state feature vector is timestamped, aligned and fused with task execution progress information, and a standardized equipment state-task progress joint vector is generated.

5. The method of claim 1, wherein, The step S3 specifically comprises: A task list in a current task queue and resource type information required by the task list are acquired, resource requirements of the task are structured modeled, and a preliminary mapping relationship between the task and the resource is established; Based on real-time equipment availability data provided by an equipment state perception module, an equipment resource state matrix is generated, and current occupation states and predicted release times of each equipment are recorded as basic input for resource conflict detection; A directed graph modeling method in graph theory is used, the task is taken as a node in the graph, and a resource request relationship between the task and the equipment is taken as an edge, and a task-resource dependency graph is constructed; Based on the task-resource dependency graph, a topological sorting algorithm is used to detect whether there is a loop structure in the graph, if there is a loop, it is determined that there is a task scheduling deadlock risk, a deadlock resolution mechanism is triggered, and a high-priority conflict task is marked; A resource conflict detection algorithm is executed on the task-resource dependency graph, all equipment nodes are traversed, the number of times that a same equipment is requested by multiple tasks in a same time window is counted, and if the number of times exceeds equipment concurrent processing capability, it is determined that there is a resource conflict event and a conflict task set is recorded.

6. The method of claim 5, wherein, In the step S3, the task-resource dependency graph is modeled in a graph theory directed graph form, the task node and the equipment node are bidirectionally connected through a resource request edge and an allocation edge, and node attributes include a task ID, a remaining duration, a priority and an equipment state code, and a release time.

7. The method of claim 1, wherein, The step S4 specifically comprises: Based on a standardized equipment state vector output by an equipment state perception module, a current equipment health degree and a load rate are extracted as key influence factors for dynamic priority evaluation; A remaining duration of each task in a task queue is normalized to generate a standardized task remaining duration parameter; Based on a task original priority weight table, a basic priority value of each task is acquired as an initial reference value for dynamic priority evaluation; A multi-objective dynamic priority evaluation function is constructed, and a function form adopts a weighted linear combination combined with a nonlinear decay factor; Parameter initialization and weight distribution are performed on the multi-objective dynamic priority evaluation function, and initial weight coefficients are set based on historical scheduling data and expert experience. 8.The method of claim 7, wherein, In the step S4, the task-resource dependency graph is modeled in a graph theory directed graph form, the task node and the equipment node are bidirectionally connected through a resource request edge and an allocation edge, and node attributes include a task ID, a remaining duration, a priority and an equipment state code, and a release time. 9.The method of claim 7, wherein, In the step S4, the input of the multi-target dynamic priority evaluation function includes the normalized device health, the normalized load rate, the normalized task remaining duration, and the normalized task priority weight, wherein the device health and the load rate participate in the weighted linear combination according to the weight parameters, and the priority score vector is generated by using the weighted superposition and the nonlinear attenuation in combination with the duration and the priority parameters.

10. The method of claim 1, wherein, The step S5 specifically includes: obtaining the device health parameter in the normalized device state vector, performing nonlinear mapping on the health by using an exponential attenuation function to generate a device availability attenuation factor; obtaining the device load rate parameter in the normalized device state vector, performing segmented weighting processing on the load rate by using a load sensitivity function to generate a device load sensitivity factor; constructing a task priority feedback adjustment model based on the device availability attenuation factor and the device load sensitivity factor in combination with the original task priority weight to generate a task priority dynamic adjustment coefficient; obtaining task remaining duration information, combining the task priority dynamic adjustment coefficient, and calculating a task priority score by using a weighted Euclidean distance algorithm to generate a task priority score vector; performing normalization processing on the task priority score vector, converting it into a task execution probability distribution by using a Softmax function to generate a task execution probability vector.

Images