A method and system for predicting energy consumption based on production operation processes of bulk and general cargo terminals

Abstract

The present application relates to the field of port energy management, and provides a kind of energy consumption prediction method and system based on bulk cargo wharf production operation process, method includes: obtaining cargo attribute information set and equipment parameter set, by context fusion, time sequence storage and data cleaning to build linkage database;Based on cargo attribute information set split process, combined with equipment archives and operation urgency coefficient to verify and collaborative division, according to theoretical and actual berthing duration to carry out risk assessment and generate link list;According to link list and real-time working condition, combined with mapping table to output candidate equipment set or compensation scheme, generate collaborative operation trigger signal, start collaborative scheme then resource allocation generates equipment allocation list;Build initial energy consumption prediction model, online iterative optimization;Calculate each operation link and final energy consumption prediction result, generate final energy consumption prediction report.The present application is suitable for port intelligent energy management and automatic scheduling by collaborative operation process dynamic programming and energy consumption adaptive prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of port energy management, and particularly relates to an energy consumption prediction method and system based on a production operation process of a bulk and general cargo terminal. BACKGROUND

[0002] With the rapid development of global economy, international trade is becoming increasingly frequent. As a key hub for cargo transportation, the energy consumption of a port is also expanding, and its energy consumption management has attracted widespread attention. According to authoritative statistical data, the production energy consumption of a bulk and general cargo terminal accounts for a considerable proportion of the total energy consumption of a port, and has shown an increasing trend in recent years. The operation process of a bulk and general cargo terminal is complex, involving various devices such as ship loaders, stacker-reclaimers, belt conveyors, bucket door machines, and door machines, and different types of cargos such as coal, ore, and grain. The running states and energy consumption of the devices are quite different, which brings great challenges to energy consumption prediction.

[0003] Chinese patent application with publication number CN120297705A provides an intelligent device scheduling system and method for a bulk and general cargo terminal, a device, and a storage medium. The scheduling system includes a data acquisition module, an information analysis and strategy generation module, a joint scheduling calculation module, a scheduling execution module, an execution feedback module, a model updating module, and a rule induction module. The method constructs a closed-loop free system that integrates semantic understanding, strategy generation, and scheduling execution to solve the problem that unstructured information cannot be effectively utilized by the scheduling system. The method can accurately extract scheduling parameters from unstructured data, even directly extract scheduling feature parameters and generate structured scheduling strategy suggestions from unstructured data, reduce the data threshold of system intelligent transformation, drive the scheduling model together with structured parameters, and introduce a feedback learning mechanism.

[0004] However, the current technology still faces many challenges. When performing device matching, the energy consumption prediction and operation planning system of a bulk and general cargo terminal usually only performs static allocation based on the rated parameters calibrated when the device is shipped. When a device is subjected to long-term high-load operation, mechanical wear or performance degradation occurs, and its actual operation capacity is often lower than the rated level. The traditional system cannot dynamically perceive such performance changes and still performs task allocation based on ideal rated parameters. Once high-intensity operation tasks are incorrectly allocated to such performance-degraded devices, not only will the efficiency of the operation link significantly decrease, forming a process bottleneck, but also may cause a chain of delays in subsequent links. Further, the allocation deviation caused by ignoring the dynamic health status of the device may make the entire ship loading and unloading operation unable to be completed within the planned port stay time, resulting in high demurrage fees and disrupting the overall ship schedule of the port, thereby seriously affecting the overall operation efficiency and core competitiveness of the terminal. SUMMARY

[0005] To achieve the above objectives, this invention provides an energy consumption prediction method based on the production operation process of a bulk cargo terminal, the specific technical solution of which is as follows:

[0006] Simultaneously acquire cargo attribute information set and equipment parameter set, and sequentially perform context association fusion, structured time-series storage and data cleaning on the cargo attribute information set and equipment parameter set to build a linked database;

[0007] Based on the cargo attribute information set of the linked database, the standard operating procedure is initially broken down. The feasibility of the initially broken-down procedure is verified and the collaborative operation links are divided by combining the equipment archive and operation urgency coefficient of the linked database. Based on the theoretical total operation time and the actual docking time of the ship, the verified procedure is risk-assessed and optimized for emergencies, and a list of links is generated.

[0008] Based on the real-time operating information of the process list and equipment archive, and combined with the preset process requirement-equipment parameter mapping table, a candidate equipment set or compensation scheme is output. Based on the output results, a branch judgment is made to generate a collaborative operation trigger signal. Based on the collaborative operation trigger signal, it is determined whether to start the collaborative scheme generation process. If it is started, resources are allocated to the equipment in the candidate equipment set, an equipment allocation list is generated, and the collaborative scheme generation process is continuously monitored and analyzed. Feedback is given when preset feedback conditions are met.

[0009] An initial energy consumption prediction model is constructed by building a multivariate regression model and a gradient boosting tree XGBoost model. The initial energy consumption prediction model is then fine-tuned through an online iterative optimization mechanism to generate an updated energy consumption prediction model.

[0010] The system calls the updated energy consumption prediction model to calculate the energy consumption prediction results for each sub-stage of the operation and the final energy consumption prediction results. Through multi-dimensional visualization analysis and customized report generation, the final energy consumption prediction report is generated.

[0011] Furthermore, the method for constructing the linked database includes:

[0012] Obtain cargo information, use a validity screening function to verify the validity of the cargo information, and if the verification result is invalid, trigger a dual protection mechanism to fill in the data and construct a cargo attribute information set.

[0013] The system synchronously acquires multi-source electrical signals and dynamic operating condition signals from the equipment, generates instantaneous energy consumption based on the multi-source electrical signals, and integrates the instantaneous energy consumption and the dynamic operating condition signals into a set of equipment parameters.

[0014] The cargo attribute information set and equipment parameter set are context-associated and fused, and then stored in a structured time sequence. An outlier detection and processing are performed on the context-associated and fused data using a hybrid verification algorithm to build a linked database.

[0015] Further, the double guarantee mechanism includes automatic matching of the knowledge base and manual maintenance;

[0016] The automatic matching of the knowledge base refers to taking the cargo type label in the cargo attribute information set as a retrieval keyword to search in the cargo attribute knowledge base. If a pre-stored entry is successfully matched, the corresponding missing cargo attribute information is automatically extracted and filled in.

[0017] The manual maintenance refers to pushing a to-do task to the operation terminal if a pre-stored entry cannot be matched in the cargo attribute knowledge base, prompting the operator to manually enter and calibrate.

[0018] Further, the generation method of the link list includes:

[0019] Based on the cargo attribute information set in the linkage database, it is determined whether the differentiated process division mechanism is triggered through a preset rule model. If the differentiated process division mechanism is triggered, the standard operation process is split into sub-links that adapt to the characteristics of the cargo, and a first intermediate link list is generated.

[0020] Each link in the first intermediate link list is traversed, and the cooperative operation trigger threshold of each link is calculated in combination with the equipment archives library and the operation urgency coefficient in the linkage database. It is determined whether the to-be-processed operation rate requirement of each link is greater than the cooperative operation trigger threshold. If it is greater, the link is split into multiple cooperative operation sub-links. If it is not greater, the link is maintained, and a second intermediate link list is generated.

[0021] The theoretical total operation time and the actual ship docking time are compared to determine whether the emergency process merging mechanism is triggered. If the emergency process merging mechanism is triggered, the second intermediate link list is merged and optimized, and the highest priority of equipment use is marked, and a link list is generated.

[0022] Further, the determination logic of the triggered emergency process merging mechanism includes multiplying the theoretical total operation time by a time compression tolerance coefficient to calculate the shortest feasible time required for the standard process to complete the standard process under extreme acceleration. The shortest feasible time and the actual ship docking time are compared. If the actual ship docking time is shorter than the shortest feasible time, it is determined that there is a significant time risk, and the emergency process merging mechanism is triggered.

[0023] The time compression tolerance coefficient is a threshold coefficient for defining an emergency state preset in the system configuration library.

[0024] Further, the generation method of the equipment allocation list includes:

[0025] According to the link demand of the link list, the corresponding equipment parameters are queried from the preset link demand-equipment parameter mapping table, the current available equipment is retrieved to form an initial candidate pool, the equipment in the initial candidate pool is traversed, the dynamic evaluation parameters are calculated in combination with the real-time working condition information of the equipment archive in the linkage database, the equipment whose dynamic evaluation parameters meet the threshold is added to the candidate equipment set, and if the candidate equipment set is empty, a compensation mechanism is triggered to generate a compensation scheme;

[0026] According to the candidate equipment set or the compensation scheme output in the previous step, branch judgment is performed, if the output is the compensation scheme, a collaborative work trigger signal of 0 is directly output, and if the output is the candidate equipment set, each equipment in the candidate equipment set is traversed, it is judged whether there is at least one candidate equipment whose dynamic trigger threshold is greater than or equal to the work rate demand, if there is, a collaborative work trigger signal of 0 is generated, and if there is not, a collaborative work trigger signal of 1 is generated;

[0027] According to the collaborative work trigger signal, it is judged whether to start the collaborative scheme generation process, if it is started, the work quantity is proportionally distributed based on the dynamic processing capacity of each equipment in the candidate equipment set, an equipment distribution list is generated, the results of the collaborative scheme generation process are analyzed for continuous monitoring and statistical analysis, and when the preset feedback condition is met, the terminal work flow and data are fed back and corrected.

[0028] Further, the generation method of the updated energy consumption prediction model comprises:

[0029] The historical work data is extracted from the linkage database as a training set, combined with the actual real energy consumption value, a multiple regression model is constructed, and a preliminary energy consumption prediction value is output;

[0030] Based on the prediction residual of the actual real energy consumption value and the preliminary energy consumption prediction value and the equipment distribution list, a gradient boosting tree XGBoost model is introduced, the multiple regression model and the gradient boosting tree XGBoost model are gradually corrected and fused to obtain an initial energy consumption prediction model, and an energy consumption prediction value is output;

[0031] The energy consumption prediction value and the actual real energy consumption value are compared, the prediction error rate is calculated, if the prediction error rate exceeds a preset threshold, an online iterative optimization mechanism is triggered, the initial energy consumption prediction model is fine-tuned based on the work data stream, an updated energy consumption prediction model is generated, and the updated energy consumption prediction model is subjected to significance test and goodness-of-fit test.

[0032] Further, the generation method of the final energy consumption prediction report comprises:

[0033] The equipment distribution list is parsed, the updated energy consumption prediction model is called for prediction calculation, and the energy consumption prediction results of each work sub-link and the final energy consumption prediction results are obtained respectively;

[0034] Based on the energy consumption prediction results of each operation sub-link and the final energy consumption prediction results, a final energy consumption prediction report containing visual analysis charts and structured data reports is generated.

[0035] An energy consumption prediction system based on bulk and general cargo terminal production operation process is used to implement the energy consumption prediction method based on bulk and general cargo terminal production operation process, comprising a multi-source data fusion module, a dynamic process planning module, an intelligent resource allocation module, an adaptive prediction module and a visual decision support module.

[0036] The multi-source data fusion module is used to synchronously acquire a set of cargo attribute information and a set of equipment parameters, and sequentially perform context association fusion, structured time sequence storage and data cleaning on the set of cargo attribute information and the set of equipment parameters, and construct a linkage database.

[0037] The dynamic process planning module is used to preliminarily split a standard operation process based on the set of cargo attribute information of the linkage database, perform feasibility verification and collaborative operation link division on the preliminarily split process in combination with the equipment archives and operation urgency coefficients of the linkage database, perform risk assessment and emergency optimization on the verified process according to the theoretical total operation time and the actual berthing time of the ship, and generate a link list.

[0038] The intelligent resource allocation module is used to output a candidate equipment set or a compensation scheme according to the link list and real-time working condition information of the equipment archives in combination with a preset link demand-equipment parameter mapping table, perform branch judgment according to the output result, generate a collaborative operation trigger signal, judge whether to start a collaborative scheme generation process according to the collaborative operation trigger signal, perform resource allocation on the equipment in the candidate equipment set if the collaborative scheme generation process is started, generate an equipment allocation list, and continuously monitor and analyze the collaborative scheme generation process, and feed back when the preset feedback condition is met.

[0039] The adaptive prediction module is used to construct a multivariate regression model and an initial energy consumption prediction model of a fusion gradient boosting tree XGBoost model, fine-tune the initial energy consumption prediction model through an online iterative optimization mechanism, and generate an updated energy consumption prediction model.

[0040] The visual decision support module is used to calculate the energy consumption prediction results of each operation sub-link and the final energy consumption prediction results by calling the updated energy consumption prediction model, generate a final energy consumption prediction report through multidimensional visual analysis and customized report generation.

[0041] Compared with the prior art, the present application has the following advantages:

[0042] The present application improves the problem that the traditional analysis method is difficult to accurately attribute and quantitatively model the micro-flow characteristics of the material and the macro-energy consumption response of the equipment due to the fragmented data sources and single dimension by real-time matching and context association of high-frequency, dynamic equipment energy consumption and working condition parameters, and macroscopic cargo physical properties and operation task information.

[0043] The present application reduces the deviation of the operation plan from the actual working condition of the terminal caused by the solidification of the process template by establishing a multi-stage judgment process planning mechanism based on cargo characteristics, equipment capacity and operation time limit, and improves the problem that the traditional single planning method cannot simultaneously consider multi-dimensional dynamic constraints, resulting in process redundancy or disconnection with reality.

[0044] The present application avoids the problem that the actual performance of the equipment is lower than the rated value due to long-term use, thereby causing the task allocation to exceed its capacity load or the collaborative operation efficiency to be low by combining the rated performance parameters of the equipment with a real-time performance attenuation coefficient reflecting the aging and wear degree of the equipment to dynamically evaluate and match the real capacity of the candidate equipment for each operation link.

[0045] The present application avoids the problem that a single model is difficult to simultaneously consider high accuracy of the prediction result and explainability of the main factors due to the limitation of the ability by combining a multivariate regression base model capturing the main linear trend with a gradient boosting tree XGBoost correction model deep learning the prediction residual to jointly predict the energy consumption.

[0046] The present application avoids the problem that the manager is difficult to quickly trace the key energy consumption link and locate the core influencing factor when facing a large amount of data by taking the composition path of the total energy consumption, the contribution ratio of each link and the flow relationship of energy between equipment as the key analysis dimension to multi-dimensionally visualize the digital energy consumption prediction result. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0048] Figure 1 is a principle flow chart of an energy consumption prediction method based on the production operation process of a bulk and general cargo terminal of the present application;

[0049] Figure 2 is a functional module diagram of an energy consumption prediction system based on the production operation process of a bulk and general cargo terminal of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0051] Embodiment 1

[0052] Please refer to Figure 1 The embodiment provides an energy consumption prediction method based on a bulk and general cargo terminal production operation process, which comprises the following steps:

[0053] S1000, synchronously acquiring a set of cargo attribute information and a set of equipment parameters , sequentially performing context correlation fusion, structured time sequence storage and data cleaning on the set of cargo attribute information and the set of equipment parameters to construct a linkage database .

[0054] Specifically, the step aims to construct a high-quality, multi-dimensional and strongly associated intelligent data acquisition and processing system, and convert originally isolated and static data into dynamic and associated linkage databases , thereby providing key data support for subsequent environment to realize accurate energy consumption prediction and attribution analysis.

[0055] Further, the step S1000 comprises the following steps:

[0056] S1100, acquiring cargo information, verifying validity of the cargo information by using an effectiveness discrimination function, triggering a double protection mechanism to perform data filling if the verification result is determined to be invalid, and constructing a set of cargo attribute information .

[0057] Specifically, the step aims to construct a structured and high-time-efficiency set of cargo attribute information from original cargo information sources from multiple channels through automatic synchronization, data verification and fusion processing , thereby providing accurate and reliable material property input for subsequent steps.

[0058] In the implementation process, this step focuses on dynamically obtaining the physical properties of the goods directly related to the energy consumption of the equipment. Through the standardized API interface with the port production planning system, the system can realize automatic and periodic synchronization of ship type, cargo list data, and ship planned docking time. The synchronization period is set to a fixed time every day, such as 00:00, 12:00, and a dynamic adjustment mechanism is established: when the urgency of the operation plan is monitored to increase, the synchronization frequency can be adaptively shortened to once an hour, ensuring the high timeliness of the data. At the same time, the urgency of the operation plan is captured and stored as a key operation demand data. The textual urgency level is mapped to a numerical operation urgency coefficient by querying the built-in configuration library of the system , which is used in the subsequent step of dynamic adjustment process to divide and match the strategy with the equipment.

[0059] To ensure the accuracy and completeness of the warehouse data, this step has a data verification mechanism. This mechanism uses an effectiveness discrimination function to automatically filter and judge each piece of cargo information record. The judgment logic of the effectiveness discrimination function is: for a given cargo information record, it is only valid when all physical properties meet two conditions at the same time, i.e., the physical property parameter value is not empty and the physical property parameter value is within the preset reasonable threshold interval. Otherwise, if any physical property does not meet the above condition, the cargo information record is invalid.

[0060] When the cargo information record is judged to be invalid, or the API interface does not return the key data, the system will start the dual protection mechanism of knowledge base automatic matching and manual maintenance to ensure the completeness of the data. First, the missing or abnormal cargo attributes are filled through the knowledge base automatic matching; for new types of goods not covered in the knowledge base, a man-machine interaction interface is provided to guide the technical personnel to manually enter and calibrate. Finally, all the verified and supplemented data are integrated into a structured cargo attribute information set , whose data structure is defined as . Among them, represents the cargo attribute information set, which is a collection of all cargo attribute information of multiple operation tasks, and each element is a multi-dimensional data record describing the characteristics of a single operation cargo; represents the operation task unique identifier, which is a unique code to distinguish different production operation tasks in the port; represents the cargo type label, which is a text or code label identifying the specific classification of the cargo, and its value is a predefined enumeration value or string, such as "coal", "ore", "food", etc.; represents the cargo density, which is a key physical property describing the mass per unit volume of the cargo, and is a floating-point value with a unit of tons per cubic meter ; represents the average particle size of the cargo, which is a physical parameter describing the average size of the particles of the bulk material, and is a floating-point value with a unit of millimeters (mm), and is used to evaluate the flowability and frictional resistance of the material during transportation and loading and unloading, which indirectly affect the energy consumption performance of the belt conveyor and other conveying equipment; represents the moisture content of the cargo, which is a physical parameter describing the percentage of water contained in the cargo, and is a floating-point value expressed in percentage (%), and is used to evaluate the additional weight and adhesion of the cargo due to the water content.

[0061] For example, in a specific application scenario, the cargo information of a batch of planned operations is obtained through an API interface, and the cargo type label of the cargo is “Indonesian lignite”. In the data verification mechanism, the validity discrimination function monitors the key parameters of the cargo information record, i.e., the cargo density is missing, so the system determines that the cargo information record is invalid. At this time, the system starts a hierarchical double protection mechanism.

[0062] First priority, knowledge base automatic matching. Taking the cargo type label “Indonesian lignite” as the search keyword, a query is performed in the cargo attribute knowledge base. If a pre-stored entry is successfully matched, the corresponding density value is automatically extracted, and the value is used to fill in the blank field in the record, completing the automatic repair of the data.

[0063] Second priority, manual maintenance. If no pre-stored entry is matched in the knowledge base, a to-do task is pushed to the operation terminal, prompting the operator to manually enter the density value according to the manifest information of the batch of cargo.

[0064] Specifically, through the double protection mechanism, it is ensured that the key physical properties of each piece of operation data entering the subsequent processing process are complete and valid. Not only the integrity of the operation data is fundamentally guaranteed, but also the problem of failure of subsequent model prediction or significant decrease in precision caused by missing key features is avoided.

[0065] Step S1200, synchronously collecting multi-source electrical signals and dynamic working condition signals of the equipment, generating instantaneous energy consumption based on the multi-source electrical signals, and integrating the instantaneous energy consumption and the dynamic working condition signals into an equipment parameter set .

[0066] Specifically, the step aims to convert the original electrical signals and state signals generated by multi-source sensors deployed on various types of equipment into a device parameter set containing precise timestamps, high-frequency device energy consumption and working condition parameters through real-time collection and structured processing , solving the technical problem of inaccurate prediction caused by the lack of real-time working condition data in traditional methods.

[0067] In the specific implementation process, the core of this step is to build a real-time monitoring network covering key equipment throughout the process to accurately capture the dynamic energy consumption and working condition changes of the equipment in actual operation at a frequency of seconds. The specific deployment is as follows: high-precision multifunctional power monitoring terminals are deployed in the motor control cabinets of key equipment such as ship loaders, stacker reclaimers, belt conveyors, and belt door machines to collect core energy consumption data in real time. In addition, specific sensors are installed at key positions of the equipment to obtain dynamic working conditions, such as installing a laser weight sensor at the head of the belt conveyor to obtain real-time load L, and installing an angle sensor on the equipment operating lever to calculate the running speed v. All collected data are aggregated through LonWorks industrial field bus technology, and 4G communication modules are redundantly deployed at key points to ensure stable and reliable data transmission.

[0068] Finally, all the calculated and collected parameters are integrated into a structured device parameter set , whose data structure is defined as . Among them, represents the device parameter set, which is a collection of state information of all monitored devices at all sampling times, and each element is a multi-dimensional data record describing the state of the device at a single time; represents the unique identifier of the device, which is used to distinguish the unique code of different physical devices in the port; represents the second-level timestamp, which is a unique time identifier marking the data collection time, usually in the form of a time date string conforming to the ISO 8601 standard, used to provide a time reference for device state data; represents the instantaneous energy consumption, which is the actual energy consumption of the device within the sampling time interval corresponding to a specific timestamp , with a floating-point value and a unit of kilowatt-hour kWh, used to accurately quantify the energy consumption of the device in a very short time; represents the real-time running speed of the device, which is a physical quantity describing the speed of the key mechanical parts of the device, such as the lifting of the hoist or the movement of the vehicle, with a floating-point value and a unit of meters per second (m / s), used to quantify the operating intensity of the device; represents the real-time load state of the device, which is a dimensionless parameter describing the ratio of the current load to the rated load, with a decimal value between 0 and 1, used to reflect the work done by the device.

[0069] The instantaneous energy consumption The calculation logic consists of two steps: First, by multiplying the instantaneous voltage, instantaneous current, and instantaneous power factor collected at the same moment, the instantaneous active power of the device at that moment is calculated; then, this instantaneous active power is multiplied by an extremely short sampling time interval to obtain the total electrical energy consumed by the device within that time interval, i.e., the instantaneous energy consumption. .

[0070] Step S1300: Set the cargo attribute information and equipment parameter set Context-related data fusion is performed, structured time-series storage is implemented, and a hybrid verification algorithm is used to detect and process outliers in the fused data, thereby constructing a linked database. .

[0071] Specifically, this step aims to integrate the cargo attribute information set collected in step S1100. and the set of device parameters collected in step S1200 By deeply integrating and correlating information from various operational processes, and through a rigorous data governance process, a high-quality, multi-dimensional, and highly interconnected database is constructed. .

[0072] The linked database A built-in device archive serves as the foundation for data support. This device archive is a static or semi-static basic database, established during the initial deployment of the entire system. The archive records the identification information of each device, such as manufacturer, model, and the rated processing capacity specified at the time of manufacture. Baseline parameters, etc. This data is typically entered once and only changed after major equipment overhauls or upgrades. Furthermore, this database is not entirely static. A background analysis module continuously processes historical data collected and stored in step S1200, such as the cumulative equipment runtime and manually entered maintenance records, and periodically calculates the latest performance degradation coefficient using a preset equipment health assessment model. This information is then updated in the database and linked to the corresponding device to dynamically reflect its actual operational capabilities.

[0073] In practice, this step is crucial for maximizing data value. Its technological innovation lies in the deep binding and integration of time-series equipment parameters and cargo attribute information within the operational process, along with rigorous data governance. Specifically, it involves the following stages:

[0074] The first phase involves data association and context annotation. This involves linking high-frequency, dynamic device parameter sets... and macro operation link cargo attribute information set Real-time matching and fusion are performed. For example, when the equipment ID number is BM01 bucket door machine is performing the operation task ID number T250919 coal unloading operation, each piece of energy consumption and working condition data generated by the equipment in this period will be automatically labeled with the composite label "unloading-coal-T250919". This linkage mechanism converts isolated equipment measurement values into multi-dimensional feature vectors containing operation scene context information.

[0075] In the second stage, structured time series storage. All data labeled through the above stages will be classified according to operation links, cargo types, etc., and stored in a structured data center based on a time series database (TSD). The time series database can efficiently process and query massive, time-stamped data. In addition, the system sets a continuous data acquisition and storage period of no less than 3 months to ensure sufficient and high-quality data support for subsequent model training and iteration.

[0076] In the third stage, data quality assurance. To ensure the quality of the data stored, a data cleaning engine is built into this stage, which integrates statistical analysis and physical law constraint hybrid verification algorithms to perform strict outlier detection and processing on the data. The core is to apply an effectiveness discrimination function for automated screening. The judgment logic of the effectiveness discrimination function is: for a given data, it must meet two conditions to be judged as valid. The first is the statistical condition, the absolute value of the deviation of the data from the historical mean of the same working condition cannot exceed the preset multiple of the historical standard deviation. The second is the physical condition, the value of the data must be greater than or equal to the minimum lower limit value set according to the physical law of the equipment. If either of the above conditions is not met, the data point is judged as invalid. For data judged as invalid, the system will use interpolation algorithms to complete and correct, so as to ensure the effectiveness, accuracy and integrity of the data in the final linkage database .

[0077] S2000, based on the cargo attribute information set of the linkage database Performs preliminary splitting of the standard operation process, combined with the equipment archives and operation urgency coefficient of the linkage database Performs preliminary splitting of the standard operation process, combined with the equipment archives and operation urgency coefficient of the linkage database Performs preliminary splitting of the standard operation process, combined with the equipment archives and operation urgency coefficient of the linkage database Performs preliminary splitting of the standard operation process, combined with the equipment archives and operation urgency coefficient of the linkage database Performs preliminary splitting of the standard operation process, combined with the equipment archives and operation urgency coefficient of the linkage database .

[0078] Specifically, the step aims to construct a linkage database based on step S1300 , the macro-port operation process is dynamically adapted and finely divided, and a machine-readable structured link list is generated , which is a key bridge connecting data perception and device-level fine control.

[0079] In the specific implementation process, the step establishes a dynamic process division mechanism driven by data through the built-in standard operation process template library of the system, and the technical personnel selects and customizes the configuration through the Web visual interface. Further, step S2000 includes:

[0080] Step S2100, based on the linkage database The set of cargo attribute information , whether the differentiated process division mechanism is triggered is determined through the preset rule model, if the differentiated process division mechanism is triggered, the standard operation process is divided into sub-links adapted to the characteristics of the cargo, and a first intermediate link list .

[0081] Specifically, the step aims to generate a first intermediate link list adapted to the characteristics of the cargo by differentiating and dynamically configuring the standard operation process template based on the set of cargo attribute information in the linkage database , to realize the pertinence and adaptability of the operation process to different materials.

[0082] In the specific implementation process, the system automatically accesses the set of cargo attribute information in the linkage database , and when it identifies that the key physical properties of a specific cargo meet the preset conditions, it will automatically trigger the preset differentiated process division mechanism.

[0083] Taking the high-density, large-grain coal unloading link as an example, when the system identifies that the cargo is coal, and its density exceeds the preset density threshold, and the average grain size d is greater than the preset average grain size threshold, the system will automatically split the unloading link into three sub-links of ship alignment, rough unloading and fine unloading due to its characteristics of large loading and unloading resistance and high cleaning difficulty. To ensure the executability of the split process, the step further quantitatively allocates the time length of each sub-link:

[0084] First, the ship alignment time length. The time length is mainly related to the ship type, and its value is determined by querying the "ship type-alignment time length" mapping table in the system configuration library. The mapping table is generated based on historical operation data statistics and maintained and updated by technical personnel.

[0085] Second, the unloading time. This time is strongly correlated with the difficulty of the unloading operation due to the physical characteristics of the cargo, and its value is calculated using a density-based dynamic adjustment rule. The calculation logic for the unloading time is as follows: First, calculate the actual density value of the current cargo. The difference between the actual density value and the baseline density value is calculated. Next, this difference is divided by a preset density increment unit, and the result is rounded down to obtain an integer density exceedance level. Then, this exceedance level is multiplied by a preset density influence coefficient to obtain a percentage increase in time. Finally, the baseline time is extended by this percentage to calculate the final actual allocation time adapted to the cargo density, i.e., the fine unloading time. The actual density value of the cargo... From the linked database The real-time parameters obtained from the data, characterizing the properties of the current work object; the reference density value is used to determine the actual density value. Whether the preset density parameter value exceeds the standard; the density increment unit is a granular parameter that disperses continuous density differences into integer levels; the density influence coefficient is a dimensionless parameter that quantifies the difference in physical properties into changes in operation time; the benchmark time is a standard operation reference time preset in the configuration library based on historical data statistics or expert experience.

[0086] Third, the rough unloading time. This time is calculated based on the margin allocation principle to ensure that the sum of the times of each sub-stage equals the theoretical total unloading time. The calculation logic for the rough unloading time is as follows: First, divide the total tonnage of the cargo in this operation by the rated unloading rate of the equipment to calculate the theoretical total time required to complete the entire unloading process; then, subtract the ship alignment time and fine unloading time determined in the previous stages from this theoretical total time, and the difference is allocated as the operation time for the rough unloading sub-stage. The total tonnage of the cargo in the operation is obtained from the core parameters of the operation task from the standardized API interface of the port production planning system; the rated unloading rate of the equipment is obtained from the linkage database. Key performance indicators of the device are obtained from the built-in device archive.

[0087] Similarly, when dealing with humidity When grains are at higher moisture levels, for example... If the value exceeds 15%, in order to meet the requirements of dust prevention and spillage prevention, the system will divide the storage process into three sub-processes: conveying, dust removal, and graded stacking. Furthermore, the start and stop thresholds of the equipment associated with the dust removal sub-process will be linked to real-time humidity and dynamic data.

[0088] In summary, through process breakdown and time calculation triggered by cargo attributes, the system updates the original standard operating procedure template into a first intermediate step list that includes specific sub-steps and their quantitative parameters. And use it as input for subsequent steps.

[0089] Step S2200: Traverse the first intermediate link list Each step in the process, combined with a linked database Equipment archives and operational urgency coefficients Calculate the collaborative operation trigger threshold for each stage, and determine whether the processing rate requirement of each stage is greater than the collaborative operation trigger threshold. If it is greater, the stage is split into multiple collaborative operation sub-stages; if it is not greater, the stage is retained, and a second intermediate stage list is generated. .

[0090] Specifically, this step aims to address the first intermediate list generated in step S2100. Combined with the linkage database generated in step S1300 The equipment archive stored in the database and the job urgency coefficient generated synchronously in step S1100 are used for the operation. Perform physical feasibility verification, automatically identify and break down collaborative operation sub-steps, and generate a list of second intermediate steps to ensure feasibility. .

[0091] In the specific implementation process, the system uses the first intermediate link list The system takes the input as input and iterates through each step. For each step, the system compares the required processing speed with a dynamically calculated collaborative operation trigger threshold in real time to ensure that the divided process planning matches the actual processing capacity of the terminal equipment, avoiding a mismatch between workload and equipment capacity. The calculation logic for the collaborative operation trigger threshold is: based on the rated processing capacity of a single piece of equipment... Based on this, and adjusted according to a comprehensive adjustment factor. This adjustment factor contains two variables: one is a performance degradation coefficient that reflects the actual capability changes of the equipment due to factors such as aging. The other is the task urgency coefficient, which reflects task priority. The final calculated result is the collaborative task trigger threshold used for the current decision. The specific process formula is as follows:

[0092] ;

[0093] Among them, the rated processing capacity It is the theoretical maximum operating rate under factory conditions or standard operating conditions, obtained from the equipment archive; the performance degradation coefficient It is a correction coefficient dynamically determined in the equipment archive based on data such as the equipment's service life, cumulative operating time, and number of maintenance operations.

[0094] In the process of traversal, the system generates a second intermediate link list according to the comparison result If the rate requirement of a link's pending work does not exceed its collaborative work trigger threshold , the link is copied to the second intermediate link list without change. If the rate requirement of a link's pending work exceeds the collaborative work trigger threshold , the system replaces the original link with multiple collaborative work sub-links and writes the new set of sub-links to the second intermediate link list. For example, in a grain loading operation, when the loading link in the link list requires a rate of 8 and the rated handling capacity of a single stacker-reclaimer is , after calculating the dynamic threshold and determining that collaborative work is needed, the system splits and replaces the original loading link with three sub-links: main loading, auxiliary loading, and combined conveying, providing clear work boundaries and task allocation basis for subsequent steps.

[0095] Step S2300, compare the theoretical total work duration and the actual ship docking duration to determine whether to trigger the emergency process merging mechanism. If the emergency process merging mechanism is triggered, merge and optimize the second intermediate link list , mark the highest priority of equipment use, and generate a link list .

[0096] Specifically, this step aims to generate a preliminary adapted second intermediate link list from step S2200, based on the ship planned docking duration obtained from the port production planning system standardized API interface in step S1100, to perform final business logic adaptation, dynamically adjust the granularity of the process, and generate a final structured link list that ensures timeliness and feasibility .

[0097] In the specific implementation process, the system sums the dynamic allocation duration of all sub-links in the input second intermediate link list to calculate the theoretical total work duration required to complete the operation under standard operation . The theoretical total work duration is compared with the obtained actual ship docking duration to determine whether the preset emergency work condition is met. This determination is automatically performed by a process merging trigger function, and its determination logic is as follows: the theoretical total work duration multiplying a time compression tolerance coefficient, a shortest feasible duration required for the standard process to complete the standard process under the limit acceleration condition is calculated. Then, the shortest feasible duration is compared with the actual berthing duration of the ship. If the actual berthing duration of the ship is still shorter than the shortest feasible duration, it is determined that there is a significant time risk, and the function triggers the process merging mechanism. The time compression tolerance coefficient is a threshold coefficient for defining an emergency state, which is preset in the system configuration library.

[0098] Once the process merging mechanism is triggered, the system will execute the preset process optimization strategy. For example, the independent unloading and stacking links are merged into one unloading and stacking linkage link, and the intermediate waiting, equipment transfer and other non-value-added sub-links between the two links are deleted in the process. At the same time, the system will mark the highest priority of the equipment used on this linkage link, such as the priority of the portal crane over the gantry crane, directly guiding the next step of equipment matching. Finally, the system generates a detailed structured link list .

[0099] S3000, according to the link list and the real-time working condition information of the equipment archive library, combining the preset link demand-equipment parameter mapping table, outputting a candidate equipment set or a compensation scheme , making branch judgment according to the output result, generating a collaborative work trigger signal , judging whether to start the collaborative scheme generation process according to the collaborative work trigger signal , if yes, allocating resources to the equipment in the candidate equipment set , generating an equipment allocation list , and continuously detecting and analyzing the collaborative scheme generation process, and feeding back when the preset feedback condition is met.

[0100] Specifically, the present step aims to scientifically and efficiently configure equipment resources for each link in the process list according to the link list output by step S2300 , combining the linkage database generated by step S1300 , outputting an executable equipment allocation list , which is the core decision-making link to ensure the efficiency and energy economy of the operation.

[0101] Further, step S3000 includes:

[0102] Step S3100, according to the link demand of the link list , querying the corresponding equipment parameters from the preset link demand-equipment parameter mapping table, retrieving the currently available equipment to form an initial candidate pool, traversing the equipment in the initial candidate pool, combining the linkage database The real-time working condition information of the equipment archive is used to calculate a dynamic evaluation parameter, and equipment whose dynamic evaluation parameter meets a threshold is added to a candidate equipment set If the candidate equipment set is empty, a compensation mechanism is triggered to generate a compensation scheme .

[0103] Specifically, this step aims to convert the abstract job requirements defined in the link list into a quantitative query of specific performance parameters of the equipment through a multi-dimensional rule mapping model, and output a candidate equipment set or a compensation scheme that meets the conditions, laying a foundation for subsequent equipment combination and collaborative decision-making.

[0104] In the specific implementation process, a link requirement-equipment parameter mapping table preset in the configuration library is used to bind the job characteristics of the link with the key performance parameters of the equipment and their quantitative thresholds. Subsequently, a candidate equipment set or a compensation scheme is generated through the following four stages:

[0105] The first stage is retrieval and filtering. All currently available equipment is retrieved from the database according to the link requirements to form an initial candidate pool.

[0106] The second stage is one-by-one verification. The dynamic evaluation parameter of each piece of equipment in the initial candidate pool is extracted. The key innovation point of this stage is that the dynamic evaluation parameter is not the fixed rated parameter in the equipment archive, but a value reflecting the current real ability of the equipment combined with the real-time working condition information in the built-in equipment archive in the linkage database . For example, for the coal rough unloading link, its requirements are mapped to "belt feeder / maximum unloading capacity ≥ 80 kN". When matching, the actual unloading capacity of a certain belt feeder after considering the aging factor is 78 kN, instead of its factory calibration of 80 kN, thereby effectively avoiding the problem of mismatch due to equipment performance degradation.

[0107] The third stage is list generation. The dynamic evaluation parameter of the equipment is compared with the quantitative threshold set in the mapping table. If the condition is met, the equipment ID is added to the candidate equipment set .

[0108] The fourth stage is gap judgment and compensation. After all candidate equipment is traversed, if the candidate equipment set is empty, it is determined that there is a device gap. At this time, the system will automatically trigger the compensation mechanism to retrieve the preset alternative scheme from the mapping table and form a compensation scheme as output.

[0109] Step S3200, according to the candidate device set output by the previous step or compensation scheme Branch judgment is performed, and if the output is the compensation scheme , a collaborative job trigger signal is directly output 0, and if the output is the candidate device set , each device in the candidate device set is traversed to determine whether there is at least one candidate device whose dynamic trigger threshold is greater than or equal to the job rate requirement, and if there is, a collaborative job trigger signal 0 is generated, and if there is not, a collaborative job trigger signal 1 is generated.

[0110] Specifically, this step aims to make a decision on the candidate device set or compensation scheme output by step S3100, and output a clear collaborative job trigger signal .

[0111] In the specific implementation process, this step first performs branch judgment on the output type of step S3100. If the input is a compensation scheme , it indicates that there is an alternative scheme at this link and no regular collaboration is needed, and this step directly outputs a collaborative job trigger signal and ends.

[0112] If the input is a candidate device set , the core decision logic is started, that is, it is determined whether any candidate device can independently complete the specified job. The core decision logic is as follows:

[0113] The system traverses each device in the candidate device set and dynamically calculates the dynamic trigger threshold of its collaborative job. The calculation logic of the trigger threshold is consistent with the judgment logic for process splitting in S2200, which is not described here, but its purpose in this step is to trigger the final device collaboration allocation. Subsequently, the system compares the job rate requirement of the link with the dynamic trigger threshold of each candidate device in the candidate device set . If there is at least one candidate device in the candidate device set whose dynamic trigger threshold is greater than or equal to the job rate requirement, it is determined that no collaboration is needed, and the collaborative job trigger signal output is 0. Otherwise, if the dynamic trigger threshold of all candidate devices in the candidate device set is less than the job rate requirement, it is determined that multi-device collaboration must be started, and the collaborative job trigger signal output is 1.

[0114] Step S3300, according to the collaborative operation trigger signal determine whether to start the collaborative scheme generation process, and if so, based on the candidate device set the dynamic processing capacity of each device, proportionally allocate the workload, and generate a device allocation list , analyze the results of the collaborative scheme generation process for continuous monitoring and statistical analysis, and when the preset feedback conditions are met, feedback and correct the terminal operation process and data.

[0115] Specifically, this step aims to allocate the final device resources based on the collaborative operation trigger signal output by step S3200 to the candidate device set output by step S3100, generate a device allocation list containing specific control parameters, and execute intelligent feedback logic to achieve closed-loop self-optimization of the entire system.

[0116] In the specific implementation process, this step includes two core stages: the collaborative scheme generation stage and the intelligent feedback optimization stage, which are as follows:

[0117] The first stage, the collaborative scheme generation stage. When the received collaborative operation trigger signal is received, this stage is started. The system not only determines the combination of devices participating in collaboration, but also proportionally allocates the workload according to the dynamic processing capacity of each candidate device in the candidate device set . For scenarios involving multiple devices collaborating, the allocation proportion of the workload for a single device is determined according to the following logic: first, sum the dynamic processing capacities of all candidate devices participating in collaborative operation to obtain the total processing capacity of the device collaboration; then, divide the dynamic processing capacity of a single device by the total processing capacity of the device collaboration to obtain the allocation proportion of the workload that the device should undertake in this collaborative task. The dynamic processing capacity is the dynamic trigger threshold of each candidate device.

[0118] The system further generates a final device allocation list based on this workload allocation proportion. The device allocation list is a structured data object, and its specific content includes three levels: the device combination list level, which clearly lists the unique identifiers of all devices participating in this collaborative operation; the workload allocation level, which clearly specifies the allocation proportion of the workload that each device in the device collaboration combination should undertake and the target operation rate after conversion; and the collaborative control instruction level, which ensures the safety and efficiency of multiple devices collaborating. ​​It also includes specific control parameters, such as speed difference control parameters, to avoid material blockage or pulling caused by speed mismatch between equipment.

[0119] The second stage is the intelligent feedback optimization stage. This stage involves continuous monitoring and statistical analysis of the combined results of device collaboration to construct a feedback loop from the execution layer to the planning layer, achieving adaptive optimization of the entire system. This intelligent feedback optimization stage mainly includes the following two feedback mechanisms:

[0120] First, feedback and correction of the terminal operation process, specifically feedback to step S2000. The system continuously monitors the long-term frequency of collaborative operations triggered in each stage. If the number of collaborations in a certain stage exceeds a preset threshold within a preset statistical period, the system determines that there is a structural bottleneck in the process division of that stage. At this time, the system will automatically generate a process optimization instruction and feed it back to step S2000, suggesting adjustments to upstream process or sub-stage parameters to stagger or smooth peak operation demands.

[0121] Similarly, if the energy consumption assessment of the combined result deviates from the preset value of the process requirement for an extended period exceeding a threshold, the system determines that the equipment capability requirements set for that process in the process requirement-equipment parameter mapping table in step S3100 are not accurate enough. At this point, the system will automatically send a correction instruction to adjust the rules in the mapping table to ensure that subsequent process division and equipment coordination better align with the actual situation of optimal energy consumption.

[0122] Second, there is the feedback from the data foundation, specifically feedback to step S1000. The system will continuously perform statistical analysis on the links in step S3100 where equipment shortages occur frequently and persistently. If a link is determined to have a persistent equipment capability deficiency, the system will generate a demand instruction for new equipment. This demand instruction for new equipment will explicitly include the required equipment type and key performance parameters, and will be pushed to the linkage database constructed in step S1300. The equipment archive provides precise and quantitative data-driven decision support for the terminal management to conduct equipment procurement and asset upgrades.

[0123] S4000, the initial energy consumption prediction model for constructing a multivariate regression model and a fused gradient boosting tree XGBoost model. The initial energy consumption prediction model is optimized through an online iterative optimization mechanism. Fine-tune the model to generate an updated energy consumption prediction model. .

[0124] Specifically, this step aims to build upon the linked database constructed in step S1300. And in conjunction with the device allocation list generated in step S3300 This is transformed into a final energy consumption prediction value that accurately quantifies the energy consumption of future operations. , generating an updated energy consumption prediction model , is the technical core of realizing port energy saving and cost optimization.

[0125] Further, step S4000 includes:

[0126] Step S4100, extracting historical operation data from the linkage database as a training set , combining the actual real energy consumption value , constructing a multiple regression model, and outputting a preliminary energy consumption prediction value .

[0127] Specifically, this step aims to extract historical operation data from the linkage database as a training set , use a multiple regression algorithm to establish a baseline model that captures the linear relationship between energy consumption and main influencing factors, and output a preliminary energy consumption prediction value .

[0128] In the specific implementation process, this step extracts effective samples from the linkage database for each operation link as a training set , each group containing device running speed v, load state L, running time t, and cargo density p and other characteristic parameters, and normalizing these different orders of magnitude of characteristic parameters to improve the efficiency and stability of model training. Subsequently, the system takes the actual real energy consumption value as the dependent variable, and the above characteristic parameters as the independent variables, to establish the following multiple regression equation:

[0129] ;

[0130] Wherein, the actual real energy consumption value is a supervisory signal in the model training process, guiding the model to learn the relationship between each feature and energy consumption, and its unit is kilowatt-hour (kWh); represents a constant term, which is the baseline prediction value of the model when all independent variables are zero; represents the regression coefficient of each independent variable, which is used to quantify the linear impact of the independent variable on energy consumption; represents a random error term, which is used to reflect random factors or nonlinear relationships that are not explained by the current model.

[0131] In the model training phase, the least squares method is used to train the training set The regression coefficients of each independent variable are estimated based on the independent variables and the known dependent variable E. After the training is completed, a validity test is performed, including a significance test, i.e., F test and P value less than 0.05, and a goodness-of-fit test, i.e., R² test and R² value greater than 0.9, to ensure that the model has statistical significance and high explanatory power.

[0132] In the model prediction stage, for new input features, the trained model will output a preliminary energy consumption prediction value .

[0133] Step S4200, based on the prediction residual of the actual real energy consumption value and the preliminary energy consumption prediction value and the equipment allocation list , introduce the gradient boosting tree XGBoost model, gradually correct and fuse the multiple regression model and the gradient boosting tree XGBoost model, obtain the initial energy consumption prediction model , output the energy consumption prediction value .

[0134] Specifically, this step aims to introduce the gradient boosting tree (Gradient Boosting Decision Tree, GBDT) typical implementation, i.e., XGBoost model, based on the actual real energy consumption value , the preliminary energy consumption prediction value generated in step S4100 and the equipment allocation list generated in step S3300, use its nonlinear learning ability to make fine corrections to capture the nonlinear laws caused by complex coupling of multiple factors, and output high-precision final energy consumption prediction value .

[0135] In the specific implementation process, the prediction residual of the multiple regression model constructed in step S4100 is taken as the learning target of the gradient boosting tree XGBoost model to capture the nonlinear laws caused by complex coupling of multiple factors that cannot be explained by the multiple regression model. The prediction residual of the multiple regression model is the difference between the actual real energy consumption value and the preliminary energy consumption prediction value , i.e. .

[0136] In the model prediction stage, the system first parses the equipment allocation list into a feature vector for each job link , fuses the preliminary energy consumption prediction value of the multiple regression model and the residual corrected by the XGBoost model in T rounds of iteration, i.e., the initial energy consumption prediction model , and calculates the final energy consumption prediction value , the specific process formula is as follows:

[0137] ;

[0138] wherein, represents the final energy consumption prediction value of the operation link i, and the unit is usually kilowatt-hour (kWh), which is used to provide accurate quantitative estimation of future operation energy consumption for the terminal management layer; represents the preliminary energy consumption prediction value of the operation link i, which is used as a reference for residual correction; k represents the index of the decision tree, and the value range is a positive integer from 1 to T, which is used to traverse each decision tree in the XGBoost model; T represents the total iteration number of the XGBoost model, that is, the total number of decision trees, which is used to determine the complexity and fitting ability of the model; represents the residual prediction value of the kth tree for the operation link i, which is the residual compensation value output by the kth decision tree in the XGBoost model for the input feature vector , and the value is equal to the weight of the leaf node in which the operation link i falls in the decision tree , and the unit is kilowatt-hour (kWh), which is used to gradually and finely correct the prediction deviation of the previous model; represents the feature vector of the operation link, which is an ordered numerical list containing all relevant characteristics of the operation link i, such as device running speed v, load state L, running time t, and cargo density p, etc.; i represents the unique index of the operation sample, which is an identifier used to uniquely identify a specific operation task record in the data set.

[0139] The calculation of the residual prediction value of the kth tree for the operation link i , the core of which is the weight of the leaf node it falls into. The weight is optimized by second-order Taylor expansion of the loss function, which not only considers the direction of the current prediction deviation, but also considers the curvature of the deviation and the complexity of the model, so as to calculate a most reasonable correction value that can effectively reduce the error and prevent overfitting of the model. The direction of the current prediction deviation is embodied by the first-order partial derivative , the curvature of the deviation is embodied by the second-order partial derivative , and the complexity of the model is controlled by the L2 regularization term coefficient . Wherein, represents the first-order partial derivative of the loss function with respect to the preliminary prediction value of the operation sample i, which is a gradient value quantifying the size and direction of the deviation between the current prediction value and the actual value, and is used to guide the direction of model correction; the preliminary prediction value is an iterative cumulative value, and for the kth decision tree, it is the sum of the preliminary energy consumption prediction value of the multivariate regression model and the correction value of the k-1th decision tree in the XGBoost model; The second-order partial derivative of the loss function with respect to the preceding predicted value of the job sample i is a gradient value that reflects the curvature near the current predicted value and is used to guide the step size of model correction.

[0140] Finally, the final energy consumption prediction values ​​for all operational steps i in the process will be calculated. Summing the results yields the final energy consumption prediction for the entire workflow. .

[0141] Step S4300: Compare the predicted energy consumption values. and actual energy consumption value The prediction error rate is calculated. If the prediction error rate exceeds a preset threshold, an online iterative optimization mechanism is triggered based on the job data stream. For the initial energy consumption prediction model Fine-tune the model to generate an updated energy consumption prediction model. The updated energy consumption prediction model Significance tests and goodness-of-fit tests were performed.

[0142] Specifically, this step aims to continuously utilize the latest job data streams through an automatically triggered online iterative optimization mechanism. and the final energy consumption prediction value output in step S4200 For the initial energy consumption prediction model deployed in step S4200 Fine-tuning and iteration are performed to generate an updated energy consumption prediction model that can adapt to environmental changes. This addresses the problem of decreased prediction accuracy caused by conceptual drift in traditional static models, such as equipment aging or changes in operating conditions.

[0143] The job data stream It is a data record containing the latest actual energy consumption results of completed tasks, generated by the continuous operation of the data acquisition and processing process constructed in steps S1000 to S3000. It is mainly used for monitoring and evaluating the initial energy consumption prediction model. The performance.

[0144] In practice, the core of this step is a data-driven closed-loop iterative process, which includes the following stages:

[0145] Phase 1: Trigger Condition Monitoring. The system continuously uses the latest job data stream. Monitoring the current initial energy consumption prediction model The predictive performance of the initial energy consumption prediction model. The final energy consumption prediction value for the output of a completed task. The actual energy consumption value generated in step S4100 The prediction error rate is calculated for comparison. When the average of the prediction error rates of 5 consecutive work links exceeds the preset threshold, the online iterative optimization mechanism of the model is automatically triggered.

[0146] Second stage, data and model preparation. Once the online iterative optimization mechanism is triggered, the latest one month of job data is intercepted from the job data stream as a new training set. At the same time, the overall architecture of the current initial fusion model is locked to ensure stability during optimization.

[0147] First, parameter fine-tuning and retraining. The system fine-tunes the current initial fusion model . Specifically, the regression coefficients of the multiple regression model and the learning rate of the XGBoost model are fine-tuned. For example, reducing the learning rate appropriately means that each new decision tree will correct the residual more conservatively, which helps to improve the generalization ability of the model.

[0148] Second, verification and deployment. After the above retraining stage is completed, the updated energy consumption prediction model must pass the significance test, i.e. F test, and the goodness-of-fit test, i.e. R² test, to verify its effectiveness. If the verification is passed, the updated energy consumption prediction model will be deployed online to replace the old version of the initial energy consumption prediction model as the engine for all subsequent prediction tasks of the system.

[0149] S5000, the updated energy consumption prediction model is called to calculate the energy consumption prediction results of each work sub-link and the final energy consumption prediction results , and through multidimensional visualization analysis and customized report generation, the final energy consumption prediction report is generated.

[0150] Specifically, this step aims to apply the updated energy consumption prediction model obtained in step S4300 to actual terminal production work planning, combined with the equipment allocation list generated in step S3300, through multidimensional visualization, to convert into intuitive management insights, and generate the final energy consumption prediction report .

[0151] Further, step S5000 includes:

[0152] Step S5100, the equipment allocation list is parsed, and the updated energy consumption prediction model is called for prediction calculation to obtain the energy consumption prediction results of each work sub-link and the final energy consumption prediction results .

[0153] Specifically, this step aims to call the trained update energy consumption prediction model to parse the equipment allocation list of the job to be predicted as input features, perform prediction calculation, and obtain energy consumption prediction results of each job sub-link and final energy consumption prediction results .

[0154] In the specific implementation process, the system parses the equipment allocation list of the job to be predicted into the input feature vector required by the model, and calls the update energy consumption prediction model for operation. This operation process strictly follows the calculation logic defined in step S4200, and finally obtains the energy consumption prediction results of each job sub-link in the job flow and the final energy consumption prediction results after summarization. These original numerical prediction results will serve as the data basis for subsequent visualization analysis and diagnosis steps.

[0155] Step S5200, based on the energy consumption prediction results of each job sub-link and the final energy consumption prediction results , generate a final energy consumption prediction report containing visual analysis charts and structured data reports .

[0156] Specifically, this step aims to take the energy consumption prediction results of each job sub-link and the final energy consumption prediction results output by step S5100 as input, convert them into intuitive management insights through an integrated result processing engine, and generate a final energy consumption prediction report containing visual analysis charts and structured data reports , providing support for port energy saving and operation decision-making.

[0157] The integrated result processing engine includes a multi-dimensional visual analysis component and a customized report generation component.

[0158] In the specific implementation process, this step is a key bridge connecting data prediction and management decision-making, taking the energy consumption prediction results of each job sub-link and the final energy consumption prediction results as input, and converting the digital prediction results into intuitive and easy-to-understand structured reports through the integrated result processing engine.

[0159] The processing engine renders the input energy consumption prediction results into a set of multi-dimensional visual analysis charts through its built-in visual analysis component to meet different analysis needs. Specific visual analysis charts include:

[0160] 1. Bar and Pie Charts: Used for horizontal comparison of energy consumption across different stages and for a visual representation of the energy consumption percentage of each piece of equipment, highlighting the top five energy-intensive devices. 2. Line Charts: Used to show the trend of energy consumption over time, such as by hour, day, or week, helping managers understand energy consumption fluctuation patterns. 3. Waterfall Charts: Used to clearly display the final energy consumption forecast results. It is constructed by gradually accumulating the energy consumption prediction results of each sub-stage, intuitively reflecting the contribution of each stage to the total energy consumption. Fourth, Sankey diagram: used to visually present the complete flow path and quantitative relationship of energy between different equipment and operation stages.

[0161] Meanwhile, the system utilizes its built-in customized report generation component to generate structured data reports containing elements such as equipment parameters, calculation processes, and error ranges based on the input energy consumption prediction results. These reports support formats such as Excel, PDF, and CSV to adapt to different application scenarios. The core feature of these reports lies in their customizable content, providing differentiated analysis for different business departments. When targeting operations managers, it offers energy consumption trend analysis based on daily, weekly, and monthly time dimensions, helping them understand energy consumption patterns for rational production scheduling. When targeting equipment maintenance departments, it provides detailed energy consumption reports for specific equipment, highlighting high-energy-consuming equipment and their operating status parameters for targeted equipment maintenance and technical upgrades.

[0162] Example 2

[0163] This embodiment, based on Embodiment 1, provides an energy consumption prediction system based on the production operation process of a bulk cargo terminal, such as... Figure 2 As shown, it includes a multi-source data fusion module, a dynamic process planning module, an intelligent resource allocation module, an adaptive prediction module, and a visualization decision support module;

[0164] The multi-source data fusion module is used to synchronously acquire cargo attribute information sets. and equipment parameter set For the set of cargo attribute information and equipment parameter set Context-related fusion, structured time-series storage, and data cleaning are performed sequentially to build a linked database. .

[0165] The dynamic process planning module is based on a linked database. cargo attribute information set Perform initial breakdown of standard operating procedures and integrate with the linked database. Equipment archives and operational urgency coefficient The feasibility of the initially broken-down process was verified and collaborative work steps were divided based on the theoretical total work time. and the actual length of time when the ship is docked to check the flow of risk assessment and emergency optimization, generating a link list .

[0166] The intelligent resource allocation module: for according to the link list and real-time working condition information of the equipment archive, combined with the preset link demand-equipment parameter mapping table, output the candidate equipment set or compensation scheme , according to the output result branch judgment, generate collaborative work trigger signal , according to the collaborative work trigger signal determine whether to start the collaborative scheme generation process, if so, allocate resources to the equipment in the candidate equipment set , generate equipment allocation list , and continuously monitor and analyze the collaborative scheme generation process, and when the preset feedback condition is met, feedback.

[0167] The adaptive prediction module: for building a multivariate regression model and an initial energy consumption prediction model based on XGBoost model , through online iterative optimization mechanism to fine-tune the initial energy consumption prediction model , generate updated energy consumption prediction model .

[0168] The visual decision support module: for calling the updated energy consumption prediction model calculate the energy consumption prediction result of each work sub-link and the final energy consumption prediction result , through multi-dimensional visual analysis and customized report generation, generate the final energy consumption prediction report .

[0169] The above technical solutions provided in the embodiments of the present application have not been described in detail, so as not to be too repetitive.

[0170] The specific embodiments described above further detail the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting energy consumption based on bulk cargo terminal production operation process, characterized in that, The method comprises the following steps: Synchronization of obtaining a cargo attribute information set and a device parameter set, context correlation fusion of the cargo attribute information set and the device parameter set in sequence, structured time sequence storage and data cleaning, and construction of a linkage database; Based on the cargo attribute information set of the linkage database, the preliminary splitting of the standard operation process is performed, the feasibility verification and collaborative operation link division of the preliminary split process are performed in combination with the device archives library and the operation urgency coefficient of the linkage database, the risk assessment and emergency optimization of the verified process are performed according to the theoretical total operation time and the actual ship berthing time, and a link list is generated; According to the link list and the real-time working condition information of the device archives library, in combination with the preset link demand-device parameter mapping table, a candidate device set or a compensation scheme is output, branch judgment is performed according to the output result, a collaborative operation trigger signal is generated, it is judged whether to start the collaborative scheme generation process according to the collaborative operation trigger signal, if yes, resource allocation is performed on the devices in the candidate device set, a device allocation list is generated, and the collaborative scheme generation process is continuously detected and analyzed, and feedback is performed when the preset condition is met; An initial energy consumption prediction model is constructed by combining a multiple regression model and an XGBoost model, and the initial energy consumption prediction model is fine-tuned through an online iterative optimization mechanism to generate an updated energy consumption prediction model; The updated energy consumption prediction model is called to calculate the energy consumption prediction results of each operation sub-link and the final energy consumption prediction results, and a final energy consumption prediction report is generated through multi-dimensional visual analysis and customized report generation.

2. The method for predicting energy consumption based on bulk cargo terminal production operation process according to claim 1, characterized in that, The construction method of the linkage database comprises: Obtaining cargo information, verifying the validity of the cargo information by using an effectiveness discrimination function, triggering a double protection mechanism for data filling if the verification result determines that it is invalid, and constructing a cargo attribute information set; Synchronously collecting multi-source electrical signals and dynamic working condition signals of the device, generating instantaneous energy consumption based on the multi-source electrical signals, and integrating the instantaneous energy consumption and the dynamic working condition signals into a device parameter set; Context correlation fusion of the cargo attribute information set and the device parameter set, structured time sequence storage, and abnormal value detection and processing of the context correlation fused data by using a hybrid verification algorithm to construct a linkage database.

3. The method for predicting energy consumption based on bulk cargo terminal production process according to claim 2, characterized in that, The double protection mechanism comprises automatic matching of a knowledge base and manual maintenance; The automatic matching of the knowledge base refers to taking the cargo type label in the cargo attribute information set as a retrieval keyword to search in the cargo attribute knowledge base, automatically extracting and filling the missing cargo attribute information if a pre-stored item is successfully matched; The manual maintenance refers to pushing a to-do task to an operation terminal if a pre-stored item cannot be matched in the cargo attribute knowledge base, prompting an operator to manually input and calibrate.

4. The method for predicting energy consumption based on bulk cargo terminal production operation process according to claim 1, characterized in that, The generation method of the link list comprises: Based on the cargo attribute information set in the linkage database, it is judged whether to trigger a differentiated process division mechanism by using a preset rule model, if the differentiated process division mechanism is triggered, the standard operation process is split into sub-links adapted to the characteristics of the cargo, and a first intermediate link list is generated; Traverse each link in the first intermediate link list, combine the equipment archives in the linkage database and the operation urgency coefficient to calculate the collaborative operation trigger threshold of each link, judge whether the rate requirement of the pending operation of each link is greater than the collaborative operation trigger threshold, if greater, split the link into multiple collaborative operation sub-links; if not greater, keep the link, and generate a second intermediate link list; Compare the theoretical total operation time length and the actual ship berthing time length to determine whether to trigger the emergency process merging mechanism, if the emergency process merging mechanism is triggered, merge and optimize the second intermediate link list, mark the highest priority of equipment use, and generate a link list.

5. The method for predicting energy consumption based on bulk cargo terminal production process according to claim 4, characterized in that, The determination logic of triggering the emergency process merging mechanism includes multiplying the theoretical total operation time length by a time compression tolerance coefficient to calculate the shortest feasible time length required for completing the standard process under extreme acceleration, comparing the shortest feasible time length with the actual ship berthing time length, and if the actual ship berthing time length is shorter than the shortest feasible time length, it is determined that there is a major time risk, and the emergency process merging mechanism is triggered; The time compression tolerance coefficient is a threshold coefficient for defining an emergency state preset in a system configuration library.

6. The method for predicting energy consumption based on bulk cargo terminal production process according to claim 1, wherein, The generation method of the equipment allocation list includes: According to the link requirements of the link list, query the corresponding equipment parameters from the preset link requirement-equipment parameter mapping table, retrieve the current available equipment to form an initial candidate pool, traverse the equipment in the initial candidate pool, combine the real-time working condition information of the equipment archives in the linkage database to calculate the dynamic evaluation parameters, add the equipment whose dynamic evaluation parameters meet the threshold to the candidate equipment set, and if the candidate equipment set is empty, trigger the compensation mechanism to generate a compensation scheme; According to the candidate equipment set or the compensation scheme output in the previous step, if the output is the compensation scheme, directly output the collaborative operation trigger signal as 0, if the output is the candidate equipment set, traverse each equipment in the candidate equipment set, judge whether there is at least one candidate equipment whose dynamic trigger threshold is greater than or equal to the operation rate requirement, if there is, generate a collaborative operation trigger signal of 0, if there is not, generate a collaborative operation trigger signal of 1; According to the collaborative operation trigger signal, judge whether to start the collaborative scheme generation process, if yes, based on the dynamic processing capacity of each equipment in the candidate equipment set, allocate the operation amount in proportion, generate an equipment allocation list, analyze the results of the collaborative scheme generation process for continuous monitoring and statistical analysis, and when the preset feedback conditions are met, feedback and correct the terminal operation process and data.

7. The method for predicting energy consumption based on bulk cargo terminal production process according to claim 1, wherein, The generation method for updating the energy consumption prediction model includes: Extract historical operation data from the linkage database as a training set, combine the actual real energy consumption value to build a multiple regression model, and output a preliminary energy consumption prediction value; Based on the prediction residual of the actual real energy consumption value and the preliminary energy consumption prediction value and the equipment allocation list, introduce the gradient boosting tree XGBoost model, gradually correct and fuse the multiple regression model and the gradient boosting tree XGBoost model to obtain an initial energy consumption prediction model, and output an energy consumption prediction value; The predicted energy consumption value is compared with the actual real energy consumption value to calculate a prediction error rate. If the prediction error rate exceeds a preset threshold, an online iterative optimization mechanism is triggered to fine-tune the initial energy consumption prediction model based on the job data stream, generate an updated energy consumption prediction model, and perform significance test and goodness-of-fit detection on the updated energy consumption prediction model.

8. The method for predicting energy consumption based on bulk cargo terminal production operation process according to claim 1, characterized in that, The method for generating the final energy consumption prediction report comprises: The device allocation list is parsed, and the updated energy consumption prediction model is called for prediction calculation to obtain energy consumption prediction results of each job sub-link and final energy consumption prediction results; Based on the energy consumption prediction results of each job sub-link and the final energy consumption prediction results, a final energy consumption prediction report containing visual analysis charts and structured data reports is generated.

9. A bulk and general cargo terminal production operation process-based energy consumption prediction system for implementing a bulk and general cargo terminal production operation process-based energy consumption prediction method according to any one of claims 1-8, characterized in that, The system comprises a multi-source data fusion module, a dynamic process planning module, an intelligent resource allocation module, an adaptive prediction module, and a visual decision support module. The multi-source data fusion module is used to synchronously acquire a set of cargo attribute information and a set of device parameters, sequentially perform context association fusion, structured time sequence storage, and data cleaning on the set of cargo attribute information and the set of device parameters, and construct a linkage database. The dynamic process planning module is used to preliminarily split a standard job process based on the set of cargo attribute information of the linkage database, perform feasibility verification and collaborative job link division on the preliminarily split process in combination with the device archives library and the job urgency coefficient of the linkage database, perform risk assessment and emergency optimization on the verified process according to the theoretical total job duration and the actual ship docking duration, and generate a link list. The intelligent resource allocation module is used to output a candidate device set or a compensation scheme according to the link list and real-time working condition information of the device archives library in combination with a preset link demand-device parameter mapping table, perform branch judgment according to the output result, generate a collaborative job triggering signal, determine whether to start a collaborative scheme generation process according to the collaborative job triggering signal, perform resource allocation on devices in the candidate device set if the collaborative scheme generation process is started, generate a device allocation list, and continuously monitor and analyze the collaborative scheme generation process to perform feedback when a preset feedback condition is met. The adaptive prediction module is used to construct an initial energy consumption prediction model of a multivariate regression model and a fusion gradient boosting tree XGBoost model, fine-tune the initial energy consumption prediction model through an online iterative optimization mechanism, and generate an updated energy consumption prediction model. The visual decision support module is used to call the updated energy consumption prediction model to calculate energy consumption prediction results of each job sub-link and final energy consumption prediction results, generate a final energy consumption prediction report through multidimensional visual analysis and customized report generation.

Images