Multi-dimensional cloud edge collaborative operation analysis method and system for dry bulk and general cargo wharf

Abstract

The present application relates to the technical field of wharf operation management, in particular to a dry bulk cargo and piece miscellaneous cargo wharf multi-dimensional cloud edge collaborative operation analysis method and system. It includes the following steps: data collection and industry and finance integration; multi-dimensional cost characteristic construction; double space decoupling intelligent benchmarking analysis; analysis result output and closed loop interaction. The present application uses double space sparse coding technology with operation mechanism quantity anchoring constraint to construct mechanism quantity portrait matrix and alternately correct with measured physical operation mechanism quantity as anchor point, realizing the binding of operation state coding and wharf real operation mechanism. The present application uses double space sparse coding technology with low correlation decoupling constraint to decouple cost characteristics into external condition coding and operation state coding, and divides comparable operation groups according to external condition coding similarity, realizing the fair benchmarking of operation efficiency after stripping off the interference of objective conditions and the accurate identification of management short board.

Description

Technical Field

[0001] This invention relates to the field of terminal operation and management technology, and more specifically, to a multi-dimensional cloud-edge collaborative operation analysis method and system for dry bulk cargo and general cargo terminals. Background Technology

[0002] Dry bulk and general cargo terminal operations encompass multiple stages, including loading and unloading, horizontal transportation, yard storage, and auxiliary operations. These operations continuously generate on-site operational data, business and financial data, and equipment physical measurement data. With the increasing digitalization and intelligence of ports, conducting terminal operation analysis and efficiency control based on multi-source data has become an important development direction in the field of port operation management.

[0003] In the field of port data processing and analysis, various data-driven technical solutions already exist. Invention patent CN202411329617.8 discloses a real-time port cargo monitoring method based on sparse representation. This method involves deploying a wireless signal transmitter and receiver outside the cargo container. The transmitter emits a specific frequency wireless signal into the container, and the receiver receives the signal transmitted after passing through the cargo. This allows for the acquisition of quantitative data on the radio wave obstruction effect of different cargo categories and cargo category labels. The method constructs a signal feature dictionary representing different cargo categories, with each category corresponding to a specific signal pattern. Sparse representation technology is then used to match the received signal features with the signal patterns in the dictionary, achieving automatic cargo category identification. Invention patent CN202610195903.2 discloses a port low-carbon scheduling method based on cloud-edge-device architecture and dual-dimensional consistency. The port control center calculates dynamic time-of-use electricity price and global power deviation signal based on the real-time operating status of the system through a nonlinear dynamic electricity price update model. The quay crane allocation unit responds to the dynamic time-of-use electricity price and generates a quay crane allocation scheme in the port area with the goal of minimizing the electricity cost of the quay crane group. The terminal equipment group includes generator group and quay crane operation group. The generator group and quay crane operation group are allocated in a dual-dimensional consistent distributed manner with generator cost increment factor and quay crane loading and unloading efficiency factor as consistency variables, respectively. The global power deviation signal is received by the main intelligent agent elected by each group, and symmetrical compensation is performed on their respective consistency variables to realize the coordinated and optimized scheduling of port low-carbon economy.

[0004] Although the above-mentioned technical solutions have their respective design advantages in their application scenarios, they also have the following technical defects: First, data-driven feature modeling and equipment physical operation mechanisms are separated, and feature representation lacks physical interpretability: The sparse representation and dictionary learning in CN202411329617.8 only perform pure data-driven pattern matching around wireless signal features. The physical meaning of dictionary atoms and codes only corresponds to the signal patterns of cargo categories, and no correlation constraint is established between feature representation and the actual physical operation mechanism of terminal equipment; Although the cloud-edge-device scheduling in CN202610195903.2 involves physical equipment control, its data processing and analysis dimensions are limited to power consumption and scheduling costs. It does not integrate the terminal operation cost features with the physical operation mechanism of equipment for modeling. Neither of the above two technical routes can obtain a physically interpretable representation of the operational status. Secondly, the interdependence of objective conditions and operational management levels in multidimensional cost characteristics makes fair benchmarking of operational efficiency impossible. CN202411329617.8 uses a single-space signal feature dictionary for matching and identification, failing to address the modeling approach of decomposing multidimensional features into different attribute spaces, thus failing to isolate the interference of external objective factors on feature representation. CN202610195903.2's scheduling optimization aims to minimize the electricity cost of the port's quay crane group, without considering the coupling relationship between objective conditions such as cargo type, ship type, and berth conditions and operational management levels in cost characteristics. Even after these complementary technical approaches, they still cannot support fair benchmarking of terminal operational efficiency and identification of management shortcomings while eliminating the interference of objective condition differences. Therefore, the key technical problem that existing technologies urgently need to solve is: how to achieve the fusion modeling of multi-source terminal operational data and equipment physical operation mechanisms under a cloud-edge collaborative architecture, and how to isolate the interference of objective condition differences in cost feature analysis to obtain physically interpretable operational status representations, thereby supporting fair benchmarking of terminal operational efficiency and identification of management shortcomings. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional cloud-edge collaborative operation analysis method and system for dry bulk and general cargo terminals, in order to solve the problems mentioned in the background art, such as the disconnect between data-driven feature modeling and equipment physical operation mechanism, the lack of physical interpretability of feature representation, and the coupling between objective condition differences and operation management level in multi-dimensional cost features, which makes it impossible to achieve fair benchmarking of operation efficiency.

[0006] To address the aforementioned technical problems, one objective of this invention is to provide a multi-dimensional cloud-edge collaborative operation analysis method for dry bulk and general cargo terminals, comprising the following steps: S1. Data collection and integration of business and finance: Real-time collection of terminal site operation data, business and financial data and equipment physical measurement data, and through on-site preprocessing at the edge and unified collection and mapping in the cloud, forming a standardized dataset at the operation unit level that includes cost attributes and terminal physical operation mechanism quantities. S2. Construction of multi-dimensional cost features: Based on the standardized dataset at the work unit level, through the division of work centers and the hierarchical calculation of cost drivers, a set of work unit-level monetary cost features covering the dimensions of departments, work links, and goods is constructed. S3. Dual-space decoupling intelligent benchmarking analysis: Based on the cost characteristics of work unit level and the physical operation mechanism quantity collected by the terminal production equipment, a dual-space sparse coding technique with low correlation decoupling constraints and operation mechanism quantity anchoring constraints is used to decouple cost characteristics and obtain the operation status code that is stripped of the influence of objective conditions; comparable work groups are divided according to the similarity of external condition codes, and operational efficiency evaluation and control shortcomings are completed within the group, outputting multi-dimensional intelligent benchmarking analysis results; S4. Analysis Result Output and Closed-Loop Interaction: Based on the multi-dimensional intelligent benchmarking analysis results, the cloud-based result encapsulation and edge-based synchronous feedback are used to generate operation optimization guidelines that can be distributed to the on-site operation terminal to achieve closed-loop optimization of terminal operation and management.

[0007] As a further improvement to this technical solution, in step S1, the data acquisition and business-finance integration and aggregation includes the following steps: S1.1 On-site operation data and equipment physical measurement data of wharf berths, loading and unloading equipment, transportation equipment, storage yards, and gate nodes are collected locally by edge computing units deployed at each node, and data cleaning, outlier removal, and operation cycle time window aggregation processing are performed to generate standardized operation data slices at the edge. S1.2. For the standardized operation data slices uploaded by each edge node, as well as the business and financial data output by the business production system and financial accounting system, the fields are mapped and associated in the cloud with the operation unit identifier as the primary key to establish the corresponding mapping relationship between operation unit, resource consumption and cost item. S1.3 Perform consistency verification and missing value marking on the multi-source data after association mapping, generate a standardized dataset at the operation unit level that simultaneously contains monetary cost attributes and physical quantity terminal physical operation mechanism quantities, and store it in the cloud data layer.

[0008] As a further improvement to this technical solution, the construction of multi-dimensional cost features in step S2 includes the following steps: S2.1. For the standardized dataset at the work unit level, the work centers are divided into four categories based on the terminal operation process: loading and unloading operations, horizontal transportation, yard storage, and auxiliary operations. The work unit identifier is used as the matching key to complete the matching of work center affiliation and determine the cost collection object of each work unit. S2.2 For cost data collected to each work center, a two-level allocation calculation logic of resource driver and work driver is adopted. First, the resource layer cost is allocated to the corresponding work center according to the resource driver. Then, the total cost of the work center is allocated to each subordinate work unit according to the work driver, thus completing the cost quantification calculation at the work unit level. S2.3. Classify and collect the quantitative cost data of the work unit according to three dimensions: department, work process, and type of goods, and construct a multi-dimensional work unit-level monetary cost feature set, which is then output to S3.

[0009] As a further improvement to this technical solution, the dual-space decoupling intelligent benchmarking analysis in S3 includes the following steps: S3.1 Using the cost feature set of work unit level as training sample, construct an external condition dictionary and an operation status dictionary that respectively carry objective conditions and operation status. Incorporate cost feature reconstruction error, dual-space sparse regularization term and low correlation decoupling constraint term into joint optimization objective for iterative training to obtain converged dual-space dictionary parameters. S3.2. Based on the activation correspondence between the dictionary atoms of each operational state in the training set and the physical operation mechanism quantities of the samples, the initial value of the mechanism quantity profile matrix is ​​generated statistically, and the profile matrix is ​​alternately corrected using the measured physical operation mechanism quantities as anchor points to obtain a stable mechanism quantity profile matrix. S3.3 For a single work unit, its monetary cost characteristics and physical work mechanism quantity are used as the encoding input. The converged external condition dictionary, operation status dictionary and stable mechanism quantity profile matrix are called. The dual-space sparse coding with low correlation decoupling constraints and work mechanism quantity anchoring constraints is used to solve the problem and output the external condition code and operation status code of the work unit. S3.4. Based on the similarity of the external condition codes of each work unit, divide the work into comparable work groups. Within each group, conduct operational efficiency evaluation and identify control shortcomings based on the operational status code, and output multi-dimensional intelligent benchmarking analysis results.

[0010] As a further improvement to this technical solution, S3.1, the dual-space dictionary construction and low-correlation constraint training, includes the following steps: S3.11. Generate initial atomic sets for the external condition dictionary and the operation status dictionary using a random initialization method. The external condition dictionary is used to carry objective condition information such as cargo type, ship type, work batch, and berth conditions, while the operation status dictionary is used to carry operation status information such as scheduling coordination, equipment organization, and resource allocation efficiency. S3.12. Using the cost feature set of the work unit level as the training input, solve the external condition code and the operation status code of each training sample in each iteration. Based on the joint optimization objective composed of cost feature reconstruction error, dual-space sparse regularization term and low correlation decoupling constraint term, alternately update the atomic parameters of the external condition dictionary and the operation status dictionary. S3.13. Monitor the iterative change of the joint optimization objective. When the cost feature reconstruction error drops to a preset threshold and the change of the joint optimization objective for a consecutive preset number of rounds is less than the preset change threshold, stop iterative training and output the converged external condition dictionary and operation status dictionary parameters.

[0011] As a further improvement to this technical solution, S3.2, the construction of the mechanistic quantitative profile and the measured anchoring correction, includes the following steps: S3.21. Calculate the average physical operation mechanism quantity of the corresponding sample when each operation state dictionary atom is activated in the training set, and generate the initial value of the mechanism quantity portrait matrix; for operation state dictionary atoms with fewer than the preset threshold, fill their portrait vector with the average physical operation mechanism quantity of the global training set. S3.22. Fix the external condition dictionary, the operational state dictionary, and the external condition encoding and operational state encoding of each training sample. Using the measured physical operation mechanism quantity as the anchor point, update the mechanism quantity profile matrix along the gradient direction that reduces the mechanism quantity reconstruction error. The update method satisfies: ; in, For the first The updated mechanism profile matrix For the first Mechanistic profile matrix of the wheel. Update the step size for the portrait matrix. The training sample is a vector of measured physical operation mechanism quantities. For the first The operational status code obtained from the first round of solving. For the first Transposition of wheel operation status code For matrix multiplication operations; after updating the fixed mechanism quantity profile matrix, an optimization objective including cost feature reconstruction term, dual-space sparse regularization term, low correlation decoupling constraint term and operation mechanism quantity anchoring constraint term is adopted to resolve the external condition code and operation status code of each training sample, and the profile matrix update and code solving are performed alternately; S3.23. Monitor the iterative change of the mechanism profile matrix and the mechanism reconstruction error. When the change of the mechanism profile matrix in two adjacent rounds is less than the preset threshold and the mechanism reconstruction error continues to decrease, stop the alternating correction and output a stable mechanism profile matrix.

[0012] As a further improvement to this technical solution, in S3.3, the constrained dual-space sparse coding includes the following steps: S3.31. Obtain the monetary cost feature vector and the measured physical operation mechanism vector of a single work unit, call the converged external condition dictionary, operation status dictionary and stable mechanism profile matrix to complete the encoding input preparation; S3.32. Construct a dual-space sparse coding optimization objective function, which sequentially includes a cost feature reconstruction term, a dual L1 sparse regularization term for external condition coding and operational status coding, a low-correlation decoupling constraint term for the two coding crossovers, and an anchoring constraint term for the operational status coding mapped by the mechanism quantity profile matrix and the measured physical operation mechanism quantity. The optimization objective function satisfies: ; in, This represents the monetary cost feature vector for a single work unit. For external condition dictionary, For the operational status dictionary, Encoding for external conditions Coded for operational status, and and Encoded vectors of the same dimension L1 sparse regularization coefficients encoded for external conditions L1 sparse regularization coefficients for encoding operational status. For low-correlation decoupling constraint coefficients, The matrix represents the mechanistic quantity profile, where m is the vector of measured physical mechanistic quantities for each work unit. Anchoring constraint coefficients for operational mechanism quantities. The square of the L2 norm of the vector. The L1 norm of the vector; S3.33 Output the external condition code and operational status code of the work unit for subsequent comparable group division and operational efficiency evaluation.

[0013] As a further improvement to this technical solution, S3.4, the comparable group division, operational efficiency evaluation, and weakness identification, includes the following steps: S3.41 Calculate the cosine similarity of the external condition codes of each work unit, and classify work units with similar objective conditions into the same comparable work group based on the similarity threshold. S3.42. Select the optimal set of work units in terms of unit cost from the set of cost features at the work unit level. Sort the atoms in the operation status dictionary in the optimal set of work units according to their activation frequency. Mark the atoms with activation frequency higher than the preset high frequency threshold as high-efficiency atoms and the atoms with activation frequency lower than the preset low frequency threshold as low-efficiency atoms. Calculate the operation efficiency score of each work unit based on the activation ratio of high-efficiency atoms and low-efficiency atoms in the operation status code. S3.43. For work units whose operational efficiency scores are lower than the group's baseline, analyze the resource consumption patterns corresponding to the inefficient atoms activated in their operational status codes and identify the corresponding control bottleneck types. S3.44. Integrate the efficiency ranking, weakness distribution and optimization direction within each comparable work group to generate multi-dimensional intelligent benchmarking analysis results, and transfer them to S4.

[0014] As a further improvement to this technical solution, in step S4, the analysis result output interacts with the closed loop, including the following steps: S4.1. The results of multi-dimensional intelligent benchmarking analysis are structured and encapsulated to generate operational optimization guidelines that include operational efficiency evaluation, types of management shortcomings and optimization directions, and stored in the cloud analysis results library; S4.2 The cloud will synchronously distribute the operation optimization guidelines to the edge computing units of the corresponding dock operation nodes, and the edge computing units will push them to the on-site operation terminals. S4.3. The edge computing unit collects the execution feedback data of the operation optimization guidelines from the field operation terminal and sends it back to the cloud. The cloud updates the operation analysis parameters based on the feedback data to realize closed-loop optimization of terminal operation management.

[0015] The second objective of this invention is to provide a multi-dimensional cloud-edge collaborative operation analysis system for dry bulk and general cargo terminals, used to execute the steps of the aforementioned multi-dimensional cloud-edge collaborative operation analysis method for dry bulk and general cargo terminals, including: The data acquisition and business-finance integration and aggregation unit is equipped with edge data access interfaces deployed at various nodes of the terminal and cloud business-finance data access interfaces. The built-in edge preprocessing logic performs on-site preprocessing on the received terminal site operation data and equipment physical measurement data, and completes the multi-source data association and alignment through cloud aggregation and mapping logic, outputting a standardized dataset at the operation unit level that includes cost attributes and terminal physical operation mechanism quantities. The cost feature construction and multi-dimensional aggregation unit loads a standardized dataset at the work unit level, has built-in work center division logic to determine the cost aggregation object of each work unit, and completes the cost quantification at the work unit level through built-in cost driver hierarchical calculation logic, and aggregates and outputs a set of monetary cost features at the work unit level according to the dimensions of department, work link, and type of goods. The dual-space decoupling and intelligent benchmarking analysis unit is equipped with a feature input interface to receive the cost features of the operation unit level and the physical operation mechanism quantity collected by the terminal production equipment. It has a built-in dual-space sparse coding logic with low correlation decoupling constraints and operation mechanism quantity anchoring constraints to decouple the cost features, obtain the operation status code stripped of the influence of objective conditions, and divide the comparable operation groups according to the similarity of external condition codes. Within the group, the operation efficiency evaluation and control shortcomings identification are completed, and multi-dimensional intelligent benchmarking analysis results are output. The result encapsulation and closed-loop interaction unit receives multi-dimensional intelligent benchmarking analysis results, generates operation optimization guidelines that can be sent to the on-site operation terminal through the cloud-built result encapsulation logic, and completes on-site execution feedback collection through the edge terminal synchronous distribution and feedback transmission link, thereby realizing closed-loop optimization of terminal operation management.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a mechanism profile matrix by using a dual-space sparse coding technique with anchoring constraints of operational mechanism quantities. It alternately corrects the matrix using measured physical operational mechanism quantities as anchor points, which solves the problem of the separation between data-driven feature modeling and equipment physical operational mechanisms. This achieves the binding of operational status coding with the actual operational mechanisms of the terminal, and improves the physical interpretability of feature representation. 2. This invention uses a dual-space sparse coding technique with low correlation decoupling constraints to decouple cost features into external condition codes and operational status codes. It also divides comparable work groups according to the similarity of external condition codes, which solves the problem of the mutual coupling between objective condition differences and operational management level in cost features. This enables fair benchmarking of operational efficiency and accurate identification of control shortcomings after removing the interference of objective conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall steps of the multi-dimensional operation analysis method of the present invention; Figure 2 This is a schematic diagram of the overall structural framework of the multi-dimensional operation analysis system of the present invention; The meanings of the labels in the diagram are as follows: 1. Data collection and business-finance integration unit; 2. Cost characteristic construction and multi-dimensional aggregation unit; 3. Dual-space decoupling and intelligent benchmarking analysis unit; 4. Result encapsulation and closed-loop interaction unit. Detailed Implementation

[0018] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, this embodiment provides a multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals, including: S1. Data Acquisition and Business-Finance Integration: Real-time acquisition of terminal site operation data, business and financial data, and equipment physical measurement data. This data is pre-processed on-site at the edge and uniformly aggregated and mapped in the cloud to form a standardized dataset at the operation unit level, containing cost attributes and terminal physical operation mechanism quantities. An operation unit refers to the smallest granularity object for cost aggregation and efficiency analysis in terminal operation management, specifically a complete loading / unloading batch or a single operation task for a certain type of cargo at a certain berth. The operation unit identifier is a field or combination of fields that uniquely identifies an operation unit. Physical operation mechanism quantities refer to physical quantity data directly collected from terminal production equipment, reflecting the physical operation process, energy consumption, and efficiency, including operating tonnage, equipment hours, transportation distance, and energy consumption readings. Monetary cost attributes refer to data representing the cost composition of the operation unit in monetary terms.

[0020] S1.1 On-site operation data and equipment physical measurement data of wharf berths, loading and unloading equipment, transportation equipment, storage yards, and gate nodes are collected locally by edge computing units deployed at each node, and data cleaning, outlier removal, and operation cycle time window aggregation processing are performed to generate standardized operation data slices at the edge. Specifically, edge computing units are deployed at wharf berths, loading and unloading equipment, transportation equipment, storage yards, and gate nodes. These edge computing units connect to sensors, programmable logic controllers (PLCs), equipment operation monitoring systems, or work management terminals at their respective nodes to collect real-time on-site operational data and equipment physical measurement data. On-site operational data includes start and end times, operation type, cargo type, equipment number, work team, and workload. Equipment physical measurement data includes equipment runtime, cumulative energy consumption readings, number of lifting actions, number of traveling actions, number of slewing actions, load capacity, cumulative belt conveyor volume, and stacker-reclaimer travel distance.

[0021] Furthermore, the edge computing unit performs data cleaning, outlier removal, and job cycle time window aggregation on the collected raw data, as follows: Data cleaning includes removing duplicate records, standardizing data formats, and eliminating records with missing key fields. Missing key fields refer to records lacking job unit identifiers or timestamps. Records lacking non-key fields are retained for unified processing in S1.3.

[0022] Outlier removal is performed according to the following rules: For each type of equipment physical measurement data, a reasonable range of values ​​is pre-defined. Reasonable value ranges should be determined primarily based on equipment nameplate parameters or design specifications; if nameplate parameters are missing, the 1st and 99th percentiles of historical operating data should be used; if neither of these is available, the value should be determined with reference to port operation quota standards and stored in the configuration parameters of the edge computing unit. If the value of this type of physical measurement data in a certain record... satisfy or If the value exceeds 1.2 times the rated lifting capacity of the gantry crane, it is considered an outlier and the record is removed. For example, if the recorded value of the single lifting load of the gantry crane exceeds 1.2 times the rated lifting capacity, it is considered an outlier and is removed; if the cumulative amount of the belt conveyor is negative within the time window and the absolute value exceeds the preset measurement error tolerance, it is considered an outlier and is removed.

[0023] The work cycle time window aggregation process defines the work cycle time window by using the start and end events corresponding to the work unit identifier as boundaries. Within the time window, continuously collected equipment physical measurement data are aggregated and calculated. Let... This is the aggregated value of the physical operation mechanism quantity. For cumulative physical quantities, including cumulative energy consumption readings and cumulative belt conveyor output, the aggregated value is the difference between the cumulative value at the end of the time window and the cumulative value at the beginning of the time window. ,in This is the cumulative value at the end of the time period. This is the cumulative value at the start time. For instantaneous physical quantities, including motor current and instantaneous power, representative values ​​are taken according to the following rules: for energy-related quantities, the average value within the time window is taken. ,in Number of points collected within the time window For the first The instantaneous values ​​at each sampling point; for load-related measurements, the maximum value within the time window is taken. If no data is collected within the time window, the corresponding physical mechanism quantity field of the work unit is marked as missing and will be handled uniformly in S1.3.

[0024] Ultimately, the aggregated data forms standardized operation data slices at the edge. Each record corresponds to one operation cycle of a work unit, and the fields include work unit identifier, operation type, type of goods, equipment number, operation time range, and aggregated physical operation quantity values. The edge computing unit uploads the standardized operation data slices to the cloud via message queues.

[0025] S1.2. For the standardized operation data slices uploaded by each edge node, as well as the business and financial data output by the business production system and financial accounting system, the fields are mapped and associated in the cloud with the operation unit identifier as the primary key to establish the corresponding mapping relationship between operation unit, resource consumption and cost item. Specifically, each edge computing unit uploads standardized operational data slices to the cloud server, and the business and financial data output by the business production system and the financial accounting system are also aggregated to the cloud. The business and financial data includes the direct material costs, direct labor costs, equipment depreciation, energy consumption costs, maintenance costs, and allocated indirect costs corresponding to the operational unit, and includes dimensional information such as operational unit identification, department code, operational process code, and product type code.

[0026] Furthermore, the cloud uses the job unit identifier as the primary key to perform field mapping and association alignment between standardized job data slices and business financial data. A predefined field mapping table unifies the naming of fields with different names in the business production system and the financial accounting system to standard field names. Specifically, the "job order number" in the business production system and the "cost object number" in the financial accounting system are both mapped to "job unit identifier," and the "job quantity" in the business production system is mapped to the standard field "physical job quantity." During association alignment, the job unit identifier in the standardized job data slice is used as the benchmark to retrieve records with the same job unit identifier in the business financial data, and the fields of both are merged. If multiple cost items in the business financial data correspond to the same job unit, they are associated as different fields according to the cost item category, forming a corresponding mapping relationship of job unit-resource consumption-cost item, clearly defining the cost item corresponding to each resource consumption. This mapping relationship is stored as a mapping table, serving as the basic configuration of the cloud data layer.

[0027] S1.3 Perform consistency verification and missing value marking on the multi-source data after association mapping, generate a standardized dataset at the operation unit level that simultaneously contains monetary cost attributes and physical quantity terminal physical operation mechanism quantities, and store it in the cloud data layer.

[0028] Specifically, consistency checks are performed on the multi-source data after association mapping, including: checking whether the work unit identifier exists and is consistent in both the standardized work data slice and the business financial data; checking whether the total cost amount is consistent with the total cost in the business financial data; checking whether the dimensions of the physical work mechanism quantity are consistent; and checking whether the work time range overlaps or conflicts.

[0029] Furthermore, inconsistent data will be processed according to the following preset rules: if the work unit identifier is inconsistent, the record will be marked as unmatched data and will not be included in the subsequent cost feature construction; if the total cost amount is inconsistent, the data output by the financial accounting system will be used for correction and a correction log will be recorded; if the units are inconsistent, they will be converted uniformly according to the preset unit conversion rules; if the work time range overlaps or conflicts, the time window will be re-determined based on the original timestamps of the work start event and work end event, and the corresponding physical quantities will be re-aggregated.

[0030] Regarding missing value marking, for fields that remain missing after association, a missing value marker and a missing reason code are added to the data record, along with the metadata information of the missing field. Data with missing value marking remains in the dataset, and subsequent steps select to skip it or fill it with the default value based on the missing value marker. Specifically, work units missing key fields such as work unit identifier, total cost, and work quantity are marked as "unavailable" and are not included in S2 cost allocation and S3 analysis; when non-key fields are missing, they are filled with the average value of the corresponding fields of the same work unit with the same work process code and the same type of goods, and the filling marker is recorded. Missing value processing is performed before S2.1 work center attribution matching.

[0031] The above processing generates a standardized dataset at the operation unit level, containing both monetary cost attributes and physical quantity terminal physical operation mechanism quantities. Each record in this dataset corresponds to an operation unit and includes an identification field (operation unit identifier), cost attribute fields (amount of each cost item), physical quantity fields (various physical operation mechanism quantities), dimension fields (department code, operation link code, cargo type code), and data quality fields (consistency check result, missing data flag). This standardized dataset is stored in the cloud data layer as input for constructing S2 multi-dimensional cost features.

[0032] S2. Construction of Multi-Dimensional Cost Features: Based on the standardized dataset at the work unit level, a set of monetary cost features at the work unit level is constructed by dividing work centers and calculating cost drivers in a hierarchical manner, covering the dimensions of departments, work links, and cargo types. In this embodiment, a work center refers to an intermediate collection level in the terminal operation process that performs similar work activities and applies the same cost collection logic; a resource driver refers to the quantitative basis for allocating resource layer costs to each work center; and a work driver refers to the quantitative basis for allocating the collection costs of work centers to their subordinate work units.

[0033] S2 specifically includes the following steps: S2.1. For the standardized dataset at the work unit level, the work centers are divided into four categories based on the terminal operation process: loading and unloading operations, horizontal transportation, yard storage, and auxiliary operations. The work unit identifier is used as the matching key to complete the matching of work center affiliation and determine the cost collection object of each work unit. Specifically, based on the stage attributes of the terminal operation process, all operational activities are divided into four types of operation centers: loading and unloading operations, horizontal transportation, yard storage, and auxiliary operations. Among them: The loading and unloading operation center covers cargo lifting, transshipment, loading and unloading operations performed by equipment such as gantry cranes and quay cranes at the berth front; The horizontal transport operations center covers the transport operations of goods from berth to yard or from yard to berth carried out by container trucks and belt conveyor systems within the port. The stockyard storage operations center covers the storage, retrieval, and repacking operations performed by stacker-reclaimers and stockyard cranes; The auxiliary operations center covers support operations beyond the three types of operations centers mentioned above, including mooring and unmooring, tank cleaning, weighing, and tallying.

[0034] Furthermore, based on the aforementioned division of work centers, a mapping table between work type codes and work center categories is pre-established in the cloud and stored in the cloud data layer. For each record in the standardized dataset at the work unit level, using the work unit identifier as the matching key, the work type field in the record is read, and the work center category to which the work unit belongs is determined according to the mapping table. The work center category field is then written into the record. In summary, each work unit is assigned to a unique work center, which is the cost aggregation object for that work unit.

[0035] S2.2 For cost data collected to each work center, a two-level allocation calculation logic of resource driver and work driver is adopted. First, the resource layer cost is allocated to the corresponding work center according to the resource driver. Then, the total cost of the work center is allocated to each subordinate work unit according to the work driver, thus completing the cost quantification calculation at the work unit level. Specifically, resource-layer costs originate from the cost attribute fields in the standardized dataset at the work unit level generated in S1.3, including direct material costs, direct labor costs, equipment depreciation, energy costs, maintenance costs, and allocated indirect costs. These cost items are categorized into five resource costs: labor costs, equipment depreciation, energy costs, maintenance costs, and material costs. Direct labor costs correspond to labor costs, equipment depreciation to equipment depreciation, energy costs to energy costs, maintenance costs to maintenance costs, and direct material costs to material costs. Allocated indirect costs are categorized according to their nature: indirect labor costs are categorized under labor costs, indirect energy costs under energy costs, and other indirect costs are categorized under their corresponding resource cost items based on the beneficiary. Let... Indicates the first The total cost of each resource item , Let this be the total number of resource cost items. Indicates the total number of work centers. Indicates the first The total cost allocated to each work center These correspond to loading and unloading operation center, horizontal transportation operation center, yard storage operation center, and auxiliary operation center, respectively.

[0036] Furthermore, the first level of allocation is the distribution of resource-level costs to activity centers. For each resource cost... The resource drivers and their consumption amounts across the four work centers are pre-determined. The determination methods and data sources for resource drivers are as follows: Labor costs are driven by the man-hours at each work center, taken from the shift attendance records in the business production system; equipment depreciation costs are driven by the machine-hours at each work center, taken from the equipment runtime field in the S1.1 standardized work data slice; energy costs are driven by the energy consumption metering values ​​at each work center, taken from the cumulative energy consumption readings in the S1.1 standardized work data slice; maintenance costs are driven by the maintenance work order hours at each work center, taken from the maintenance work order records in the business production system; and material costs are driven by the material requisition amounts at each work center, taken from the material requisition records in the business production system. All the above resource driver consumption amounts are aligned and mapped using the fields in S1.2, and then stored in the cloud data layer along with the standardized dataset in S1.3. For resource driver consumption data collected by work unit (such as equipment hours and energy consumption metering values), the work unit-level consumption is summed according to the work center affiliation determined in S2.1. For resource driver consumption data recorded by department or work group (such as labor hours, maintenance work order hours, and material requisition), it is aggregated to the corresponding work center based on the correspondence between department / work group and work center. This yields the resource driver consumption data for each work center. .set up Indicates the first Resource costs in the first The resource driver consumption of each work center, then the first The first work center was allocated the first The resource cost is: ; in For the first The total consumption of a resource driver across all work centers. If If a resource has no consumption record in the current period, then the resource cost item will not participate in the allocation, and the reason will be recorded in the allocation log. The total cost allocated to each work center is: .

[0037] Furthermore, the second level of allocation is the distribution of activity center costs to activity units. For the... Each work unit under each work center predetermines its work motivation. The determination method and data source for work motivation are as follows: the work motivation of the loading and unloading work center and the yard storage work center is the work tonnage, taken from the work quantity field in the S1.1 standardized work data slice; the work motivation of the horizontal transport work center is the transport ton-kilometer, obtained by multiplying the work quantity in tons by the transport distance in kilometers, where the work quantity is taken from the work quantity field in the S1.1 standardized work data slice, and the transport distance is taken from the physical mechanism quantity field in the S1.1 standardized work data slice; the work motivation of the auxiliary work center is the work batch quantity, taken from one work task corresponding to the work unit. In this embodiment, for the sake of simplicity, the work batch quantity is used as the unified work motivation of the auxiliary work center; in actual applications, the work units under the auxiliary work center can be further subdivided into work motivations according to the auxiliary work type, such as mooring and unmooring operations by vessel number, tank cleaning operations by hatch number, and weighing and tallying operations by batch quantity. The subdivided motivation quantities are still calculated according to the above-mentioned second-level allocation formula. Let... For the first The collection of all work units under each work center. For the first The work-related factors of each work unit Then the first Cost allocated to each work unit for: ; in For the first The total number of work drivers for all work units under a work center. If If a work center has no activity driver records for the current period, its costs are not allocated to work units; instead, they remain at the work center level and are marked as unallocated costs, with an allocation log recorded. Unallocated costs are added to the total work center costs for the current period when the work center generates activity driver records in subsequent accounting periods, and then allocated a second time according to the second-level allocation formula described above. If... If it is an empty set, that is, the work center has no subordinate work units in the current period, the same treatment method as above applies when the total number of work drivers is zero.

[0038] Through the above two-level allocation, each work unit obtains a quantified cost value. The cost quantification calculation at the work unit level is completed. The allocation results and allocation logs are stored together in the cloud data layer.

[0039] S2.3. Classify and collect the quantitative cost data of the work unit according to three dimensions: department, work process, and type of goods, and construct a multi-dimensional work unit-level monetary cost feature set, which is then output to S3.

[0040] Specifically, in S2.2, the quantified cost values ​​for each work unit are obtained. Based on this, using each work unit record as the basic row, the quantitative cost data is categorized and collected according to three dimensions: department, work process, and commodity type. The department dimension is taken from the department code field in the S1.3 standardized dataset, the work process dimension is taken from the work center category field determined in S2.1, and the commodity type dimension is taken from the commodity code field in the S1.3 standardized dataset. After collection, each work unit record simultaneously carries three dimension labels: department code, work process category, and commodity type code, along with the corresponding cost value. .

[0041] Furthermore, for each work unit record, its cost value is merged with the physical mechanism quantity field in the S1.3 standardized dataset to form a work unit-level monetary cost feature vector. The numerical cost feature components participating in the S3 dictionary encoding operation in this feature vector include: the allocated amount of each cost item and the total cost; the work unit identifier, department code, work process category, and cargo type code are stored as dimension label fields along with each feature vector, used for dimensional analysis after S3.4 grouping and correlation with S4 results, and do not participate in the matrix operations of dictionary encoding; the values ​​of various physical operation mechanism quantities are used as independent physical operation mechanism quantity vectors for S3 mechanism quantity anchoring constraints. The feature vectors of all work units constitute a work unit-level monetary cost feature set, stored in the cloud data layer, and used as input for S3 dual-space decoupled intelligent benchmarking analysis.

[0042] S3. Dual-space decoupling intelligent benchmarking analysis: Based on the cost characteristics of work unit level and the physical operation mechanism quantity collected by the terminal production equipment, a dual-space sparse coding technique with low correlation decoupling constraints and operation mechanism quantity anchoring constraints is used to decouple cost characteristics and obtain the operation status code that is stripped of the influence of objective conditions; comparable work groups are divided according to the similarity of external condition codes, and operational efficiency evaluation and control shortcomings are completed within the group, outputting multi-dimensional intelligent benchmarking analysis results; Specifically, the cost of a terminal operation unit is determined by two factors: objective conditions and operational status. Objective conditions include cargo type, vessel type, throughput, and berth conditions, which are external factors beyond the control of the operation unit itself. Operational status includes scheduling and coordination, equipment organization, and resource allocation efficiency, which are internal factors that can be improved through management. Both types of factors are reflected in the cost data of the operation unit. If direct horizontal benchmarking is performed based on cost data, differences in objective conditions can interfere with the judgment of operational management level. An operation unit with superior objective conditions but extensive operational management may have lower costs than an operation unit with unfavorable objective conditions but refined operational management. The benchmarking conclusions drawn from this are unlikely to reflect true operational efficiency. Traditional benchmarking methods usually group by a single dimension or directly compare absolute cost values, and have limited ability to handle situations where multiple objective conditions have overlapping influences.

[0043] To address the aforementioned issues, S3 employs dual-space sparse coding to decouple cost features. A low-correlation decoupling constraint forces the external condition coding and operational status coding to be independent in the representation space, preventing leakage of objective condition information and operational status information between the two dictionaries. A mechanism anchoring constraint maps the operational status coding to the physical mechanism space via a mechanism profile matrix, using measured physical quantities as anchors to verify the physical consistency of the operational status coding, providing a verification basis independent of cost data for the decoupling results. After decoupling, comparable work groups are formed based on the similarity of external condition coding, ensuring that work units within a group are at similar levels in terms of the comprehensive impact of multi-dimensional objective conditions. Differences in efficiency scores within a group can be attributed to differences in operational status. Furthermore, based on the activation frequency of operational status dictionary atoms in the optimal work set, efficient and inefficient atoms are labeled, and the operational efficiency score of each work unit is calculated based on the coding amplitude to identify management shortcomings. Thus, this embodiment, through the decoupling, grouping, and benchmarking link, avoids the one-sidedness of single-dimensional grouping and the distortion problems of absolute cost comparison, ensuring that the operational efficiency evaluation results are not affected by differences in objective conditions.

[0044] In this embodiment, the external condition dictionary refers to the basis matrix stored in the form of atomic sets for sparsely representing the objective condition information of the work unit; the operation status dictionary refers to the basis matrix stored in the form of atomic sets for sparsely representing the operation management status of the work unit; the external condition code refers to the sparse representation vector of the work unit on the external condition dictionary; the operation status code refers to the sparse representation vector of the work unit on the operation status dictionary; the mechanism quantity portrait matrix refers to the matrix that maps the atoms of the operation status dictionary to the physical mechanism quantity space; activation refers to the absolute value of a certain component in the encoding vector exceeding a preset non-zero judgment threshold.

[0045] Furthermore, before entering S3.1, the cost feature set at the work unit level is preprocessed: the cost feature vector is standardized by dimension, so that the mean of each dimension is 0 and the standard deviation is 1; the physical operation mechanism vector is normalized by dimension to [0,1]. The mean, standard deviation, minimum, and maximum values ​​of each dimension calculated during standardization and normalization are stored in the cloud data layer along with the convergence dictionary obtained from S3.1 training and the stable mechanism profile matrix obtained from S3.2 correction; during online encoding in S3.3, the same standardization / normalization parameters are called to transform the features of the work unit to be analyzed, ensuring that the feature space of training and inference is consistent. Number of training samples Should meet ,in This represents the number of atoms in the dictionary. If the actual number of samples is insufficient, historical data will be merged or the data will be allowed to accumulate.

[0046] S3.1 Using the cost feature set at the work unit level as training samples, construct an external condition dictionary and an operational status dictionary that respectively carry objective conditions and operational status. Incorporate cost feature reconstruction error, dual-space sparse regularization term, and low-correlation decoupling constraint term into the joint optimization objective for iterative training to obtain converged dual-space dictionary parameters; including the following steps: S3.11. Generate initial atomic sets for the external condition dictionary and the operation status dictionary using a random initialization method. The external condition dictionary is used to carry objective condition information such as cargo type, ship type, work batch, and berth conditions, while the operation status dictionary is used to carry operation status information such as scheduling coordination, equipment organization, and resource allocation efficiency. Specifically, numerical cost feature components are extracted from the set of monetary cost features at the work unit level generated in S2.3, assuming the number of training samples is... Each cost feature vector has a dimension of 1. Let's define an external condition dictionary. Operational Status Dictionary ,in The number of atoms in the dictionary. exist Select from the range, example: In the low-correlation decoupling constraint term, the vector inner product operation requires the two encoded vectors to have the same dimension. The initial atomic sets of the two dictionaries are generated using a random initialization method: external condition dictionary. Each atomic component independently has a mean of 0 and a standard deviation of Sampling is performed on a normal distribution, and after sampling, each atom is normalized using the L2 norm so that the modulus of each atom is 1; Operational state dictionary The initial atoms are generated in the same way. The purpose of atom normalization is to eliminate the scale interference of atomic modulus on the encoding solution, and to keep the constraint strength of the sparse regularization term consistent among all atoms.

[0047] Furthermore, the semantic division of labor between the external condition dictionary and the operational status dictionary is implemented as follows: During dictionary initialization, both dictionaries are randomly generated without pre-defined semantic labels; during training, low-correlation decoupling constraints force the representation spaces of the two dictionaries to be independent. After training convergence, dictionary semantic verification and correction are performed: using existing objective condition fields such as cargo type and ship type in the standardized dataset, each category's objective condition field is one-hot encoded, and the Pearson correlation coefficients between the external condition code and the operational status code and each one-hot encoded field are calculated respectively; for each category field, the maximum absolute value of the correlation coefficients of all one-hot dimensions is taken as the correlation index of that field, and the mean of the correlation indices of all objective condition fields is taken as the comprehensive correlation index. If the comprehensive correlation index of the operational status code is higher than that of the external condition code, the semantic division of labor is reversed, and the two dictionaries and their encoding roles are swapped. This step ensures that the external condition dictionary carries objective condition information and the operational status dictionary carries operational status information.

[0048] S3.12. Using the cost feature set of the work unit level as the training input, solve the external condition code and the operation status code of each training sample in each iteration. Based on the joint optimization objective composed of cost feature reconstruction error, dual-space sparse regularization term and low correlation decoupling constraint term, alternately update the atomic parameters of the external condition dictionary and the operation status dictionary. Specifically, let the first The cost feature vector of each training sample is Its external condition is encoded as The operating status code is The joint optimization objective during the training phase is: ; in , ; L1 sparse regularization coefficients encoded for external conditions L1 sparse regularization coefficients for encoding operational status. The decoupling constraint coefficients are low-correlation. Cost characteristic reconstruction term. Measure the degree to which a linear combination of two dictionaries restores the original cost feature vector; L1 sparse regularization term. and Constrain the number of non-zero components in each of the two codes separately; low correlation decoupling constraint term The square of the inner product of two encoding vectors is constrained to approach zero to prevent objective condition information and operational status information from leaking between dictionaries.

[0049] Furthermore, iterative training is performed using the following alternating update strategy: First, fix the current dictionary. and For all The encoding is solved for each training sample. and The encoding solution employs proximal gradient descent: for each training sample, a fixed... renew Then fix renew Execute alternately. To update For example, first calculate the objective function with respect to... The gradient includes the gradients of the reconstruction error term and the low-correlation term. Then, a soft thresholding operator is applied to the L1 term for near-end update. The convergence condition for the encoding solution is the change in encoding between two adjacent rounds. Or it can reach the maximum number of iterations, 100.

[0050] Secondly, fix all the solved codes and update the external condition dictionary using gradient descent. Let the residuals be... External condition dictionary The gradient of each atom is Update step size The example value is 0.01. After the update, L2 normalization is performed on this atom.

[0051] Then, fixing all the solved codes, update the operation status dictionary in the same way, and the operation status dictionary is updated to the [number]th [number]. The gradient of each atom is Update step size same After updating, normalization is performed.

[0052] Finally, the objective function value for this round of joint optimization is calculated.

[0053] In addition, the coefficients in this embodiment , Example value: 0.1 The example value is 0.05, and all values ​​are unrestricted.

[0054] S3.13. Monitor the iterative change of the joint optimization objective. When the cost feature reconstruction error drops to a preset threshold and the change of the joint optimization objective for a consecutive preset number of rounds is less than the preset change threshold, stop iterative training and output the converged external condition dictionary and operation status dictionary parameters.

[0055] Specifically, the average reconstruction error is calculated after each iteration. Preset reconstruction threshold , To reconstruct the threshold scaling factor (0.05 in the example), the joint optimization objective function value is also recorded. When both conditions are met and continuous wheel When the time is right, stop training and output the convergence dictionary. and Then, semantic verification and correction in S3.11 are performed.

[0056] S3.2. Based on the activation correspondence between the dictionary atoms of each operational state in the training set and the physical operation mechanism quantities of the samples, the initial value of the mechanism quantity profile matrix is ​​statistically generated, and the profile matrix is ​​alternately corrected using the measured physical operation mechanism quantities as anchor points to obtain a stable mechanism quantity profile matrix; including the following steps: S3.21. Calculate the average physical operation mechanism quantity of the corresponding sample when each operation state dictionary atom is activated in the training set, and generate the initial value of the mechanism quantity portrait matrix; for operation state dictionary atoms with fewer than the preset threshold, fill their portrait vector with the average physical operation mechanism quantity of the global training set. Specifically, let the dimension of the physical operation mechanism quantity be... , No. The measured physical mechanism vector of each sample is Mechanistic Quantitative Profiling Matrix , No. List For the first A profile vector of each operational state dictionary atom. Activation determination threshold. Take the 90th percentile of the absolute values ​​of all encoded components in the training set. For example, if the data volume is normal, it can be approximated as 10. -3 For the first Atoms, if Then it is determined to be activated. Number of activations. .like (Example) If 1% of the sample is positive, then the initial profile is the mean of the physical mechanism quantity of that atomic activation sample. Otherwise, take the global mean. .

[0057] S3.22. Fix the external condition dictionary, the operational state dictionary, and the external condition encoding and operational state encoding of each training sample. Using the measured physical operation mechanism quantity as the anchor point, update the mechanism quantity profile matrix along the gradient direction that reduces the mechanism quantity reconstruction error. The update method satisfies: ; in, For the first The updated mechanism profile matrix For the first Mechanistic profile matrix of the wheel. Update the step size for the portrait matrix. The training sample is a vector of measured physical operation mechanism quantities. For the first The operational status code obtained from the first round of solving. For the first Transposition of wheel operation status code For matrix multiplication operations; after updating the fixed mechanism quantity profile matrix, an optimization objective including cost feature reconstruction term, dual-space sparse regularization term, low correlation decoupling constraint term and operation mechanism quantity anchoring constraint term is adopted to resolve the external condition code and operation status code of each training sample, and the profile matrix update and code solving are performed alternately; Specifically, batch gradient descent is used: for each training sample Calculate the increment Take the average ,renew Update step size The example value is 0.01.

[0058] Alternating correction cycle: First, with the image matrix A and dictionary fixed, a warm start is performed using the encoding output from S3.12 as the initial values. A complete optimization objective function with five constraints is then used to resolve the encoding of each training sample using the Alternating Directional Multiplier Method (ADMM), consistent with S3.32. The objective function is: ; in This is the anchoring constraint coefficient for the operational mechanism quantity; for example, it is set to 0.1.

[0059] Then, with a fixed encoding, update the portrait matrix according to the batch update formula described above.

[0060] The above two steps constitute one round of alternating correction.

[0061] S3.23. Monitor the iterative change of the mechanism profile matrix and the mechanism reconstruction error. When the change of the mechanism profile matrix in two adjacent rounds is less than a preset threshold and the mechanism reconstruction error continues to decrease, stop the alternating correction and output a stable mechanism profile matrix. Specifically, calculate after each round... and When continuous Wheel satisfies and The process stops when the previous descent occurs, or is forcibly stopped when the maximum number of iterations (100) is reached. If the convergence condition is not met even after reaching the maximum number of iterations, the current profile matrix is ​​output and marked as "non-converged," and subsequent iterations can be increased or the update step size adjusted to re-execute the alternating correction. If the convergence condition is met, a stable profile matrix is ​​output. .

[0062] S3.3. For a single work unit, using its monetary cost characteristics and physical work mechanism quantities as encoding inputs, the converged external condition dictionary, operational status dictionary, and stable mechanism quantity profile matrix are invoked. A dual-space sparse coding method with low-correlation decoupling constraints and work mechanism quantity anchoring constraints is employed to solve the problem, outputting the external condition code and operational status code for that work unit. This includes the following steps: S3.31. Obtain the monetary cost feature vector and the measured physical operation mechanism vector for a single work unit, call the converged external condition dictionary, operating status dictionary, and stable mechanism profile matrix to complete the encoding input preparation; specifically, read the cost feature vector of the work unit to be analyzed. and measured physical mechanism vector , call .

[0063] S3.32. Construct a dual-space sparse coding optimization objective function. The optimization objective function successively includes a cost feature reconstruction term, a dual L1 sparse regularization term for external condition coding and operational status coding, a low-correlation decoupling constraint term for the two coding crossovers, and an anchoring constraint term for the operational status coding mapped by the mechanism quantity profile matrix and the measured physical operation mechanism quantity. The optimization objective function satisfies: ; in, This represents the monetary cost feature vector for a single work unit. For external condition dictionary, For the operational status dictionary, Encoding for external conditions Coded for operational status, and and Encoded vectors of the same dimension L1 sparse regularization coefficients encoded for external conditions L1 sparse regularization coefficients for encoding operational status. For low-correlation decoupling constraint coefficients, A mechanistic profile matrix, This is a vector of measured physical operation mechanism quantities for the work unit. Anchoring constraint coefficients for operational mechanism quantities. The square of the L2 norm of the vector. The L1 norm of the vector is used; the cost feature reconstruction term constrains the linear combination of the two dictionaries to restore the original cost features of the work unit; the double L1 sparse regularization term constrains the number of non-zero components of the two codes; the low correlation decoupling constraint term suppresses the square of the inner product of the two coding vectors; the mechanism quantity anchoring constraint term constrains the consistency between the predicted value of the physical quantity after the operation status code is mapped by the portrait matrix and the measured physical quantity.

[0064] Furthermore, the Alternating Direction Multiplier Method (ADMM) is used to solve the above-mentioned objective function. Auxiliary variables are introduced. An augmented Lagrangian function is constructed, and the original, auxiliary, and dual variables are updated alternately. A soft threshold operator is used to update the auxiliary variables. Penalty parameters... Example value: 1.0. Convergence condition: original residuals. And dual residuals The maximum number of iterations is 1000.

[0065] S3.33. Output the external condition code and operational status code of this work unit for subsequent comparable group division and operational efficiency evaluation. Specifically, output... and The data is then stored in the cloud. If the L2 norm of the external condition encoding is zero, it is marked as an anomalous sample and will not participate in the S3.4 grouping.

[0066] S3.4. Based on the similarity of the external condition codes of each work unit, comparable work groups are divided. Within each group, operational efficiency evaluation and control weakness identification are carried out based on the operational status code, and multi-dimensional intelligent benchmarking analysis results are output. This includes the following steps: S3.41 Calculate the cosine similarity of the external condition codes of each work unit, and classify work units with similar objective conditions into the same comparable work group based on the similarity threshold. Specifically, for any two work units and ,calculate If any encoding norm is zero, the similarity is 0, and the sample is labeled abnormally and not included in the grouping. Similarity threshold. The example value is 0.85. Further, the connected component method can be used for grouping: with the work unit as the node and the edge as the similarity (not lower than a threshold), the connected components of the graph are taken as comparable work groups.

[0067] For the grouping results, if the number of work units in a certain group is less than If the units within a group are found to have significantly different overall objective conditions after grouping, the similarity threshold can be appropriately increased and the connected component method can be re-executed to ensure the comparability of the work units within the group.

[0068] S3.42. Select the optimal set of work units in terms of unit cost from the set of cost features at the work unit level. Sort the atoms in the operation status dictionary in the optimal set of work units according to their activation frequency. Mark the atoms with activation frequency higher than the preset high frequency threshold as high-efficiency atoms and the atoms with activation frequency lower than the preset low frequency threshold as low-efficiency atoms. Calculate the operation efficiency score of each work unit based on the activation ratio of high-efficiency atoms and low-efficiency atoms in the operation status code. Specifically, each comparable work group is screened separately: work units within the group are categorized by unit cost. Sort in ascending order, take the first few. Example: 20% constitutes the optimal subset of jobs in this group; summing all subsets forms the optimal job set. If the total number of all job units... ,but Take 50%; if If the atomic labeling is not performed, the efficiency evaluation will be output directly in order of unit cost.

[0069] Furthermore, within the optimal job set, the activation frequency of dictionary atoms for each operational state is statistically analyzed. For each comparable job group, sort them by the frequency of the best subset within that group, with the top 20% labeled as efficient atoms and the bottom 20% as inefficient atoms.

[0070] For the There are 1 work unit, and its comparable work group is denoted as . The efficient atomic index set of this group is Inefficient atomic index set is Then the efficiency score If the denominator is zero, then Take the average of the valid scores within the group.

[0071] S3.43. For work units whose operational efficiency scores are lower than the group's baseline, analyze the resource consumption patterns corresponding to the inefficient atoms activated in their operational status codes and identify the corresponding control bottleneck types. Specifically, the baseline within a group is the median of the operational efficiency scores of all work units within that group. If the sample size within a group is not less than... If a baseline is not calculated, then no intra-group benchmarking results are output; otherwise, it is marked as insufficient sample size. For job cells with scores below the baseline, their activated inefficient atoms are read. Let the first... The image vector of each inefficient atom is , its first Each component Corresponding to the Each physical mechanism dimension; let... The vector of the measured physical mechanism quantity of all work units within the comparable work group is the vector of the measured physical mechanism quantity at the th... The mean across each dimension The coefficient of determination for significantly higher physical quantities is 1.5 in the example: a profile vector of inefficient atoms. Based on the average physical mechanism values ​​of the atomic activation samples in the S3.2 training set, this reflects the typical physical consumption characteristics corresponding to this operational state mode; if ; If the example value is 1.5, then the physical quantity is determined to be significantly high. Based on the physical mechanism quantity profiles of historical work units and the pre-established mapping rules corresponding to management problem records, the type of bottleneck is determined. For example, a significantly high energy consumption component corresponds to "insufficient equipment energy efficiency management." These mapping rules are established based on the correlation analysis between abnormal patterns of physical mechanism quantities in historical work units and corresponding operational improvement cases. After review by port operation experts, they are stored as a bottleneck type rule library, which supports regular updates based on new operational cases.

[0072] S3.44. Integrate the efficiency ranking, weakness distribution and optimization direction within each comparable work group to generate multi-dimensional intelligent benchmarking analysis results, and transfer them to S4.

[0073] Specifically, for each comparable work group, an efficiency descending ranking table, a distribution statistics of bottleneck types, and optimization directions are integrated. Optimization directions use structured parameter templates, including the equipment type to be adjusted, operational parameters, target values, or adjustment directions, and are associated with the bottleneck type. For example, "insufficient equipment energy efficiency management" corresponds to "adjusting the upper limit of idle standby time for loading and unloading equipment from T1 to T2." Here, T1 is the currently configured upper limit of idle standby time for that equipment type, and T2 is a recommended value based on the statistical analysis of parameters corresponding to efficient work units within the comparable work group (exemplarily taking the median of the efficient work set), or a preset configuration by terminal operators based on actual working conditions. The results for all groups are summarized into multi-dimensional intelligent benchmarking analysis results, stored in the cloud data layer, and transmitted to S4.

[0074] S4. Analysis Result Output and Closed-Loop Interaction: Based on multi-dimensional intelligent benchmarking analysis results, the results are encapsulated in the cloud and synchronously distributed to the edge, generating operational optimization guidelines that can be sent to the on-site operation terminals to achieve closed-loop optimization of terminal operation management. This includes the following steps: S4.1. The results of multi-dimensional intelligent benchmarking analysis are structured and encapsulated to generate operational optimization guidelines that include operational efficiency evaluation, types of management shortcomings and optimization directions, and stored in the cloud analysis results library; Specifically, three data blocks are extracted from the multi-dimensional intelligent benchmarking analysis results generated by S3.44: operational efficiency evaluation, types of management and control shortcomings, and optimization directions. Among them: Operational efficiency evaluation includes the efficiency ranking and efficiency score of work units within each comparable work group; The types of control shortcomings include the shortcomings identified by each work unit in S3.43; The optimization direction includes structured parameter templates associated with the type of bottleneck, which contain the equipment type to be adjusted, operating parameters, target values, or adjustment direction.

[0075] Furthermore, the three data blocks mentioned above are associated according to the work unit identifier and encapsulated into operation optimization guidelines. Each operation optimization guideline corresponds to a work unit and includes the work unit identifier, the comparable work group to which it belongs, the operation efficiency score, the type of management weakness, and the optimization direction. The operation optimization guidelines are generated based on the benchmarking analysis results of completed work units and are used to guide the configuration of operation parameters and operation management of subsequent similar work units (with the same work type, cargo type, and work center combination) at this work node. After encapsulation, the operation optimization guidelines are stored in the cloud analysis result database, and the retrieval relationship is established using the work unit identifier as the index key.

[0076] S4.2 The cloud will synchronously distribute the operation optimization guidelines to the edge computing units of the corresponding dock operation nodes, and the edge computing units will push them to the on-site operation terminals. Specifically, the cloud determines the corresponding dock operation node for each operation unit based on the operation unit identifier in the operation optimization guide, and synchronously sends the operation optimization guide to the edge computing unit of that node via a message queue. Upon receiving the operation optimization guide, the edge computing unit pushes it to the on-site operation terminal according to preset operation terminal push rules. The on-site operation terminal is an operation management terminal or mobile terminal deployed at the dock operation site, used to display the content of the operation optimization guide to on-site personnel.

[0077] S4.3. The edge computing unit collects the execution feedback data of the operation optimization guidelines from the field operation terminal and sends it back to the cloud. The cloud updates the operation analysis parameters based on the feedback data to realize closed-loop optimization of terminal operation management.

[0078] Specifically, after the operation optimization guidelines are displayed, the on-site operation terminal collects feedback data on the execution of these guidelines by on-site personnel. This feedback data includes the execution status (executed, not executed, or partially executed), execution time, and physical mechanism data of subsequent work units applying the optimization guidelines (the change is obtained by comparing with the baseline value of similar historical work units that did not apply the guidelines). The feedback data carries an optimization guide identifier and an identifier for the subsequent work unit applying the guidelines. The edge computing unit collects this feedback data and transmits it back to the cloud. The cloud then associates the feedback data with the original analysis results using the optimization guide identifier.

[0079] Furthermore, after receiving the execution feedback data, the cloud compares the execution feedback data with the expected optimization directions in the operation optimization guidelines, and updates the operation analysis parameters based on the comparison results: if the execution feedback data indicates that the operation optimization guidelines are effective, the current operation analysis parameters are maintained or strengthened; if the execution feedback data indicates that the operation optimization guidelines are ineffective or have little effect, the operation analysis parameters are adjusted, including the similarity threshold in S3.41, the ratio of efficient / inefficient atomic annotations in S3.42, and related parameters in the efficiency score calculation.

[0080] The validity determination is based on the following criteria: if the unit cost of the subsequent work unit following the application of the optimization guidance decreases by a percentage less than a preset threshold (5% for example), or the operational efficiency score improves by a percentage less than a preset threshold (0.1 for example), it is considered valid; otherwise, it is considered invalid or the effect is not significant. Parameter adjustments are based on the following methods: if invalid, the similarity threshold in S3.41 is reduced by a preset step size (0.05 for example) to expand the range of comparable work groups, or the proportion of efficient atomic annotations in S3.42 is increased by 5 percentage points; if valid, the current parameters are maintained or finely adjusted in the same direction based on the current parameters. Thus, the cloud-based operational analysis parameters are iteratively updated based on on-site execution feedback, forming a closed-loop optimization of terminal operation management.

[0081] like Figure 2 As shown, this embodiment also provides a multi-dimensional cloud-edge collaborative operation analysis system for dry bulk and general cargo terminals, used to execute the steps of the above-described multi-dimensional cloud-edge collaborative operation analysis method for dry bulk and general cargo terminals, including: The data acquisition and business-finance integration aggregation unit 1 is equipped with edge data access interfaces deployed at various nodes of the terminal and cloud business-finance data access interfaces. The built-in edge preprocessing logic performs on-site preprocessing on the received terminal site operation data and equipment physical measurement data, and completes the multi-source data association and alignment through cloud aggregation and mapping logic, outputting a standardized dataset at the operation unit level that includes cost attributes and terminal physical operation mechanism quantities. Cost Feature Construction and Multi-Dimensional Aggregation Unit 2: This unit loads a standardized dataset at the work unit level, uses built-in work center division logic to determine the cost aggregation objects for each work unit, and completes work unit-level cost quantification through built-in cost driver hierarchical calculation logic. It aggregates and outputs work unit-level monetary cost feature sets by department, work process, and type of goods. The dual-space decoupling and intelligent benchmarking analysis unit 3 is equipped with a feature input interface to receive the cost features of the operation unit level and the physical operation mechanism quantity collected by the dock production equipment. It has a built-in dual-space sparse coding logic with low correlation decoupling constraints and operation mechanism quantity anchoring constraints to decouple the cost features, obtain the operation status code stripped of the influence of objective conditions, and divide the comparable operation groups according to the similarity of external condition codes. Within the group, the operation efficiency evaluation and control shortcomings identification are completed, and multi-dimensional intelligent benchmarking analysis results are output. The result encapsulation and closed-loop interaction unit 4 receives multi-dimensional intelligent benchmarking analysis results. The cloud-based built-in result encapsulation logic generates operation optimization guidelines that can be sent to the on-site operation terminal. The guidelines are then sent to the edge terminal and sent back to the feedback link to complete the on-site execution feedback collection, thereby realizing closed-loop optimization of terminal operation and management.

[0082] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for multi-dimensional cloud edge collaborative operation analysis of dry bulk and general cargo terminals, characterized in that, include: S1. Data collection and integration of business and finance: Real-time collection of terminal site operation data, business and financial data and equipment physical measurement data, and through on-site preprocessing at the edge and unified collection and mapping in the cloud, forming a standardized dataset at the operation unit level that includes cost attributes and terminal physical operation mechanism quantities. S2. Construction of multi-dimensional cost features: Based on the standardized dataset at the work unit level, through the division of work centers and the hierarchical calculation of cost drivers, a set of work unit-level monetary cost features covering the dimensions of departments, work links, and goods is constructed. S3. Dual-space decoupling intelligent benchmarking analysis: Based on the cost characteristics of work unit level and the physical operation mechanism quantity collected by the terminal production equipment, a dual-space sparse coding technique with low correlation decoupling constraints and operation mechanism quantity anchoring constraints is used to decouple cost characteristics and obtain the operation status code that is stripped of the influence of objective conditions; comparable work groups are divided according to the similarity of external condition codes, and operational efficiency evaluation and control shortcomings are completed within the group, outputting multi-dimensional intelligent benchmarking analysis results; S4. Analysis Result Output and Closed-Loop Interaction: Based on the multi-dimensional intelligent benchmarking analysis results, the cloud-based result encapsulation and edge-based synchronous feedback are used to generate operation optimization guidelines that can be distributed to the on-site operation terminal to achieve closed-loop optimization of terminal operation and management.

2. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 1, characterized in that, In step S1, the data collection and integration of business and financial data includes the following steps: S1.1 On-site operation data and equipment physical measurement data of wharf berths, loading and unloading equipment, transportation equipment, storage yards, and gate nodes are collected locally by edge computing units deployed at each node, and data cleaning, outlier removal, and operation cycle time window aggregation processing are performed to generate standardized operation data slices at the edge. S1.

2. For the standardized operation data slices uploaded by each edge node, as well as the business and financial data output by the business production system and financial accounting system, the fields are mapped and associated in the cloud with the operation unit identifier as the primary key to establish the corresponding mapping relationship between operation unit, resource consumption and cost item. S1.3 Perform consistency verification and missing value marking on the multi-source data after association mapping, generate a standardized dataset at the operation unit level that simultaneously contains monetary cost attributes and physical quantity terminal physical operation mechanism quantities, and store it in the cloud data layer.

3. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 1, characterized in that, In S2, the construction of multi-dimensional cost features includes the following steps: S2.

1. For the standardized dataset at the work unit level, the work centers are divided into four categories based on the terminal operation process: loading and unloading operations, horizontal transportation, yard storage, and auxiliary operations. The work unit identifier is used as the matching key to complete the matching of work center affiliation and determine the cost collection object of each work unit. S2.2 For cost data collected to each work center, a two-level allocation calculation logic of resource driver and work driver is adopted. First, the resource layer cost is allocated to the corresponding work center according to the resource driver. Then, the total cost of the work center is allocated to each subordinate work unit according to the work driver, thus completing the cost quantification calculation at the work unit level. S2.

3. Classify and collect the quantitative cost data of the work unit according to three dimensions: department, work process, and type of goods, and construct a multi-dimensional work unit-level monetary cost feature set, which is then output to S3.

4. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 1, characterized in that, In S3, the dual-space decoupling intelligent benchmarking analysis includes the following steps: S3.1 Using the cost feature set of work unit level as training sample, construct an external condition dictionary and an operation status dictionary that respectively carry objective conditions and operation status. Incorporate cost feature reconstruction error, dual-space sparse regularization term and low correlation decoupling constraint term into joint optimization objective for iterative training to obtain converged dual-space dictionary parameters. S3.

2. Based on the activation correspondence between the dictionary atoms of each operational state in the training set and the physical operation mechanism quantities of the samples, the initial value of the mechanism quantity profile matrix is ​​generated statistically, and the profile matrix is ​​alternately corrected using the measured physical operation mechanism quantities as anchor points to obtain a stable mechanism quantity profile matrix. S3.3 For a single work unit, its monetary cost characteristics and physical work mechanism quantity are used as the encoding input. The converged external condition dictionary, operation status dictionary and stable mechanism quantity profile matrix are called. The dual-space sparse coding with low correlation decoupling constraints and work mechanism quantity anchoring constraints is used to solve the problem and output the external condition code and operation status code of the work unit. S3.

4. Based on the similarity of the external condition codes of each work unit, divide the work into comparable work groups. Within each group, conduct operational efficiency evaluation and identify control shortcomings based on the operational status code, and output multi-dimensional intelligent benchmarking analysis results.

5. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 4, characterized in that, In S3.1, the dual-space dictionary construction and low-correlation constraint training include the following steps: S3.

11. Generate initial atomic sets for the external condition dictionary and the operation status dictionary using a random initialization method. The external condition dictionary is used to carry objective condition information such as cargo type, ship type, work batch, and berth conditions, while the operation status dictionary is used to carry operation status information such as scheduling coordination, equipment organization, and resource allocation efficiency. S3.

12. Using the cost feature set of the work unit level as the training input, solve the external condition code and the operation status code of each training sample in each iteration. Based on the joint optimization objective composed of cost feature reconstruction error, dual-space sparse regularization term and low correlation decoupling constraint term, alternately update the atomic parameters of the external condition dictionary and the operation status dictionary. S3.

13. Monitor the iterative change of the joint optimization objective. When the cost feature reconstruction error drops to a preset threshold and the change of the joint optimization objective for a consecutive preset number of rounds is less than the preset change threshold, stop iterative training and output the converged external condition dictionary and operation status dictionary parameters.

6. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 4, characterized in that, In step S3.2, the construction of the mechanistic quantitative profile and the measured anchoring correction include the following steps: S3.

21. Calculate the average physical operation mechanism quantity of the corresponding sample when each operation state dictionary atom is activated in the training set, and generate the initial value of the mechanism quantity portrait matrix; for operation state dictionary atoms with fewer than the preset threshold, fill their portrait vector with the average physical operation mechanism quantity of the global training set. S3.

22. Fix the external condition dictionary, the operational state dictionary, and the external condition encoding and operational state encoding of each training sample. Using the measured physical operation mechanism quantity as the anchor point, update the mechanism quantity profile matrix along the gradient direction that reduces the mechanism quantity reconstruction error. The update method satisfies: ； in, For the first The updated mechanism profile matrix For the first Mechanistic profile matrix of the wheel. Update the step size for the portrait matrix. The training sample is a vector of measured physical operation mechanism quantities. For the first The operational status code obtained from the first round of solving. For the first Transposition of wheel operation status code For matrix multiplication operations; after updating the fixed mechanism quantity profile matrix, an optimization objective including cost feature reconstruction term, dual-space sparse regularization term, low correlation decoupling constraint term and operation mechanism quantity anchoring constraint term is adopted to resolve the external condition code and operation status code of each training sample, and the profile matrix update and code solving are performed alternately; S3.

23. Monitor the iterative change of the mechanism profile matrix and the mechanism reconstruction error. When the change of the mechanism profile matrix in two adjacent rounds is less than the preset threshold and the mechanism reconstruction error continues to decrease, stop the alternating correction and output a stable mechanism profile matrix.

7. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 4, characterized in that, In step S3.3, the constrained dual-space sparse coding includes the following steps: S3.

31. Obtain the monetary cost feature vector and the measured physical operation mechanism vector of a single work unit, call the converged external condition dictionary, operation status dictionary and stable mechanism profile matrix to complete the encoding input preparation; S3.

32. Construct a dual-space sparse coding optimization objective function, which sequentially includes a cost feature reconstruction term, a dual L1 sparse regularization term for external condition coding and operational status coding, a low-correlation decoupling constraint term for the two coding crossovers, and an anchoring constraint term for the operational status coding mapped by the mechanism quantity profile matrix and the measured physical operation mechanism quantity. The optimization objective function satisfies: ； in, This represents the monetary cost feature vector for a single work unit. For external condition dictionary, For the operational status dictionary, Encoding for external conditions Coded for operational status, and and Encoded vectors of the same dimension L1 sparse regularization coefficients encoded for external conditions L1 sparse regularization coefficients for encoding operational status. For low-correlation decoupling constraint coefficients, The matrix represents the mechanistic quantity profile, where m is the vector of measured physical mechanistic quantities for each work unit. Anchoring constraint coefficients for operational mechanism quantities. The square of the L2 norm of the vector. The L1 norm of the vector; S3.33 Output the external condition code and operational status code of the work unit for subsequent comparable group division and operational efficiency evaluation.

8. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 4, characterized in that, In step S3.4, the comparable group division, operational efficiency evaluation, and weakness identification include the following steps: S3.41 Calculate the cosine similarity of the external condition codes of each work unit, and classify work units with similar objective conditions into the same comparable work group based on the similarity threshold. S3.

42. Select the optimal set of work units in terms of unit cost from the set of cost features at the work unit level. Sort the atoms in the operation status dictionary in the optimal set of work units according to their activation frequency. Mark the atoms with activation frequency higher than the preset high frequency threshold as high-efficiency atoms and the atoms with activation frequency lower than the preset low frequency threshold as low-efficiency atoms. Calculate the operation efficiency score of each work unit based on the activation ratio of high-efficiency atoms and low-efficiency atoms in the operation status code. S3.

43. For work units whose operational efficiency scores are lower than the group's baseline, analyze the resource consumption patterns corresponding to the inefficient atoms activated in their operational status codes and identify the corresponding control bottleneck types. S3.

44. Integrate the efficiency ranking, weakness distribution and optimization direction within each comparable work group to generate multi-dimensional intelligent benchmarking analysis results, and transfer them to S4.

9. The multi-dimensional cloud-edge collaborative operation analysis method for dry bulk cargo and general cargo terminals according to claim 1, characterized in that, In step S4, the analysis results are output and interact with the closed loop, including the following steps: S4.

1. The results of multi-dimensional intelligent benchmarking analysis are structured and encapsulated to generate operational optimization guidelines that include operational efficiency evaluation, types of management shortcomings and optimization directions, and stored in the cloud analysis results library; S4.2 The cloud will synchronously distribute the operation optimization guidelines to the edge computing units of the corresponding dock operation nodes, and the edge computing units will push them to the on-site operation terminals. S4.

3. The edge computing unit collects the execution feedback data of the operation optimization guidelines from the field operation terminal and sends it back to the cloud. The cloud updates the operation analysis parameters based on the feedback data to realize closed-loop optimization of terminal operation management.

10. A multi-dimensional cloud-edge collaborative operation analysis system for dry bulk cargo and general cargo terminals, characterized in that: When the computer program of the multi-dimensional cloud-edge collaborative operation analysis system for dry bulk and general cargo terminals is executed, it performs the steps of the multi-dimensional cloud-edge collaborative operation analysis method for dry bulk and general cargo terminals as described in any one of claims 1-9, including: The data collection and business-finance integration collection unit (1) is equipped with edge data access interfaces and cloud business-finance data access interfaces deployed at each node of the terminal. The built-in edge preprocessing logic performs on-site preprocessing on the received terminal site operation data and equipment physical measurement data, and completes multi-source data association and alignment through cloud collection and mapping logic, outputting a standardized dataset at the operation unit level that includes cost attributes and terminal physical operation mechanism quantities. Cost feature construction and multi-dimensional aggregation unit (2), the cost feature construction and multi-dimensional aggregation unit (2) loads the standardized dataset at the operation unit level, the built-in operation center division logic determines the cost aggregation object of each operation unit, and completes the cost quantification at the operation unit level through the built-in cost driver hierarchical calculation logic, and aggregates and outputs the operation unit level monetary cost feature set according to the dimensions of department, operation link and commodity type; The dual-space decoupling and intelligent benchmarking analysis unit (3) is equipped with a feature input interface to receive the cost features of the operation unit level and the physical operation mechanism quantity collected by the dock production equipment. It has a built-in dual-space sparse coding logic with low correlation decoupling constraints and operation mechanism quantity anchoring constraints to decouple the cost features, obtain the operation status code stripped of the influence of objective conditions, and divide the comparable operation groups according to the similarity of external condition codes. Within the group, it completes the operation efficiency evaluation and control shortcomings identification, and outputs multi-dimensional intelligent benchmarking analysis results. The result encapsulation and closed-loop interaction unit (4) receives multi-dimensional intelligent benchmarking analysis results, generates operation optimization guidelines that can be sent to the on-site operation terminal through the cloud-built result encapsulation logic, and completes on-site execution feedback collection through the edge terminal synchronous distribution and feedback transmission link, thereby realizing closed-loop optimization of terminal operation management.

Images