Dynamic equipment scheduling method and device based on cultural event perception and equipment

Abstract

The invention relates to the technical field of equipment dynamic scheduling based on cultural event awareness, and discloses an equipment dynamic scheduling method, device and equipment based on cultural event awareness, and the method comprises the steps: constructing a cultural event feature vector, and carrying out prediction through employing a large-scale language model, thereby improving the timeliness and accuracy of employment gap prediction, and improving the efficiency of employment gap prediction. And dynamic optimization is carried out in combination with an integer programming model, so that real-time and optimal adjustment of the starting proportion of the target equipment is realized. The method has the beneficial effects that the operation interruption risk and extra manpower cost caused by sudden shortage of manpower are remarkably reduced, the cultural adaptability and the overall elasticity of the scheduling system are enhanced, and the stability of the operation efficiency and the efficient utilization of resources in special periods such as cultural events are ensured; and a solution is provided for solving a cross-culture region operation management problem.

Description

Technical Field

[0001] This invention relates to the field of dynamic equipment scheduling technology based on cultural event perception, and in particular to a method, apparatus and equipment for dynamic equipment scheduling based on cultural event perception. Background Technology

[0002] In cross-border trade, cultural events in certain regions can cause drastic and sudden fluctuations in labor attendance, posing a severe challenge to the continuous operation of industries such as warehousing and manufacturing. Existing scheduling systems are mostly based on fixed historical rules, making it difficult to predict and respond to such culturally sensitive labor shortages. At the same time, some equipment (such as automated guided vehicles) is usually configured in a fixed ratio, lacking flexibility. During periods of labor shortages, equipment may be idle or overloaded, resulting in resource waste and decreased efficiency. Furthermore, traditional analysis methods struggle to effectively integrate and uncover the deep connections between various channels of information on cultural events affecting labor, leading to insufficient decision-making basis. Summary of the Invention

[0003] Therefore, it is necessary to propose a method, device, and equipment for dynamic equipment scheduling based on cultural event awareness, in response to the existing problem of dynamic equipment scheduling based on cultural event awareness.

[0004] A method for dynamic device scheduling based on cultural event awareness, the method comprising: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

[0005] Furthermore, the step of extracting and fusing features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors includes: Semantic features are extracted from the multi-source heterogeneous data using a multilingual pre-trained language model. The structured features of cultural events in the specified target area are obtained, and the semantic features are fused with the structured features to obtain fused features; wherein, the structured features include the time information of cultural events in the specified target area and the population density distribution information of the content in the specified target area; The fused features are vectorized and encoded to generate a fixed-dimensional feature vector of the cultural event.

[0006] Furthermore, before the step of inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain the target programming model, the method further includes: Obtain device information for each target device in the specified target area; Based on the total number of target devices in the device information, a decision variable is defined to indicate whether each target device will be enabled within a preset future time period; Based on each decision variable, an objective function is constructed with the goal of minimizing the total operating cost; wherein the total operating cost consists of the equipment operating cost and the labor cost; The integer programming model is constructed by setting initial constraints based on the device information.

[0007] Furthermore, after solving the target planning model to obtain the optimal dynamic activation ratio of each target device, the method further includes: A device scheduling scheme is generated based on the optimal dynamic activation ratio. The equipment scheduling scheme is input into a preset digital twin simulation model and run. Obtain the efficiency evaluation index output by the digital twin simulation model; The equipment scheduling scheme is optimized and adjusted based on the efficiency evaluation indicators. The scheduling is executed based on the optimized and adjusted equipment scheduling scheme.

[0008] Furthermore, after the step of generating a device scheduling scheme based on the optimal dynamic activation ratio and performing scheduling execution, the method further includes: Obtain actual employment data and calculate the deviation between the actual employment data and the predicted employment gap data; The deviation is used as a feedback signal to adjust the parameters of the large language model through a reinforcement learning framework.

[0009] Furthermore, the step of obtaining the target programming model by inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model, further includes: The labor shortage data is quantified into the workload requirements that need to be completed within the preset future time period; Based on the equipment information, determine the unit operating cost of each target device when it is activated; The cost of hiring a unit of human resources; The workload requirement, the unit operating cost, and the unit human resource hiring cost are used as parameters and substituted into the objective function and constraints of the integer programming model to generate the objective programming model.

[0010] Furthermore, the integer programming model is a mixed integer programming model, and the objective function of the integer programming model is to minimize the total operating cost, which includes the start-up, shutdown and operation costs of each target device and the cost of replenishing temporary human resources; the constraints of the integer programming model include at least the warehouse operation throughput demand constraint, the maximum number of target devices in operation constraint, and the upper limit constraint of temporary human resources employment.

[0011] A device dynamic scheduling apparatus based on cultural event perception, the apparatus comprising: The acquisition module is used to acquire multi-source heterogeneous data related to cultural events in a specified target area; The generation module is used to extract and fuse features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors. The prediction module is used to obtain historical employment data of a specified area and input the cultural event feature vector and the historical employment data into a pre-trained large-scale language model to predict the employment gap data in the future preset period. The input module is used to input the labor shortage data, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain a target programming model; wherein, the integer programming model is preset with equipment information of each target device in the specified target area; The solution module is used to solve the target planning model to obtain the optimal dynamic activation ratio of each target device; The execution module is used to generate a device scheduling scheme based on the optimal dynamic activation ratio and to perform scheduling execution.

[0012] An electronic device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

[0013] A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

[0014] The beneficial effects of this invention are as follows: By constructing a feature vector of cultural events and using a large-scale language model for prediction, the timeliness and accuracy of labor shortage prediction are improved. Combined with an integer programming model for dynamic optimization, the real-time and optimal adjustment of the target equipment activation ratio is achieved, which significantly reduces the risk of operational interruption and additional labor costs caused by sudden shortages of manpower. It also enhances the cultural adaptability and overall flexibility of the scheduling system, ensuring the stability of operational efficiency and the efficient use of resources during special periods such as cultural events, and providing a solution to the operational management challenges in cross-cultural regions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] in: Figure 1 This is an application environment diagram of a device dynamic scheduling method based on cultural event awareness in one embodiment. Figure 2 This is a flowchart of a device dynamic scheduling method based on cultural event awareness in one embodiment; Figure 3 This is a structural block diagram of a device dynamic scheduling device based on cultural event perception in one embodiment; Figure 4 This is a structural block diagram of an electronic device in one embodiment. Detailed Implementation

[0017] 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, and 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.

[0018] Figure 1 This is a diagram illustrating a device dynamic scheduling application environment based on cultural event awareness in one embodiment. (Refer to...) Figure 1 This cultural event-aware dynamic device scheduling method is applied to a cultural event-aware dynamic device scheduling system. The system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal; a mobile terminal can be at least one of a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to acquire multi-source heterogeneous data, and the server 120 is used to generate device scheduling schemes and execute them.

[0019] like Figure 2 As shown, in one embodiment, a method for dynamic device scheduling based on cultural event awareness is provided. This method can be applied to both terminals and servers; this embodiment uses terminal application as an example. The method specifically includes the following steps: S1: Acquire multi-source heterogeneous data related to cultural events in a specified target area; S2: Extract and fuse features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors; S3: Obtain historical employment data for a specified area, and input the cultural event feature vector and the historical employment data into a pre-trained large-scale language model to predict the employment gap data for a future preset period. S4: Based on the labor shortage data, and constrained by equipment operating costs and labor costs, input the data into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; S5: Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; S6: Generate a device scheduling scheme based on the optimal dynamic activation ratio and execute the scheduling.

[0020] As described in step S1 above, multi-source heterogeneous data related to cultural events in the specified target area is acquired. In this embodiment, a multi-source data acquisition layer needs to be built first to capture information related to cultural events from various data sources in real time or at regular intervals. Data sources include, but are not limited to: announcements, holiday arrangements, comments, reports, and records from the company's internal sign-in / scheduling / attendance system. Data collection can employ a hybrid approach, including subscription to public application programming interfaces (APIs), access to collaborative data interfaces, and synchronization with local sensors. The acquired data types should cover structured data (announcement dates, holiday tables, population distribution tables), semi-structured web pages, CSV (Comma-Separated Values) logs, and unstructured data (Weibo tweets, comments). Metadata (timestamps, geographic tags, language identifiers, source credibility scores) should also be collected. Preliminary validation and deduplication should be performed during data acquisition, and collection latency and packet loss rates should be recorded. Ideally, the system should cover over 90% of the relevant sources for the target cultural event, with a single collection latency of less than 15 minutes. When key data sources are delayed or missing, historical data interpolation or conservative prediction strategies based on partial information can be used to ensure the system's basic operation under suboptimal data conditions, thereby guaranteeing the timeliness and completeness of subsequent predictions.

[0021] As described in step S2 above, feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors. The acquired data, after preprocessing, enters the feature engineering module. Unstructured textual data is first cleaned (denoising, word segmentation, language detection and translation), named entity recognition is performed, and sentiment analysis is conducted. Then, a multilingual pre-trained language model (e.g., a pre-trained transformer model or its variants) is used to extract semantic vector representations. For less commonly spoken languages, specific pre-trained models (e.g., XLM-Roberta) or fine-tuning strategies can be used. For structured features (e.g., population density), methods such as grid coding and embedding representation can be used to convert them into a vector form that the model can process. Structured data (e.g., holiday countdowns, local population density, historical holiday frequency, population age distribution) is standardized and mapped to numerical features. A fusion strategy is adopted for multimodal features: first, the semantic vector and structured features are aligned (time window synchronization, regional normalization), and then weighted concatenation or attention mechanism is used for fusion to generate cultural event feature vectors with fixed dimensions (e.g., 128 dimensions). In order to improve the discriminative power, feature selection and dimensionality reduction (principal component analysis or autoencoder) can be introduced and the vectors are normalized. Finally, the prediction model is provided with an input representation with high signal-to-noise ratio and cross-language and cross-regional compatibility.

[0022] As described in step S3 above, historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset time period. Historical employment data includes store or warehouse attendance records, shift logs, overtime and leave records, temporary employment history, and historical capacity / throughput time series. Before inputting into the pre-trained large-scale language model (e.g., a model based on a Transformer architecture and fine-tuned with industry data), the historical time series needs to be seasonally decomposed, imputed for missing values, and normalized. It is then concatenated with the cultural event feature vector generated in step S2 according to a time window or used as contextual input. The model can adopt a multimodal input structure and output the probability distribution of labor shortages by hour or time period for the next few days (e.g., 7 days) or the expected number of people / workload in short supply. The prediction module outputs confidence intervals and uncertainty estimates at the same time and supports threshold alarms. To ensure the model's effectiveness, this embodiment adopts online fine-tuning and periodic backtracking training. Historical prediction errors (e.g., mean absolute percentage error) are used to monitor model performance and trigger retraining or incremental updates.

[0023] As described in step S4 above, based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model. The integer programming model contains preset equipment information for each target device in the specified target area. The labor shortage obtained in S3 is quantified into the workload requirements for each future time period (e.g., number of packages processed per hour), and parameters for various target devices are read from the equipment information database: maximum processing capacity of a single device, start-up and shutdown costs, unit operating energy consumption and consumable costs, available quantity, response / switching time, and maintenance window, etc. Simultaneously, the unit hiring cost of manpower, the number of available temporary workers, and the response time are obtained. These parameters are mapped to the parameter set of the integer programming model to construct decision variables (e.g., binary variables representing whether a device is activated in a certain time period, and integer variables representing the number of activated devices), and a target programming model with specific numerical parameters is formed accordingly. In this embodiment, the target programming model is a mixed integer linear programming or mixed integer nonlinear programming model, with a clearly defined objective function and constraints for subsequent solution.

[0024] As described in step S5 above, the target planning model is solved to obtain the optimal dynamic activation ratio of each target device. For the target planning model, an appropriate solver (such as the Gurobi optimizer, CPLEX optimizer, or an open-source branching solver) is selected for solving. To improve solution efficiency, engineering methods can be adopted: model partitioning and time-period decomposition, heuristic initial solutions, parallel solving, and contextualized or robust optimization extensions for uncertain demands. The solution objective is usually to minimize the total cost (the sum of equipment operation and labor costs) while satisfying throughput and service level constraints. The obtained solution includes the number or activation probability (i.e., activation ratio) of each type of device in each time period. The solution process needs to record the solution duration, the difference between the optimal and suboptimal solutions, and feasibility. In actual deployment, a rolling time-domain strategy can be adopted, distributing the optimal decision for the near future period for execution while retaining the long-term strategy for subsequent correction.

[0025] As described in step S6 above, an equipment scheduling scheme is generated based on the optimal dynamic activation ratio and executed. The optimal activation ratio is converted into executable scheduling instructions, which are then distributed to each target device or automated guided vehicle controller through the warehouse management system / warehouse control system or equipment management system to perform specific start / stop, task allocation, and path adjustment operations. Before distribution, the scheduling scheme can be quickly simulated and verified in a digital twin environment to detect potential congestion or conflicts and make fine adjustments. The execution layer should include safety and compliance checks, smooth start / stop strategies (considering equipment thermal state and energy consumption), and manual intervention interfaces to handle anomalies. During implementation, execution data (equipment status, actual capacity, employee attendance) is continuously collected and compared with the prediction results to form a closed-loop feedback, providing training signals for subsequent model calibration and reinforcement learning. At the same time, privacy protection and localization measures are ensured during data processing to comply with regulatory requirements.

[0026] In one embodiment, step S2, which involves extracting and fusing features from the multi-source heterogeneous data to generate a corresponding cultural event feature vector, includes: S201: Use a multilingual pre-trained language model to extract semantic features from the multi-source heterogeneous data to obtain semantic features; S202: Obtain the structured features of cultural events in the specified target area, and fuse the semantic features with the structured features to obtain fused features; wherein, the structured features include the time information of cultural events in the specified target area and the population density distribution information of the content in the specified target area; S203: The fused features are vectorized and encoded to generate a fixed-dimensional feature vector of the cultural event.

[0027] As described in step S201 above, a multilingual pre-trained language model is used to extract semantic features from the multi-source heterogeneous data to obtain semantic features. Text or semi-structured text from different sources (such as press releases, comment information, short texts in local languages, etc.) is first cleaned and preprocessed, including denoising (such as hypertext markup language tags, Uniform Resource Locators, and duplicate content), language detection and sentence segmentation, word segmentation or tokenization, Uniform Character Encoding, and necessary translation or normalization. Subsequently, a multilingual pre-trained language model (such as multilingual BERT, XLM-R multilingual model, or a multilingual Transformer fine-tuned with in-domain corpus) is used to perform contextual semantic encoding on the processed text. Specifically, the process includes converting the text into a model input sequence, obtaining the context vector or CLS (Classification Token Vector) vector for each tag through forward inference, and generating sentence-level or document-level semantic vectors using pooling strategies (such as CLS pooling, average pooling, or attention-based weighted pooling) as needed. Furthermore, this step allows for the parallel extraction of various semantically relevant features: sentiment scores, topic distribution, named entity recognition (such as holiday names), event time and location entities, and topic popularity indicators. These additional semantic signals can be output along with the basic semantic vectors, along with source credibility and timestamps, providing high-dimensional semantic representation and quality assessment information for subsequent fusion. To meet real-time requirements, model inference can employ batch processing, GPU acceleration, and vector caching strategies. Domain adaptation or few-sample fine-tuning should be used for low-resource languages ​​or dialects to improve representation quality.

[0028] As described in step S202 above, the structured features of cultural events in the specified target area are obtained, and the semantic features are fused with the structured features to obtain fused features. First, numerical or categorical features directly related to cultural events are read from the structured data source, such as holiday countdown (hours / days between the start of the event), event duration, event type (cultural event, statutory holiday, etc.), population density distribution of the target area (statistics by grid or administrative division), and attendance rate fluctuations in the same period over the years. Spatial mapping of geographic information (e.g., mapping coordinates or addresses to administrative regions or latitude / longitude grids) is performed, and population density is normalized or binned. Subsequently, the semantic vectors and structured features output by S201 are spatiotemporally aligned: aggregation is performed according to a unified time window (e.g., hour or day), merging is based on geographic labels, and compensation is made for time delays from different sources. Various fusion strategies can be employed: linear transformation and normalization are performed after feature-level concatenation; attention-based cross-modal fusion, such as calculating the attention weights of semantic vectors to structured features through a single cross-modal attention network; or a gating mechanism is used to adjust the contribution of different features based on source credibility, temporal distance, and semantic relevance. To enhance robustness, statistical features (e.g., rolling mean, variance) and cross features can be introduced, and imputation or feature masks are applied to missing values ​​to maintain the stability of the fusion network.

[0029] As described in step S203 above, the fused features are vectorized and encoded to generate a fixed-dimensional cultural event feature vector. After completing multimodal fusion, the high-dimensional and heterogeneous fused features need to be mapped into a unified, fixed-dimensional vector for subsequent prediction model input and storage. First, scaling transformation (such as standardization or min-max scaling) and category feature encoding are performed on the fused features. Then, dimensionality reduction and semantic compression are performed through one to multiple fully connected neural networks or autoencoders. The network can incorporate batch normalization and random deactivation to prevent overfitting, and activation functions (ReLU function, GELU function, etc.) are used to improve nonlinear expression capabilities. To obtain stable and semantically dense output, principal component analysis or further compression based on variational autoencoders can be used. Finally, the vector dimension is fixed to a preset value (e.g., 128 or 256 dimensions), and L2 normalization or standardization is performed to facilitate distance calculation and similarity retrieval. After encoding, the vector, along with meta-information (timestamp, regional label, version number, source confidence), is written into a feature library or vector database (such as FAISS) to support fast retrieval and online inference. Simultaneously, a feature version management and drift detection mechanism is established to monitor changes in feature distribution and trigger retraining or feature regeneration processes, ensuring that the generated cultural event feature vectors are stable and interpretable in the long term.

[0030] In one embodiment, before step S4, which involves inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain the target programming model, the method further includes: S301: Obtain device information for each target device in the specified target area; S302: Based on the total number of target devices in the device information, define a decision variable to indicate whether each target device will be enabled within a preset time period in the future; S303: Based on each decision variable, construct an objective function with the goal of minimizing the total operating cost; wherein the total operating cost consists of the equipment operating cost and the labor cost; S304: Based on the device information, set initial constraints and construct the integer programming model.

[0031] As described in step S301 above, acquire the equipment information of each target device in the designated target area. Constructing an equipment information database aims to provide a complete set of parameters for subsequent modeling. Specifically, static and dynamic attributes of each or each type of target device should be collected from channels such as equipment management systems, warehouse management systems / control systems, equipment manufacturer documentation, and field sensors. Static attributes include equipment model / type (e.g., automated guided vehicle controller, sorting robot arm, conveyor line unit), rated processing capacity (number of items or work processes that can be processed per unit time), maximum concurrent tasks, physical dimensions and work adaptation area, and equipment identification and network interface information. Dynamic attributes include unit operating cost (including energy consumption, consumables, depreciation and amortization), start-up and shutdown costs or start-up delay (including preheating time and cost), minimum continuous operating time and minimum downtime, average failure rate and maintenance cycle, availability time window (maintenance or statutory downtime), response time, current health status, and predicted maintenance time. In addition, the matching relationship between equipment and work type, inter-equipment coordination constraints (e.g., only one device is allowed to operate in the same channel), and equipment location and movement path information should be collected. After collection, the data is cleaned, formatted, and versioned, and stored in a structured equipment information table or asset management database, with timestamps and credibility assessments attached, so that it can be used as deterministic or probabilistic parameters in modeling.

[0032] As described in step S302 above, based on the total number of target devices in the device information, a decision variable is defined to indicate whether each target device is enabled within a preset time period in the future. The device information is mapped to the basic decision unit of mathematical programming. First, the time domain is divided into discrete time periods (such as every hour or every 15 minutes), and the devices are indexed by single unit or by type, for example, device i∈I and time period t∈T. Based on the total number of configurations, a binary decision variable x{i,t}∈{0,1} is defined to indicate whether device i is enabled (1) or disabled (0) in time period t; or an integer decision variable y{k,t} can be defined for devices of the same type, indicating the number of k-th type devices enabled in time period t (0≤y{k,t}≤total number of configurations). To characterize the load distribution, a continuous variable z{i,t} can also be defined to indicate the actual processing volume or load ratio of device i in time period t (bounded to [0,capacity_i]). At the same time, auxiliary variables are introduced to handle start-up and shutdown costs (such as u{i,t}, which indicates that the start-up and shutdown action occurs at time t), and state transition variables with coherence constraints are introduced to meet the timing characteristics such as shortest running time and cooling time. The variable definition stage should take into account both scalability and solution efficiency, and decide whether to perform fine modeling on a single unit or aggregate modeling by type to balance accuracy and computational complexity.

[0033] As described in step S303 above, an objective function is constructed based on each decision variable, with the goal of minimizing the total operating cost; wherein the total operating cost consists of the equipment operating cost and the labor cost. The decision variables and cost parameters are combined to construct the optimization objective. The objective function is typically formalized as minimizing the total cost, i.e., a weighted summation of the unit hour operating cost of the equipment, one-time start-up and shutdown costs, load factoring, the number of temporary workers required multiplied by the unit hiring cost, and the portion of the labor shortage not replaced by equipment. Other cost items can also be added: maintenance costs accumulated over operating hours, penalties for breach of contract or delays, equipment switching costs, or penalties for load imbalance. If multiple objectives (such as cost and service level) need to be considered simultaneously, a weighted summation or hierarchical optimization strategy can be adopted. To improve robustness, uncertainties can be incorporated into the objective function through expected costs, insurance costs, or scenario-based weighted averages, forming a stochastic or robust optimization problem.

[0034] As described in step S304 above, initial constraints are set based on the equipment information to construct the integer programming model. Equipment capabilities and business requirements are mapped to specific constraints to ensure the feasibility of the solution. After defining the constraints and objective function, a complete mixed integer programming (mixed integer programming model) model is formed and exported to the solver at the modeling interface. To facilitate solving and online application, the constraint matrix should be sparsified, preprocessed (e.g., irrelevant constraint removal), and numerical tolerance and initial solution strategies should be set to improve the solution speed and ensure practicality and executability.

[0035] In one embodiment, after step S5 of solving the target planning model to obtain the optimal dynamic activation ratio of each target device, the method further includes: S601: Generate a device scheduling scheme based on the optimal dynamic activation ratio; S602: Input the equipment scheduling scheme into the preset digital twin simulation model and run it; S603: Obtain the efficiency evaluation index output by the digital twin simulation model; S604: Optimize and adjust the equipment scheduling scheme based on the efficiency evaluation indicators; S605: Execute scheduling based on the optimized and adjusted equipment scheduling scheme.

[0036] As described in step S601 above, an equipment scheduling scheme is generated based on the optimal dynamic activation ratio. The continuous or probabilistic "optimal dynamic activation ratio" obtained in S5 is converted into an executable, discretized scheduling scheme. First, the target activation ratio of each equipment category or single equipment is read in time segments (e.g., every hour or every 15 minutes). The ratio is multiplied by the total number of configurations in that category or region and rounded to obtain the specific number of equipment to be activated in each time segment. The rounding strategy can use the nearest integer, rounding up to ensure service level, or allocation based on priority (critical operations first). Second, equipment-level task allocation and start-up sequence are generated, including equipment start-up and shutdown time points, warm-up and cool-down time, minimum continuous operation constraints, shift switching window, and interface mapping with the warehouse management system / warehouse control system (pointing to specific work queues, workstations, or routes). When generating a plan, the equipment maintenance window, energy consumption peak limit, human coordination and cooperation and safety interval (to avoid grid impact or traffic congestion caused by simultaneous start-up and shutdown) should be considered. At the same time, a scheduling visualization table and difference description (compared with the conventional fixed configuration) should be generated, and priority labels, emergency rollback thresholds and manual approval points should be generated to facilitate the review by operation and maintenance personnel or manual intervention in emergencies.

[0037] As described in step S602 above, the equipment scheduling scheme is input into a preset digital twin simulation model for execution. After the scheme is generated, it is first verified in a "sandbox" environment in the digital twin environment to reduce on-site risks. The time series of the scheduling scheme, equipment start / stop commands, task flow and path allocation, etc., are input into the digital twin model (which should include high-fidelity physical behavior, collision detection, automated guided vehicle controller path planning logic and concurrent load model), and run in simulation time steps. The simulation needs to load the topology of the real venue, shelf layout, aisle restrictions, operation arrival curves and manual operation rules to visualize and reproduce potential bottlenecks, congestion points, and cross-conflicts. To improve the effectiveness of verification, multiple scenarios should be run: nominal scenario (based on predicted demand), high-demand scenario (demand amplification), disturbance scenario (equipment failure or partial shutdown), and random disturbance to evaluate robustness. The simulation process should also support multiple parallel trials, record event logs, and trigger alarms or rollback suggestions in abnormal situations.

[0038] As described in step S603 above, obtain the efficiency evaluation indicators output by the digital twin simulation model. After the simulation runs, automatically collect and calculate multi-dimensional efficiency evaluation indicators to determine whether the solution meets business and security objectives. Commonly used indicators include order processing throughput (number of orders completed per unit time), average and 95th / 99th percentile job latency, equipment utilization and idle rate, task waiting time, congestion frequency and duration, channel occupancy rate, number of task switching times (affecting equipment lifespan), energy consumption and peak power, number of fault / conflict events, and service level agreement breach rate or potential penalty estimates, etc. Also output should be a visual heatmap, time series curve, and key event replay to support gradual bottleneck identification. For uncertain scenarios, output statistical summaries (mean, variance, confidence interval) to evaluate the robustness of the solution. The evaluation results need to be compared with preset thresholds and key business performance indicators, and a suggestion level (pass / needs optimization / not executable) and corresponding optimization direction prompts (such as adding equipment, changing paths, adjusting start / stop timing, or manual intervention) should be generated.

[0039] As described in step S604 above, the equipment scheduling scheme is optimized and adjusted based on the efficiency evaluation indicators. According to the indicators and recommendation levels output by the simulation, an automatic or semi-automatic optimization and adjustment process is initiated. If the indicators are within acceptable limits, minor parameter tuning can be performed (such as delaying certain start / stop times, fine-tuning the number of devices, and changing priority weights), followed by another short-cycle simulation verification. If significant congestion or service level agreement risks are found, structural adjustments should be triggered, such as reallocating devices to bottleneck areas, changing task allocation strategies (by batch / by priority / dynamic sharding), or enabling backup manual shifts. The optimization process can use heuristic rules, local search, or metaheuristic algorithms (genetic algorithms, simulated annealing) to quickly explore scheduling variables. Alternatively, the optimizer can be called to re-solve specific time periods, for example, by fine-tuning using a local mixed integer programming model. All adjustments must be logged, and the impact of changes on energy consumption and cost must be assessed. Multiple options should be provided to decision-makers for manual selection when necessary. Optimization should also consider switching overhead during execution to avoid frequent start / stop operations introducing higher start / stop costs or equipment wear.

[0040] As described in step S605 above, scheduling is executed based on the optimized and adjusted equipment scheduling scheme. The verified and optimized final scheme is issued to the execution layer (warehouse management system / warehouse control system, equipment controller, and on-site operation and maintenance interface) for actual execution. A transactional, batch confirmation mechanism is adopted during the issuance process: first, sub-schemes that can be executed in the near future (e.g., within 1 hour) are issued and equipment / controller confirmation is obtained, and then long-term plans are issued according to the rolling time domain. The execution phase includes issuing equipment start and stop commands, task allocation, and path updates, while simultaneously generating manual operation instructions (e.g., temporarily adding manual packing positions). Execution monitoring requires enabling real-time telemetry (equipment status, location, task progress), and key indicators are sent back to the scheduling engine to support online re-optimization. If an anomaly occurs on-site (equipment failure, insufficient personnel, unexpected blockage), the system automatically triggers emergency plans or manual alarms according to the preset rollback strategy and records the event for subsequent learning. The entire execution process must ensure safety and compliance (collision prevention, energy consumption limits, and adherence to maintenance windows) and be recorded in a closed loop to provide real data support for the next round of prediction and optimization.

[0041] In one embodiment, after step S6 of generating a device scheduling scheme based on the optimal dynamic activation ratio and performing scheduling execution, the method further includes: S701: Obtain actual employment data and calculate the deviation between the actual employment data and the predicted employment gap data; S702: The deviation is used as a feedback signal to adjust the parameters of the large language model through a reinforcement learning framework.

[0042] As described in step S701 above, actual employment data is obtained, and the deviation between the actual employment data and the predicted employment gap data is calculated. First, actual employment-related data is collected and summarized from the execution layer, including the actual number of employees present, the actual amount of work completed in each time period, the number of temporary employees, and manual processing capacity, etc., and aligned with the same time granularity as the prediction period (e.g., by hour or 15 minutes). After alignment, data cleaning is required: abnormal records are removed, missing data is filled, and atypical fluctuations caused by sudden events (such as extreme weather or equipment failure) are identified and labeled to avoid misleading model updates. Then, the deviation term is calculated. Commonly used metrics include absolute error (AE), root mean square error (RMSE), mean absolute percentage error (MAPE), and risk-weighted error (with increased weight for critical periods or key business). To support fine-grained analysis, the deviation distribution should be calculated by equipment type, shift, workstation, and region, and deviation time series and spatial heatmaps should be generated to determine whether the inadequacy in positioning prediction is due to systematic deviation or random noise. The deviation calculation results need to be accompanied by confidence and statistical significance tests. Only when the deviation exceeds a preset threshold or abnormal patterns occur continuously should it be used as a signal to trigger reinforcement learning updates, so as to avoid frequent or unnecessary fine-tuning of the model. At the same time, deviation logs, raw data, and context (such as the announcement time of cultural events and sudden public opinion on social media) should be archived for subsequent offline analysis and regulatory audits.

[0043] As described in step S702 above, the deviation is used as a feedback signal to adjust the parameters of the large language model through a reinforcement learning framework. Using the deviation obtained in S701 as environmental feedback, a reinforcement learning (RL) process is constructed or driven to enhance prediction and decision-making capabilities. Specifically, two implementation paths can be adopted: First, policy-based fine-tuning, adding a trainable policy layer or calibration network (policy head) on top of the large language model (LLM), using the cultural event feature vector and the initial output of the LLM as state inputs, and the policy network outputting predicted calibration values ​​or assigning different feature weights. Stable policy gradient algorithms such as proximal policy optimization and A2C (policy gradient method) are employed, utilizing the negative / positive value of the deviation and the business cost function as reward signals (e.g., using the predicted cost reduction or improved service level agreement compliance as a positive reward); Second, direct fine-tuning based on value or behavior, through RLHF (Reinforcement Learning from...). The Human / Environment Feedback method is used to fine-tune LLM parameters, employing offline batch processing and safety strategies (low learning rate, Gaussian noise constraints on parameters, low-rank adaptation, and other efficient parameter fine-tuning techniques). Long-term policy performance is simulated in a digital twin environment to estimate the true reward, avoiding high-risk online weight updates in the production environment. A safety net is implemented during training: rollback strategies, A / B testing, canary releases, and simulation verification to ensure the new policy performs no worse than the baseline on several key performance indicators. Model evaluation includes both standard model metrics (MAPE, RMSE) and business-oriented metrics (equipment utilization, labor replenishment costs, throughput retention). Finally, the trained policy / fine-tuned parameters are periodically or triggered to the online model, with version management, drift monitoring, and compliance auditing to ensure transparent and traceable updates, conducted within privacy and regulatory requirements (e.g., anonymizing sensitive attendance data or using federated learning).

[0044] In one embodiment, step S4, which involves inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain the target programming model, further includes: S401: Quantify the labor shortage data into the workload requirements that need to be completed within the future preset time period; S402: Based on the device information, determine the unit operating cost of each target device when it is activated; S403: Cost of acquiring a unit of human resources; S404: The workload requirement, the unit operating cost, and the unit human resource employment cost are used as parameters and substituted into the objective function and constraints of the integer programming model to generate the objective programming model.

[0045] As described in step S401 above, the labor shortage data is quantified into the workload requirement to be completed within the preset future time period. The predicted labor shortage (usually represented by the number of people or probability distribution) is converted into a workload requirement index that can be used to optimize the model. First, the prediction results are aligned with the time granularity consistent with the planning model (e.g., by hour or 15 minutes). Then, the number of people in shortfall is multiplied by the average workload that can be completed per unit of manpower within the time period (e.g., the processing volume per piece / hour, or per person / shift), thereby obtaining the workload requirement in units such as "number of work pieces / number of processes / processing hours to be completed". During the process, abnormal prediction values ​​need to be smoothed or pruned, while considering the rhythm and lag effects (e.g., short-term labor shortages). A conservative strategy is adopted for predictions containing confidence intervals (e.g., taking the upper confidence bound or scenario set). To support robust optimization, several scenarios (baseline, high demand, extremely low supply) can also be generated and weights assigned to each scenario so that the subsequent model can use expected / worst-case handling in the objective function or constraints. The final output is a serialized, time-segmented workload demand vector, which can be directly input into an integer programming model or used as a constraint right-hand side term in scenario-based solutions.

[0046] As described in step S402 above, based on the equipment information, the unit operating cost of each target device when it is activated is determined. The unit operating cost typically includes energy consumption cost (calculated based on energy consumption curves and time-of-use electricity prices), cost allocated to manual monitoring or maintenance personnel, routine depreciation and amortization, consumable consumption, and additional wear and tear costs caused by start-up, shutdown, or load changes. In practice, parameters need to be extracted from equipment operation logs, energy consumption meters, maintenance records, and procurement / depreciation policies, and standardized into "cost per device per hour" or "cost per device per unit" by time period or by task billing method. In addition, the one-time start-up and shutdown costs should be amortized into the relevant time period (as a start-up cost item), and the energy efficiency differences under different load rates should be segmented and modeled (for example, the unit processing cost is higher at low loads). In actual calculations, dynamic electricity prices, seasonal maintenance plans, and shared cost allocation caused by concurrent operation of equipment should be considered, and cost transparency and auditable records should be maintained to facilitate subsequent sensitivity analysis and cost updates.

[0047] As described in step S403 above, the unit human resource employment cost is obtained, and human resource-related expenses are refined into unit cost inputs acceptable to the model. The unit employment cost includes not only basic wages, but also social security and benefits, recruitment and training amortization, temporary worker agency fees, overtime pay, commuting allowances, and potential recruitment delay costs (such as hidden costs that are difficult to increase manpower in the short term). In practice, data from the enterprise's human resource system and financial system should be integrated to calculate standardized "per person per hour" or "per workpiece" cost, and tiered cost coefficients should be set according to different job types, shifts (night shifts, holidays), and employment forms (contract workers, temporary dispatch). To facilitate comparison with equipment substitution, human productivity (processing volume per person per hour) should be converted into unit operating cost (cost / piece). In regions with outsourcing or labor market fluctuations, market premium and employment elasticity parameters should be introduced. In compliance-sensitive scenarios, the costs of statutory minimum guarantees and emergency reserve manpower should also be included to ensure the legality and feasibility of the decision.

[0048] As described in step S404 above, the workload requirement, the unit operating cost, and the unit human resource hiring cost are used as parameters and substituted into the objective function and constraints of the integer programming model to generate the target programming model. The aforementioned quantification results are mapped to specific parameters and right-hand side terms of the optimization model: workload requirement is used as the right-hand side vector of the demand constraint, equipment unit operating cost and start-up / shutdown cost are used as linear or piecewise cost coefficients in the objective function, and human resource unit cost is used as a gap compensation or penalty term. To enhance robustness, scenario-based constraints can be introduced or a penalty term for unmet requirements can be added to the objective function to avoid constraint infeasibility. After modeling, data normalization, numerical stability checks, and feasibility pre-solution (such as relaxation checks) should be performed, and the model should be exported to the solver interface (including initial solution and time limit settings) to ensure that the target programming model can reflect real costs and capacity constraints and is suitable for real-time or near-real-time solutions.

[0049] In one embodiment, the integer programming model is a mixed integer programming model, and the objective function of the integer programming model is to minimize the total operating cost, which includes the start-up, shutdown and operation costs of each target device and the cost of replenishing temporary human resources; the constraints of the integer programming model include at least the warehouse operation throughput demand constraint, the maximum number of target devices in operation constraint, and the upper limit constraint of temporary human resources employment.

[0050] Mixed-integer programming (MIPC) models are mathematical models that combine integer variables (usually binary or integer, representing discrete decisions such as the number of devices started / stopped or activated) with continuous variables (such as equipment load and labor input) within the same optimization framework. In this embodiment, the MIPC model is used to characterize the coupling relationship between the discreteness of device start / stop and the continuity of capacity / cost. Examples of decision variables include binary variables x{i,t} (whether device i is activated in time period t), integer variables y{k,t} (the number of devices of type k activated in time period t), and continuous variables z{i,t} (the actual throughput of device i in time period t) or ht (the number of temporary workers hired in time period t). MIPC models can directly incorporate start / stop one-time costs, linear operating costs, and labor costs into the objective function, and express capacity, timing, and availability constraints through linear or piecewise linear constraints. In practice, to improve solvability, strategies such as time segmentation, type clustering, heuristic initial solutions, relaxation testing and rolling horizon are often used, along with commercial solvers (such as Gurobi / CPLEX) or decomposition algorithms (Benders decomposition) to achieve industrial-grade solutions.

[0051] The objective function aims to minimize total operating costs, which typically consist of several quantifiable components: 1) the operating costs of the target equipment (charged hourly or per processing volume during the usage period, including energy consumption, consumables, and depreciation); 2) the start-up and shutdown costs or start-up delay costs (one-time expenses or time costs incurred due to warm-up / switching); 3) the cost of replenishing temporary human resources (charged per number of temporary workers or per hour, including agency fees, overtime pay, and training amortization); and 4) penalties for breach of contract / delay or service level agreement (costs introduced when demand is not met) set to ensure service levels. This objective balances short-term operational economics with long-term equipment lifespan / maintenance costs. To address demand uncertainty, the objective function can be extended to expected costs (scenario-weighted) or robust / risk-sensitive forms, thereby achieving a balance between cost minimization and robustness.

[0052] Throughput requirement constraints ensure that, within any given planning period, the sum of equipment processing capacity and manual processing capacity at least meets the predicted workload demand. This constraint must consider timing synchronization (consistent time granularity), processing occupancy (equipment is not 100% used for the target job, requiring a utilization coefficient), and task parallel / serial characteristics (some jobs require multiple processes to be sequentially executed). If the demand is random or has a confidence interval, scenario constraints or slack variables can be introduced, and the unmet quantity can be represented in the objective function as a penalty to ensure that the model can produce feasible and controllable alternatives even in infeasible situations. This constraint is core to maintaining service levels and throughput capacity, directly affecting the allocation decisions for manual and equipment substitution. Maximum operating quantity constraints limit the activation upper limit of a certain type or all target equipment within any given period, reflecting physical limitations such as physical configuration, space constraints, or power / channel capacity. This constraint can also be refined into area / workstation level constraints (the upper limit of the number of devices running simultaneously in a certain channel or workstation), shared resource constraints (the limit of the number of devices maintained simultaneously by the same maintenance team), and concurrency safety constraints (avoiding the simultaneous activation of multiple automated guided vehicle controllers in collision zones). In addition, to reflect the minimum continuous operation / downtime limit, a timing chaining constraint is usually introduced. For example, if x{i,t} changes from 0 to 1, it is required to remain at 1 for the next minimum continuous period to avoid equipment wear, energy consumption peaks, or maintenance burden caused by frequent start-ups and shutdowns. This constraint improves the feasibility of the model by limiting the decision space and ensures that the solution can be implemented safely and compliantly in the actual field.

[0053] The temporary workforce hiring cap constraint specifies the upper limit on the number of temporary workers that can be hired within any given time period, expressed as ht ≤ Hmaxt, where ht represents the number of temporary workers at time t, and Hmaxt (the temporary worker hiring cap at time t) is determined by labor market supply, employment compliance policies, or corporate contractual restrictions. This constraint reflects real-world constraints: rapid recruitment of temporary workers faces time and cost limitations, legal restrictions on continuous working hours and the ratio of temporary workers to temporary workers, and issues related to the quality and training time of temporary workers. Therefore, this constraint not only affects the optimal solution for minimizing costs but also directly determines whether it is necessary to replace manual labor by increasing equipment utilization. The model can also introduce recruitment delay and lag variables (e.g., new temporary workers need to wait for a certain number of time periods to become effective), as well as tiered costs (temporary workers on night shifts and holidays are more expensive). If Hmaxt cannot meet the demand at time t in certain time periods, a gap at time t can be introduced, and a high penalty or the triggering of alternative solutions (e.g., outsourcing, overtime) can be imposed in the objective function to ensure that the model is always solvable and to provide decision-makers with practical and actionable contingency suggestions.

[0054] Reference Figure 3 The present invention also provides a device for dynamic scheduling of equipment based on cultural event perception, the device comprising: The acquisition module 902 is used to acquire multi-source heterogeneous data related to cultural events in a specified target area; The generation module 904 is used to extract and fuse features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors. The prediction module 906 is used to acquire historical employment data of a specified area and input the cultural event feature vector and the historical employment data into a pre-trained large-scale language model to predict the employment gap data in a future preset period. The input module 908 is used to input the labor shortage data into a preset integer programming model based on the constraints of equipment operating costs and labor costs to obtain a target programming model; wherein, the integer programming model is preset with equipment information of each target device in the specified target area; The solution module 910 is used to solve the target planning model to obtain the optimal dynamic activation ratio of each target device; The execution module 912 is used to generate a device scheduling scheme based on the optimal dynamic activation ratio and to perform scheduling execution.

[0055] In one embodiment, the generation module 904 includes: The semantic feature acquisition submodule is used to extract semantic features from the multi-source heterogeneous data using a multilingual pre-trained language model to obtain semantic features. The structured feature acquisition submodule is used to acquire the structured features of cultural events in the specified target area, and fuse the semantic features with the structured features to obtain fused features; wherein, the structured features include the time information of cultural events in the specified target area and the population density distribution information of the content in the specified target area; The vectorization encoding submodule is used to perform vectorization encoding on the fused features to generate a fixed-dimensional feature vector of the cultural event.

[0056] In one embodiment, the device dynamic scheduling device based on cultural event perception further includes: The device information acquisition module is used to acquire device information of each target device in the specified target area; The definition module is used to define decision variables that indicate whether each target device will be enabled within a preset time period in the future, based on the total number of target devices configured in the device information. The objective function construction module is used to construct an objective function based on various decision variables, with the goal of minimizing the total operating cost; wherein the total operating cost consists of the equipment operating cost and the labor cost; The integer programming model acquisition module is used to set initial constraints based on the device information and construct the integer programming model.

[0057] In one embodiment, the device dynamic scheduling device based on cultural event perception further includes: The equipment scheduling scheme generation module is used to generate an equipment scheduling scheme based on the optimal dynamic activation ratio; The operation module is used to input the equipment scheduling scheme into a preset digital twin simulation model for execution; An efficiency evaluation index acquisition module is used to acquire the efficiency evaluation index output by the digital twin simulation model. The equipment scheduling scheme adjustment module is used to optimize and adjust the equipment scheduling scheme based on the efficiency evaluation index; The scheduling execution module is used to perform scheduling based on the optimized and adjusted equipment scheduling scheme.

[0058] In one embodiment, the device dynamic scheduling device based on cultural event perception further includes: The deviation calculation module is used to acquire actual employment data and calculate the deviation between the actual employment data and the predicted employment gap data. The parameter adjustment module is used to adjust the parameters of the large language model by taking the deviation as a feedback signal through a reinforcement learning framework.

[0059] In one embodiment, the input module 908 further includes: The workload demand conversion submodule is used to quantify the labor shortage data into the workload demand that needs to be completed within the future preset time period; The unit operating cost determination submodule is used to determine the unit operating cost of each target device when it is activated, based on the device information. The Employment Cost Acquisition Submodule is used to acquire the employment cost per unit of human resources. The goal programming model generation submodule is used to take the workload requirement, the unit operating cost, and the unit human resource employment cost as parameters, and substitute them into the objective function and constraints of the integer programming model to generate the goal programming model.

[0060] In one embodiment, the integer programming model is a mixed integer programming model, and the objective function of the integer programming model is to minimize the total operating cost, which includes the start-up, shutdown and operation costs of each target device and the cost of replenishing temporary human resources; the constraints of the integer programming model include at least the warehouse operation throughput demand constraint, the maximum number of target devices in operation constraint, and the upper limit constraint of temporary human resources employment.

[0061] Figure 4An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a dynamic device scheduling method based on cultural event awareness. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement the dynamic device scheduling method based on cultural event awareness. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0062] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

[0063] By constructing feature vectors of cultural events and using large-scale language models for prediction, the timeliness and accuracy of labor shortage prediction were improved. Combined with integer programming models for dynamic optimization, the real-time and optimal adjustment of the target equipment utilization ratio was achieved, which significantly reduced the risk of operational interruption and additional labor costs caused by sudden shortages of manpower. It also enhanced the cultural adaptability and overall flexibility of the scheduling system, ensuring the stability of operational efficiency and the efficient use of resources during special periods such as cultural events, and providing a solution to the operational management challenges in cross-cultural regions.

[0064] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following steps: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

[0065] By constructing feature vectors of cultural events and using large-scale language models for prediction, the timeliness and accuracy of labor shortage prediction were improved. Combined with integer programming models for dynamic optimization, the real-time and optimal adjustment of the target equipment utilization ratio was achieved, which significantly reduced the risk of operational interruption and additional labor costs caused by sudden shortages of manpower. It also enhanced the cultural adaptability and overall flexibility of the scheduling system, ensuring the stability of operational efficiency and the efficient use of resources during special periods such as cultural events, and providing a solution to the operational management challenges in cross-cultural regions.

[0066] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for dynamic equipment scheduling based on cultural event perception, characterized in that, The method includes: Acquire multi-source heterogeneous data related to cultural events in a specified target area; Feature extraction and fusion are performed on the multi-source heterogeneous data to generate corresponding cultural event feature vectors; Historical employment data for a specified region is obtained, and the cultural event feature vector and the historical employment data are input into a pre-trained large-scale language model to predict the employment gap data for a future preset period. Based on the labor shortage data, and constrained by equipment operating costs and labor costs, the data is input into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information for each target device in the specified target area; Solve the target planning model to obtain the optimal dynamic activation ratio of each target device; Based on the optimal dynamic activation ratio, a device scheduling scheme is generated and executed.

2. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, The step of extracting and fusing features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors includes: Semantic features are extracted from the multi-source heterogeneous data using a multilingual pre-trained language model. The structured features of cultural events in the specified target area are obtained, and the semantic features are fused with the structured features to obtain fused features; wherein, the structured features include the time information of cultural events in the specified target area and the population density distribution information of the content in the specified target area; The fused features are vectorized and encoded to generate a fixed-dimensional feature vector of the cultural event.

3. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, Before the step of inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain the target programming model, the method further includes: Obtain device information for each target device in the specified target area; Based on the total number of target devices in the device information, a decision variable is defined to indicate whether each target device will be enabled within a preset future time period; Based on each decision variable, an objective function is constructed with the goal of minimizing the total operating cost; wherein the total operating cost consists of the equipment operating cost and the labor cost; The integer programming model is constructed by setting initial constraints based on the device information.

4. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, After solving the target planning model to obtain the optimal dynamic activation ratio of each target device, the method further includes: A device scheduling scheme is generated based on the optimal dynamic activation ratio. The equipment scheduling scheme is input into a preset digital twin simulation model and run. Obtain the efficiency evaluation index output by the digital twin simulation model; The equipment scheduling scheme is optimized and adjusted based on the efficiency evaluation indicators. The scheduling is executed based on the optimized and adjusted equipment scheduling scheme.

5. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, After the step of generating a device scheduling scheme based on the optimal dynamic activation ratio and executing the scheduling, the method further includes: Obtain actual employment data and calculate the deviation between the actual employment data and the predicted employment gap data; The deviation is used as a feedback signal to adjust the parameters of the large language model through a reinforcement learning framework.

6. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, The step of obtaining the target programming model by inputting the data on the labor shortage, constrained by equipment operating costs and labor costs, into a preset integer programming model, further includes: The labor shortage data is quantified into the workload requirements that need to be completed within the preset future time period; Based on the equipment information, determine the unit operating cost of each target device when it is activated; The cost of hiring a unit of human resources; The workload requirement, the unit operating cost, and the unit human resource hiring cost are used as parameters and substituted into the objective function and constraints of the integer programming model to generate the objective programming model.

7. The device dynamic scheduling method based on cultural event perception according to claim 1, characterized in that, The integer programming model is a mixed integer programming model. The objective function of the integer programming model is to minimize the total operating cost, which includes the start-up, shutdown, and operation costs of each target device and the cost of replenishing temporary human resources. The constraints of the integer programming model include at least the warehouse operation throughput demand constraint, the maximum number of target devices in operation constraint, and the upper limit constraint of temporary human resources employment.

8. A dynamic equipment scheduling device based on cultural event perception, characterized in that, The device includes: The acquisition module is used to acquire multi-source heterogeneous data related to cultural events in a specified target area; The generation module is used to extract and fuse features from the multi-source heterogeneous data to generate corresponding cultural event feature vectors. The prediction module is used to obtain historical employment data of a specified area and input the cultural event feature vector and the historical employment data into a pre-trained large-scale language model to predict the employment gap data in the future preset period. The input module is used to input the labor shortage data, constrained by equipment operating costs and labor costs, into a preset integer programming model to obtain a target programming model; wherein, the integer programming model contains preset equipment information of each target device in the specified target area; The solution module is used to solve the target planning model to obtain the optimal dynamic activation ratio of each target device; The execution module is used to generate a device scheduling scheme based on the optimal dynamic activation ratio and to perform scheduling execution.

9. A computer-readable storage medium, characterized in that, The device stores a computer program that, when executed by a processor, causes the processor to perform the steps of the device dynamic scheduling method based on cultural event awareness as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The device includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the device dynamic scheduling method based on cultural event awareness as described in any one of claims 1 to 7.

Images