A method and system for diagnosing energy efficiency of a dry bulk cargo terminal based on multi-source data

Abstract

The application discloses a kind of based on multi-source data's dry bulk cargo wharf energy efficiency diagnosis method and system, it is related to port intelligent technology field, including, acquisition and fusion multi-source heterogeneous data, form standardization multi-source data event stream;Quantitative causal logic in dry bulk cargo wharf energy efficiency causal knowledge graph is injected into basic digital twin by using causal discovery algorithm to standardization multi-source data event stream carries out automation analysis, constructs dry bulk cargo wharf energy efficiency causal knowledge graph;Based on wharf physical layout, equipment parameter and process logic constructs basic digital twin, forms wharf process digital twin that is enhanced by causal knowledge by quantitative causal logic in dry bulk cargo wharf energy efficiency causal knowledge graph into basic digital twin;The detailed parameters of operation task are input into wharf process digital twin that is enhanced by causal knowledge with real-time environmental data and carry out simulation deduction, generate individualized dynamic energy efficiency benchmark curve.The application improves the sensitivity of energy efficiency anomaly detection under complex working condition, the accuracy of positioning and the effectiveness of decision support.

Description

Technical Field

[0001] This invention relates to the field of intelligent port technology, and in particular to a method and system for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data. Background Technology

[0002] In the development of intelligent and green ports, energy efficiency management of dry bulk cargo terminals is of paramount importance. A representative technical solution is to build a digital twin system that integrates physical models and real-time data. By establishing a three-dimensional virtual model of the terminal and combining equipment mechanism and material flow simulation, dynamic simulation of the process flow and theoretical energy consumption calculation can be achieved. Then, online monitoring and energy efficiency assessment can be completed by comparing actual data, which reflects the mainstream direction of technological development in this field.

[0003] Existing methods face challenges in the accuracy of energy efficiency diagnosis. Energy efficiency assessment benchmarks are difficult to accurately adapt to complex and ever-changing real-time operating conditions. The energy efficiency of dry bulk cargo terminal operations is affected by a variety of dynamic factors such as cargo type, environment, equipment status, and scheduling strategies. Existing digital twin energy consumption models are mostly based on ideal parameters or historical statistics, which are difficult to accurately depict the subtle and nonlinear effects of these complex factors. The energy efficiency benchmarks generated by these models are often static or generalized and cannot sensitively reflect the unique context of each specific operation. This makes it difficult to distinguish between real energy efficiency degradation and reasonable operating condition fluctuations when detecting anomalies, thus limiting the sensitivity and accuracy of diagnosis. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data to solve the problem that existing energy efficiency benchmarks are static and universal, unable to accurately adapt to complex and ever-changing real-time operating conditions, resulting in insufficient sensitivity and accuracy of anomaly detection.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data, which includes collecting and fusing multi-source heterogeneous data to form a standardized multi-source data event stream; By using causal discovery algorithms to automatically analyze standardized multi-source data event streams, a causal knowledge graph of energy efficiency at dry bulk terminals is constructed. Based on the physical layout of the terminal, equipment parameters and process logic, a basic digital twin is constructed. The quantitative causal logic in the energy efficiency causal knowledge graph of the dry bulk terminal is injected into the basic digital twin to form a terminal process digital twin enhanced with causal knowledge. Detailed parameters of the work tasks and real-time environmental data are input into a digital twin of the dock process enhanced with causal knowledge for simulation and deduction, generating personalized dynamic energy efficiency benchmark curves. During the actual execution of the task, real-time energy efficiency data is collected and compared with the personalized dynamic energy efficiency benchmark curve to detect abnormal energy efficiency events. Using the energy efficiency causal knowledge graph of dry bulk cargo terminals as a reasoning map, intelligent backtracking analysis is performed on energy efficiency anomalies to locate and quantify root cause variables, and a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables is integrated and output.

[0007] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method involves: collecting and fusing multi-source heterogeneous data to form a standardized multi-source data event stream, including the following steps: Multi-source heterogeneous data is obtained from the equipment monitoring unit, production management unit, video monitoring unit, environmental monitoring unit, and energy metering unit of the dry bulk terminal, and the data is cleaned. After data cleaning, multi-source heterogeneous data is mapped and bound to entities in the digital twin model. The mapped and bound multi-source heterogeneous data is then associated and fused based on spatiotemporal labels and business semantics to form a standardized multi-source data event stream.

[0008] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method includes: automating the analysis of standardized multi-source data event streams using a causal discovery algorithm to construct a causal knowledge graph of dry bulk cargo terminal energy efficiency, comprising the following steps: From standardized multi-source data event streams, fields describing equipment status, work instructions, environmental conditions, and energy consumption are extracted and reorganized into a sequence of causal variables; The stripped and recombined causal variable sequences are fed into a pre-screening layer with transit entropy as the correlation metric, and the information flow intensity between each pair of causal variable sequences is calculated. The pre-screened causal variable sequences enter the causal direction discrimination stage based on convergent cross mapping. The embedding theorem is used to verify the existence of a unidirectional driving relationship between the causal variable sequences in the reconstructed phase space. The verified unidirectional driving relationship is marked as a directed causal hypothesis. For each directed edge in the causal hypothesis network, a nonparametric Granger causality test combined with the Bayesian information criterion is used to quantify the predictive contribution of the causal variable sequence to the outcome causal variable sequence under the condition of controlling for confounding causal variable sequences, and this contribution is used as the causal strength coefficient. By combining the topology of the causal hypothesis network with the causal strength coefficients of each edge, a causal knowledge graph of energy efficiency for dry bulk cargo terminals is constructed.

[0009] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method includes: constructing a basic digital twin based on the terminal's physical layout, equipment parameters, and process logic; injecting the quantified causal logic from the dry bulk cargo terminal's energy efficiency causal knowledge graph into the basic digital twin to form a terminal process digital twin enhanced with causal knowledge; and comprising the following steps: Based on the physical layout diagram of the wharf, the geometric and dynamic parameter library of the equipment, and the process logic rule library of material transportation, a basic digital twin is constructed. The causal knowledge graph of energy efficiency in dry bulk terminals is analyzed, and the directed edge relationships in the graph are mapped to the dynamic influence links between corresponding entities in the basic digital twin. The quantitative causal strength coefficient attached to each directed edge in the causal knowledge graph of energy efficiency in dry bulk terminals is extracted. The quantified causal strength coefficient is transformed into a real-time correction parameter for the corresponding dynamic influence link in the basic digital twin; Using the real-time correction parameters after transformation, the simulation operation parameters and interaction rules of the corresponding entities in the basic digital twin are dynamically adjusted to form a digital twin of the dock process enhanced with causal knowledge.

[0010] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method includes the following steps: Detailed parameters of operational tasks and real-time environmental data are input into a digital twin of the terminal process enhanced with causal knowledge for simulation and deduction to generate a personalized dynamic energy efficiency benchmark curve. Receive detailed parameters of the work tasks, including cargo type, flow rate, equipment assignment and process path, as well as real-time environmental data obtained from the environmental monitoring terminal; Configure the detailed parameters of the task and real-time environmental data as the initial conditions and boundary constraints for the simulation of the digital twin of the dock process enhanced with causal knowledge. Under the initial conditions and boundary constraints of the simulation, the digital twin of the dock process enhanced with causal knowledge is driven to run at the virtual clock synchronization rate to deduce the complete execution process of the operation task in the virtual environment. During the simulation, the instantaneous power data of each energy-consuming virtual entity in the digital twin of the dock process, which has been enhanced with causal knowledge, is recorded at each virtual moment. The instantaneous power data stream recorded in the digital twin simulation, arranged according to virtual timestamps, is resampled and windowed according to the time resolution of the actual work plan to generate a simulation energy efficiency curve describing the change of theoretical energy consumption of the work task over time. Based on the start time of the work task plan, the simulated energy efficiency curve is aligned and mapped on the time axis to generate a personalized dynamic energy efficiency benchmark curve.

[0011] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method involves: real-time collection of actual energy efficiency data during the actual execution of operational tasks, including the following steps: When a task is started, the real-time data acquisition channel of all equipment on the corresponding process path is activated to capture the energy meter pulse signal and the running status word of the equipment. The pulse signal of the energy meter is translated into a cumulative energy reading through the pulse-energy conversion relationship and encapsulated into a time-stamped data frame. The encapsulated time-stamped data frames are sorted and cached according to the process flow order. The sorted and cached time-stamped data frames are output as actual energy efficiency data at fixed time intervals.

[0012] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method involves comparing actual energy efficiency data with a personalized dynamic energy efficiency benchmark curve to detect energy efficiency anomalies, including the following steps: The system matches the baseline energy efficiency value at the same time point from the personalized dynamic energy efficiency baseline curve, extracts the current energy consumption reading from the actual energy efficiency data, and forms a data pair with the matched baseline energy efficiency value at the same time point. The system judges the current energy consumption reading and the baseline energy efficiency value, and records the duration of the exceedance. When the duration of the excess exceeds the preset tolerance time window, a primary anomaly marker is triggered. The primary anomaly markers that are triggered are aggregated. If the density of primary anomaly markers exceeds the threshold within a specific process segment, an energy efficiency anomaly event is confirmed.

[0013] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method includes the following steps: using a causal knowledge graph of dry bulk cargo terminal energy efficiency as a reasoning map, performing intelligent backtracking analysis on energy efficiency anomalies, locating and quantifying root cause variables. Based on the energy efficiency causal knowledge graph of dry bulk cargo terminals, the terminal energy consumption nodes associated with energy efficiency anomalies are located, the primary cause nodes that directly affect the terminal energy consumption nodes are retrieved, and the actual observation sequence of the primary cause nodes during the energy efficiency anomaly event period is obtained. Extract the baseline expected value sequence of primary cause nodes on the personalized dynamic energy efficiency baseline curve within the same time period; By comparing the actual observation sequence of the primary cause node with its baseline expected value sequence, the primary cause node that causes the persistent deviation can be identified. Based on the causal influence intensity indicated in the causal knowledge graph of dry bulk terminal energy efficiency, the contribution score of each primary cause node of persistent deviation is evaluated. The primary cause nodes that cause persistent bias are sorted according to their contribution scores, and the sorted list of primary cause nodes is output as the root cause variable result.

[0014] As a preferred embodiment of the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data described in this invention, the method includes the following steps: integrating and outputting a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables: Collect information on the timing, location, and deviation descriptions of energy efficiency anomalies, and correlate and aggregate the contribution scores of root cause variables; The collected event and cause information is aligned and integrated according to the timeline. Based on the aligned and integrated information, a structured report draft is generated, and the structured report draft is rendered into a visual diagnostic report.

[0015] Secondly, the present invention provides an energy efficiency diagnosis system for dry bulk cargo terminals based on multi-source data, including a data processing module that collects and integrates multi-source heterogeneous data to form a standardized multi-source data event stream. The analysis module uses causal discovery algorithms to automatically analyze standardized multi-source data event streams and construct a causal knowledge graph of energy efficiency for dry bulk terminals. The module constructs a basic digital twin based on the terminal's physical layout, equipment parameters, and process logic. It then injects the quantified causal logic from the dry bulk terminal's energy efficiency causal knowledge graph into the basic digital twin, forming a terminal process digital twin enhanced with causal knowledge. The simulation module inputs detailed parameters of the operation task and real-time environmental data into the digital twin of the dock process enhanced with causal knowledge for simulation and simulation, and generates personalized dynamic energy efficiency benchmark curves. The inspection module collects actual energy efficiency data in real time during the actual execution of the task, compares the actual energy efficiency data with the personalized dynamic energy efficiency benchmark curve, and detects abnormal energy efficiency events. The quantification module uses the energy efficiency causal knowledge graph of dry bulk cargo terminals as the reasoning map to conduct intelligent backtracking analysis on energy efficiency anomalies, locate and quantify root cause variables, and integrate and output a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables.

[0016] The beneficial effects of this invention are as follows: By integrating heterogeneous data from multiple sources such as equipment, production, and environment into a standardized event stream, and using causal discovery algorithms to automatically construct a dynamic knowledge graph that quantifies the causal relationships between equipment, environment, operation, and energy efficiency, the causal logic in this graph is injected into a basic digital twin built on a physical model, forming a causal knowledge-enhanced twin that can integrate data-driven real-world patterns. Real-time parameters and environmental data are injected into each specific task, and high-fidelity, situation-adaptive, personalized dynamic energy efficiency benchmark curves are generated through simulation and deduction of the enhanced twin. During operation, by comparing real-time data with personalized benchmarks, real energy efficiency anomalies are sensitively detected. Using the causal knowledge graph as a reasoning map, anomalies are intelligently traced back, the contribution of root cause variables is located and quantified, and a visual diagnostic report is output. This achieves a leap from static universal assessment to dynamic personalized diagnosis, and from phenomenon monitoring to causal tracing, improving the sensitivity of energy efficiency anomaly detection, the accuracy of location, and the effectiveness of decision support under complex operating conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data.

[0019] Figure 2 This is a schematic diagram of an energy efficiency diagnostic system for dry bulk cargo terminals based on multi-source data. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0023] Reference Figures 1-2 This is one embodiment of the present invention, which provides a method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data, including the following steps: S1. Collect and merge multi-source heterogeneous data to form a standardized multi-source data event stream.

[0024] S1.1 Obtain multi-source heterogeneous data from the equipment monitoring unit, production management unit, video monitoring unit, environmental monitoring unit, and energy metering unit of the dry bulk terminal, and perform data cleaning; Furthermore, data acquisition involves obtaining equipment operating status and process parameters from the equipment monitoring unit of the dry bulk terminal, work plans and scheduling instructions from the production management unit, images and video streams from the video monitoring unit, meteorological and dust concentration data from the environmental monitoring unit, and electricity consumption readings from the energy metering unit. Data cleaning is performed on this diverse and heterogeneous multi-source data, including removing outliers that significantly exceed the physical range, filling in random missing values ​​caused by brief communication interruptions using linear interpolation of data from previous and subsequent times, and resampling data streams with different sampling frequencies based on the lowest common frequency to achieve preliminary timestamp alignment. This results in cleaned multi-source heterogeneous data. S1.2. Map and bind the cleaned multi-source heterogeneous data with entities in the digital twin model. After mapping and binding, the multi-source heterogeneous data is associated and fused with spatiotemporal labels and business semantics to form a standardized multi-source data event stream.

[0025] Furthermore, the cleaned multi-source heterogeneous data is mapped and bound to entities in the digital twin model. A mapping table is established, which clearly defines which specific equipment entity or component attribute in the digital twin model corresponds to each sensor signal from the equipment monitoring unit, and which segment of process logic in the digital twin model triggers each work instruction from the production management unit. The multi-source heterogeneous data that has been mapped and bound is then associated and fused based on spatiotemporal tags and business semantics. The fusion of spatiotemporal tags refers to assigning all data entries a precise timestamp based on a unified clock source and geographical coordinates based on the unified coordinate system of the wharf. The fusion of business semantics refers to logically grouping and associating data entries describing the same business activity, such as a ship unloading instruction, the corresponding ship unloader current data, and the wind speed data at that moment, based on process flow knowledge. The cleaned, mapped, bound, and spatiotemporally and business-related data entries are arranged in chronological order to form a standardized multi-source data event stream with spatiotemporal context and business semantics.

[0026] S2. Utilize causal discovery algorithms to automatically analyze standardized multi-source data event streams and construct a causal knowledge graph of energy efficiency for dry bulk cargo terminals.

[0027] S2.1 Extract and reorganize the fields describing equipment status, work instructions, environmental conditions and energy consumption from the standardized multi-source data event stream into a causal variable sequence; Furthermore, from the standardized multi-source data event stream, based on predefined data pattern recognition rules, fields describing equipment status such as motor current and bearing temperature, fields describing work instructions such as work order number and flow rate setpoint, fields describing environmental conditions such as wind speed and humidity, and fields describing energy consumption such as total active power and sub-meter electricity consumption are extracted. These extracted field values ​​are reorganized according to their inherent timestamp order, and each field forms an independent and equally time-interval time series of data, i.e., a causal variable sequence, such as the causal variable sequence of the rotary motor current of the No. 1 stacker-reclaimer, the causal variable sequence of the instantaneous flow rate of the BC5 belt conveyor, the causal variable sequence of the instantaneous wind speed at the wharf front, and the causal variable sequence of the total active power of the feeder of the No. 3 substation.

[0028] Specifically, semantically precise extraction and recombination are performed from event streams to construct dedicated data entities for causal relationship discovery. Traditional data analysis often operates directly on raw data columns, ignoring the different roles of semantically different data in causal analysis. Through extraction and recombination, candidate causes (equipment, environment, instructions) and candidate effects (energy consumption) are clearly distinguished. The existing spatiotemporal alignment and business semantic foundation of the multi-source data event stream are standardized. Based on the dimensional division of business understanding, it ensures that the causal variable sequence not only contains numerical values ​​but also inherits clear physical or business meanings. This makes the causal relationships discovered subsequently directly interpretable and avoids the problem of mining meaningless statistical associations from mixed data. For example, wind speed is extracted from the event stream as an independent sequence, rather than being mixed with equipment status, so that the algorithm can clearly explore the direct impact of wind speed on energy consumption, rather than the indirect association presented by other intermediate variables.

[0029] S2.2 The stripped and recombined causal variable sequences are fed into a pre-screening layer with transit entropy as the correlation metric, and the information flow strength between each pair of causal variable sequences is calculated. Furthermore, the stripped and recombined causal variable sequences are fed into a pre-screening layer that uses transit entropy as the correlation metric. For any pair of causal variable sequences, such as the causal variable sequences on the cause side... and the causal variable sequence on the outcome side Calculate from arrive Transitive entropy And from arrive Transitive entropy The transfer entropy is calculated by estimating the probability function using statistical methods. This involves the value of the result sequence at the next time step, the historical state vector of the result sequence itself, and the historical state vector of the causal sequence; the information flow intensity, i.e., the transfer entropy value, is calculated based on comparing the amount of uncertainty reduction in the future value of the predicted result after adding the historical information of the cause, given the historical conditions of the result itself; after the calculation is completed, a threshold is set for the information flow intensity of all sequence pairs, and only causal variable sequence pairs whose information flow intensity exceeds this threshold are retained, and they are considered to have a statistical dependency relationship, thus passing the pre-screening.

[0030] Specifically, transfer entropy is used as a pre-screening metric, and a pre-screening layer is constructed as a filtering mechanism. Traditional causal discovery algorithms (such as the PC algorithm) have high computational complexity and are easily affected by a large number of weakly correlated or irrelevant variables when dealing with high-dimensional, multivariate time-series data from dry bulk terminals. Before formally determining the causal direction, transfer entropy is used to quantify the intensity of directional information flow between variables. Transfer entropy can capture nonlinear and non-Gaussian dependencies, which is more adaptable to the interactions under complex terminal conditions than traditional linear correlation or covariance. This reduces the number of variable pairs that need to enter the subsequent complex causal direction determination stage, focuses on candidate pairs that may have strong causal relationships, improves the overall analysis efficiency, and reduces the risk of misjudging the causal direction in noise. Essentially, it is a two-stage strategy of coarse screening followed by fine judgment, ensuring the feasibility and robustness of the method in complex industrial scenarios.

[0031] The expression for information flow strength is: ; in, For the causal variable sequence from the cause side causal variable sequence to the outcome side Information flow intensity, For the cause-and-effect variable sequence, For the causal variable sequence on the outcome side, Let be a probability function. For the causal variable sequence on the outcome side at the next future time... The value of , Let be the state vector of the causal variable sequence on the outcome side at the current and past time points. causal variable sequence for the cause side The state vectors at the current and past moments. causal variable sequence for the outcome side Embedding dimension, causal variable sequence for the cause side Embedding dimension, For discrete time indexes, For the current analysis time The next moment in the future.

[0032] S2.3. The pre-screened causal variable sequences enter the causal direction discrimination stage based on convergent cross mapping. The embedding theorem is used to verify the existence of a unidirectional driving relationship between the causal variable sequences in the reconstructed phase space. The verified unidirectional driving relationship is marked as a directed causal hypothesis. Furthermore, the pre-screened causal variable sequences enter the causal direction discrimination stage based on convergent cross-mapping. For each pair of pre-screened sequences, such as causal variable sequence C and causal variable sequence D, the phase space is reconstructed based on them. In the phase space reconstructed by causal variable sequence C, it is examined whether the state point of causal variable sequence D can be predicted more accurately; conversely, in the phase space reconstructed by causal variable sequence D, it is examined whether the state point of causal variable sequence C can be predicted more accurately. According to the principle of convergent cross-mapping, if causal variable sequence C is the cause of causal variable sequence D, then in the phase space reconstructed by causal variable sequence D, the prediction skill of causal variable sequence C should converge and improve as the embedding dimension used for reconstruction increases. By comparing the convergence and convergence level of the bidirectional prediction skill, the unidirectional driving relationship between causal variable sequences is verified, for example, verifying that the wind speed causal variable sequence unidirectionally drives the belt conveyor power causal variable sequence. Each verified unidirectional driving relationship is labeled as a directed causal hypothesis.

[0033] Specifically, the application of convergent cross-mapping for causal direction determination effectively addresses the issues of confounding variables and common drivers. Traditional Granger causality tests assume linearity and are sensitive to unobserved confounding factors. In the dockside scenario, many variables are affected by common unobserved factors (such as grid voltage fluctuations and management strategies). Based on dynamical system theory, the principle that the shadows of causal variables can be better reconstructed in the phase space of the outcome variable is used to determine the causal direction. This can reveal causal relationships in nonlinear dynamical systems. For example, it can effectively distinguish whether an increase in wind speed leads to an increase in conveyor belt power, or whether both change simultaneously simply due to increased workload. This results in a more reliable determination of causal direction, laying a solid foundation for constructing an accurate causal knowledge graph.

[0034] S2.4 For each directed edge in the causal hypothesis network, a nonparametric Granger causality test combined with the Bayesian information criterion is used to quantify the predictive contribution of the causal variable sequence to the outcome causal variable sequence under the condition of controlling for confounding causal variable sequences, and this contribution is used as the causal strength coefficient. Furthermore, for each directed edge in the causal hypothesis network, such as the directed edge from the causal variable sequence E on the cause side to the causal variable sequence F on the effect side, a nonparametric Granger causality test combined with the Bayesian information criterion is used to quantify the causal strength coefficient. This process first constructs a prediction model for the causal variable sequence F, including historical values ​​of the causal variable sequence E and other confounding variables, while controlling for other relevant confounding causal variable sequences. Simultaneously, a restricted prediction model is constructed that does not include historical values ​​of the causal variable sequence E. The nonparametric Granger causality test evaluates the predictive contribution of the causal variable sequence E by comparing the prediction errors of the complete model and the restricted model. The Bayesian information criterion is used to strike a balance between model complexity and goodness of fit, helping to determine the optimal lag number for the prediction model. Finally, the predictive contribution of the causal variable sequence E to the causal variable sequence F is calculated and standardized, serving as the causal strength coefficient of that directed edge.

[0035] Specifically, this method integrates the nonparametric Granger causality test with the Bayesian information criterion to achieve robust causal strength quantification. The use of nonparametric tests avoids biases caused by incorrectly specified parameter models (such as linear autoregression) and is better suited to the complex nonlinear relationships in port data. Combined with the Bayesian information criterion, it automatically and objectively determines the optimal historical lag number, solving a key and challenging parameter selection problem in time series analysis. This ensures the robustness and reliability of the quantification results, enabling unbiased and adaptive estimation of the strength of specific causal relationships while controlling for potential confounding factors. The resulting causal strength coefficient is not merely a statistic but a crucial bridge between qualitative causal discovery and quantitative causal applications.

[0036] S2.5. Combine the topology of the causal hypothesis network with the causal strength coefficients of each edge to construct a causal knowledge graph of energy efficiency for dry bulk cargo terminals.

[0037] Furthermore, all directed edges and their connected nodes in the causal hypothesis network that have undergone direction discrimination and intensity quantification are integrated. Here, nodes represent different sequences of causal variables, directed edges represent verified causal relationships, and each edge is attached with a quantified causal strength coefficient. The network structure that integrates nodes, directed edges, and the causal strength coefficients attached to the edges is constructed as a causal knowledge graph for energy efficiency in dry bulk cargo terminals. This graph, in the form of a graph structure, intuitively and quantitatively expresses the potential causal relationship network between various equipment parameters, operating instructions, environmental factors, and energy efficiency indicators in dry bulk cargo terminals.

[0038] Specifically, discrete causal findings are integrated into a structured, computable knowledge graph. Traditional causal analysis outputs are often isolated lists or matrices of causal relationships, lacking overall structure and computability. By using a graph model knowledge graph, which serves as a form of knowledge representation, not only stores information about cause and effect and the extent of impact, but its graph structure also enables graph-based traversal, reasoning, and querying, allowing knowledge to move from qualitative to quantitative. This completes the construction of a causal knowledge graph for energy efficiency at dry bulk terminals.

[0039] S3. Based on the physical layout of the terminal, equipment parameters and process logic, construct a basic digital twin, and inject the quantitative causal logic in the energy efficiency causal knowledge graph of the dry bulk terminal into the basic digital twin to form a terminal process digital twin enhanced with causal knowledge.

[0040] S3.1 Based on the physical layout diagram of the wharf, the geometric and dynamic parameter library of the equipment, and the process logic rule library of material transportation, construct a basic digital twin.

[0041] Furthermore, based on the physical layout of the wharf, static three-dimensional geometric models of the wharf area, berths, storage yards, roads, and buildings are established in the simulation environment. Based on the geometric and dynamic parameter library of the equipment, driveable three-dimensional models containing precise dimensions, kinematic joints, mass, inertia, and dynamic characteristics of motors and transmission mechanisms are created for major process equipment such as ship unloaders, ship loaders, stacker-reclaimers, and belt conveyors. Based on the process logic rule library of material conveying, the conveying path logic, equipment start-up and shutdown and interlocking rules, and material flow conservation and transfer logic are defined for the material flow from the unloading point through the belt conveyor to the storage yard and then through the reclaiming and loading process. The static three-dimensional geometric model, the driveable three-dimensional model, and the process logic rules are integrated and coupled, enabling the virtual equipment to move and process virtual material flow in the virtual space according to the process logic, thereby constructing a basic digital twin with geometric, physical, and logical simulation capabilities.

[0042] Specifically, this is reflected in the modeling and coupling of the process logic rule base, which goes beyond simple 3D visualization or equipment dynamics simulation. By defining rules such as material conveying paths and equipment interlocking, the basic digital twin can simulate the overall dynamic behavior of the terminal as a complex logistics system, rather than the movement of isolated equipment. For example, it can simulate the working condition of downstream conveyor belts running idle and waiting for materials due to upstream equipment failure, demonstrating its ability to simulate system-level interactions.

[0043] S3.2 Analyze the causal knowledge graph of energy efficiency in dry bulk cargo terminals, map the directed edge relationships in the graph to the dynamic influence links between corresponding entities in the basic digital twin, and extract the quantified causal strength coefficient attached to each directed edge in the causal knowledge graph of energy efficiency in dry bulk cargo terminals.

[0044] Furthermore, the energy efficiency causal knowledge graph of the dry bulk terminal is analyzed to identify the physical or business entity corresponding to the causal variable sequence represented by each node in the graph. For example, the wind speed causal variable sequence node is mapped to the environmental wind field simulation entity in the basic digital twin, and the BC5 belt conveyor drive power causal variable sequence node is mapped to the BC5 belt conveyor drive motor entity. Then, each directed edge in the graph is traversed, and the directed edge relationship is mapped to a dynamic influence link between the corresponding two entities in the basic digital twin. For example, the directed edge from wind speed to BC5 belt conveyor drive power is mapped to a dynamic influence link from the environmental wind field simulation entity to the BC5 belt conveyor drive motor entity. The attached quantitative causal strength coefficient is extracted from each directed edge of the energy efficiency causal knowledge graph of the dry bulk terminal. This coefficient represents the magnitude of the influence of the cause node on the result node.

[0045] Specifically, it establishes a mapping mechanism from statistical causal relationships to simulation model parameters. By analyzing the causal knowledge graph, it explicitly maps statistically discovered, potentially implicit, and difficult-to-express causal relationships (such as the quantitative impact of wind speed on the power of a belt conveyor in a specific open section) into dynamic influence links between entities in the simulation model. This not only connects entities but, more importantly, establishes an influence channel, enabling the subsequent quantitative causal strength coefficients to dynamically adjust the simulation through this channel. This is equivalent to grafting data-driven neural networks onto the mechanistic model, giving it the ability to learn and integrate complex relationships in the real world.

[0046] S3.3. Transform the quantified causal intensity coefficient into the real-time correction parameter of the corresponding dynamic influence link in the basic digital twin.

[0047] Furthermore, the quantified causal strength coefficient is transformed into a real-time correction parameter for the corresponding dynamic influence link in the basic digital twin. This transformation process is based on the simulation model characteristics of the two entities connected by the dynamic influence link. For example, for the dynamic influence link from the environmental wind field simulation entity to the BC5 belt conveyor drive motor entity, its quantified causal strength coefficient describes the sensitivity of wind speed changes to the drive power. This coefficient is transformed into a real-time correction function or correction multiplier. The input of the real-time correction function is the current simulated wind speed value output by the environmental wind field simulation entity, and the output is a correction amount for the theoretical load torque or resistance coefficient of the BC5 belt conveyor drive motor entity. The transformation process ensures that the statistical meaning of the quantified causal strength coefficient is correctly translated into executable parameter adjustment logic in the simulation model.

[0048] Specifically, by translating abstract statistical coefficients into executable simulation correction logic, the computational injection of causal knowledge is realized. The quantification of causal strength coefficient is itself just a numerical value, such as a regression coefficient. Depending on the model characteristics of the entities at both ends of the link, for example, for a mechanical model, the coefficient may be transformed into a correction multiplier for the drag coefficient; for a thermodynamic model, it may be transformed into a correction term for the heat transfer coefficient. This requires a deep understanding of the intrinsic variables and interaction interfaces of the simulation model. The translation mechanism enables the non-mechanical correlation patterns mined from the data to be encoded into the simulation kernel based on physical laws, thereby reproducing the complex coupling effects observed in reality in the simulation. This realizes the core operation of causal knowledge enhancement, allowing the digital twin to evolve from an ideal physical model into a hybrid model that integrates real-world laws.

[0049] S3.4 Using the real-time correction parameters after the transformation, dynamically adjust the simulation operation parameters and interaction rules of the corresponding entities in the basic digital twin to form a dock process digital twin enhanced with causal knowledge.

[0050] Furthermore, using the transformed real-time correction parameters, the simulation operation parameters and interaction rules of the corresponding entities in the basic digital twin are dynamically adjusted. During simulation, when the state of one entity in the dynamic influence link changes, the real-time correction parameter is activated and a correction value is obtained. This correction value is immediately applied to the relevant simulation operation parameters of the entity at the other end of the link, or the interaction rules between the two entities are modified. For example, when the environmental wind field simulation entity calculates that the current wind speed increases, the load torque correction increment for the BC5 belt conveyor drive motor entity is calculated through the corresponding real-time correction function, and this increment is immediately added to the theoretical load torque currently calculated by the motor entity, thereby dynamically simulating the phenomenon that the increase in wind speed leads to an increase in drive power. All dynamic influence links operate in this way, so that the simulation behavior of the basic digital twin is continuously controlled in real time by the quantified causal logic in the causal knowledge graph of dry bulk terminal energy efficiency, forming a terminal process digital twin enhanced with causal knowledge.

[0051] Specifically, by embedding real-time correction logic into the simulation loop, the impact of causal knowledge is made immediate, continuous, and linked to the simulation state, constructing a micro-closed loop of perception-computation-correction: entities in the simulation environment perceive state changes (such as wind speed changes), and through the mapped dynamic influence links and the transformed real-time correction parameters, derive the correction amount for other entity parameters and apply it immediately, thereby changing the simulation evolution trajectory. This makes the digital twin of the dock process enhanced with causal knowledge no longer a fixed model, but an intelligent simulation entity with conditioned reflex capabilities that can dynamically simulate complex causal influences in reality. For example, it can simulate the nonlinear impact of different wind speeds and different material humidity combinations on overall energy consumption in real time, and the simulated energy consumption trajectory will be closer to the complex response of the real world.

[0052] S4. Input the detailed parameters of the operation task and real-time environmental data into the digital twin of the dock process enhanced with causal knowledge for simulation and deduction, and generate a personalized dynamic energy efficiency benchmark curve.

[0053] S4.1 Receive detailed parameters of the work tasks, including cargo type, flow rate, equipment assignment, and process path, as well as real-time environmental data obtained from the environmental monitoring terminal.

[0054] Furthermore, the system receives detailed operational parameters from the production management unit, including cargo type, planned flow rate, designated unloader and loader numbers, and the complete process path from a berth through a specific conveyor belt to a storage yard and then through another process to the loading point. It also obtains real-time environmental data, including wind speed, wind direction, temperature, and humidity, from environmental monitoring terminals deployed at the dock site.

[0055] Specifically, high-fidelity, high-granularity scenario inputs are collected for simulation. Unlike traditional methods that use static averages or typical operating conditions, this approach receives and integrates two types of dynamic inputs: task parameters and real-time environment. Detailed task parameters define the simulation script, while real-time environmental data provides the stage background for the simulation. For example, for the task of unloading 100,000 tons of iron ore from berth 5 to yard A, not only are the type and quantity of the cargo known, but also the specific unloader and conveyor belt path to be used. Combined with the actual wind speed at the dock at that moment, this refined input ensures that the simulation is tailored to the specific needs of each task, responding to subtle differences in equipment assignment and path selection, as well as instantaneous environmental fluctuations, thus generating a personalized benchmark rather than a universal one.

[0056] S4.2 Configure the detailed parameters of the operation task and the real-time environmental data as the simulation initial conditions and boundary constraints of the digital twin of the dock process enhanced with causal knowledge.

[0057] Furthermore, the detailed parameters of the operation tasks and real-time environmental data are configured as the initial conditions and boundary constraints for the simulation of the terminal process digital twin enhanced with causal knowledge. The cargo type and flow parameters are transformed into the attributes and generation rate of the virtual material flow. The equipment assignment parameters are mapped to activate the corresponding equipment model in the digital twin and set its initial state. The process path parameters are transformed into control logic that drives the virtual material to move along the specified path in the digital twin. Real-time environmental data, such as wind speed and temperature values, are directly assigned to the corresponding environmental simulation entity in the digital twin as dynamic boundary conditions for the simulation.

[0058] Specifically, a parameter configuration engine was established to achieve dynamic mapping and configuration from production instructions and environmental readings to identifiable parameters of the simulation model. This engine automatically translates business-level descriptions (such as using Ship Unloader No. 3) and physical readings (such as southeast wind at level 5) into specific variable assignments and logical triggers required to drive the simulation model. This allows the digital twin of the terminal process, enhanced with causal knowledge, to be quickly and accurately set to a virtual starting state and external conditions that are completely consistent with the actual operation that is about to occur. The dynamic configuration capability transforms the digital twin from a fixed demonstration model into a predictive tool that can be initialized on demand, making it possible to perform one-time, high-fidelity pre-simulation simulations.

[0059] S4.3 Under the initial conditions and boundary constraints of the simulation, drive the digital twin of the dock process enhanced with causal knowledge to run at the virtual clock synchronization rate, and deduce the complete execution process of the operation task in the virtual environment.

[0060] Furthermore, under the initial simulation conditions and boundary constraints, the digital twin of the dock process, enhanced with causal knowledge, is driven to run at a virtual clock synchronization rate. The virtual clock advances in fixed steps. Within each clock step, all virtual entities in the digital twin perform state calculations and interactions based on their dynamic equations, process logic, and injected causal correction logic. This process is continuously simulated to depict the complete execution process in the virtual environment from the start of the operation, equipment startup, material flow, process connection, to the completion of the task.

[0061] Specifically, the core execution link of predictive simulation using enhanced digital twins combines causal knowledge enhancement with virtual clock synchronization. Traditional simulations are often based on ideal models, and the simulation results may deviate significantly from reality. During simulation, the dynamic influence links injected by the causal knowledge graph within the digital twin continuously operate, adjusting equipment parameters (such as the load of virtual motors) in real time according to the simulated environmental state (such as virtual wind speed). The simulation is no longer an open-loop, deterministic script playback, but a closed-loop, data-driven, causal logic-dynamically adjusted virtual experiment. For example, in the simulation, when the virtual wind field is enhanced, the virtual drive motor of the open-air section belt conveyor will experience an additional load defined by causal knowledge, thereby consuming more virtual power. This allows the deduced energy consumption process to reflect complex dynamic coupling effects, improving the fidelity of the simulation.

[0062] S4.4 During the simulation, record the instantaneous power data of each energy-consuming virtual entity in the digital twin of the dock process enhanced with causal knowledge at each virtual moment.

[0063] Furthermore, during the simulation, the instantaneous power data of each energy consumption virtual entity in the digital twin of the dock process enhanced with causal knowledge is recorded at each virtual moment; the energy consumption virtual entities include all drive motor models, lighting and auxiliary equipment models; at the end of each virtual clock step, the instantaneous power consumption values ​​obtained from these entity models are synchronously collected and recorded together with the current virtual timestamp to form an instantaneous power data sequence strictly ordered by virtual timestamp.

[0064] Specifically, it captures fine-grained theoretical energy consumption trajectories from high-fidelity simulations, performs instrumented measurements on the simulation process on a virtual time scale, and sets data acquisition points on virtual entities, just like installing power sensors on real equipment, recording their theoretical energy consumption at the same virtual clock frequency. This ensures that the recorded instantaneous power data is completely synchronized with the dynamic process of the simulation, including all the details of energy consumption fluctuations caused by process logic, equipment interaction, and causal reinforcement logic. For example, it can record instantaneous power spikes caused by events such as the start-up of the virtual material handling machine and the start-up of the heavy-load section of the virtual conveyor belt. The detailed recording provides a rich data foundation for the subsequent construction of energy efficiency curves, enabling the generated benchmark curves to reflect the theoretical dynamic changes during operation, rather than just a total energy consumption or average power.

[0065] S4.5. The instantaneous power data stream recorded in the digital twin simulation and arranged according to the virtual timestamp is resampled and windowed according to the time resolution of the actual work plan to generate a simulation energy efficiency curve that describes the change of theoretical energy consumption of the work task over time.

[0066] Furthermore, the instantaneous power data of all energy-consuming virtual entities are summed at each virtual moment to obtain the total instantaneous power of the digital twin at that virtual moment. The total instantaneous power data stream is resampled according to the time interval matching the actual energy efficiency data acquisition, and the power average value within each resampling time window is obtained or the energy value is obtained by integrating the power over time. By connecting the power or energy values ​​corresponding to all resampling time points, a time series curve is formed, which is the simulated energy efficiency curve describing the change of energy consumption of the task over time under theoretical conditions.

[0067] Specifically, the raw simulation data undergoes engineering processing to match the data format of actual management. The virtual clock step of the simulation may be very small, generating massive amounts of high-frequency data, but actual energy efficiency management usually focuses on energy consumption on a minute-level or longer time scale. Through resampling and window accumulation, the micro-details of the simulation are aggregated into a macro-level, manageable energy efficiency trajectory. The aggregation is based on the time resolution of the actual work plan. The generated simulation energy efficiency curve is comparable to the actual energy efficiency data to be collected in the subsequent time scale, ensuring that the baseline curve and the actual data are measured under the same time scale and energy scale, avoiding comparison distortion caused by inconsistent data frequencies.

[0068] S4.6. Using the start time of the work task plan as a reference, align and map the simulated energy efficiency curve on the time axis to generate a personalized dynamic energy efficiency benchmark curve.

[0069] Furthermore, using the planned start time of the task as a benchmark, the simulated energy efficiency curve is aligned and mapped on the time axis; the planned start time of the task in the production scheduling plan is obtained, and this time point is used as the zero point of the time axis of the simulated energy efficiency curve; the entire time series of the simulated energy efficiency curve is shifted to the absolute time axis starting from the planned start time, so that the absolute timestamp of each point on the simulated energy efficiency curve corresponds to the predicted time on the planned execution timeline of the task; the simulated energy efficiency curve after time alignment and mapping is defined as the personalized dynamic energy efficiency benchmark curve for this specific task.

[0070] Specifically, the simulation runs independently of real-time operation, and its internal time (virtual time) is relative. By binding the starting point of the simulation curve to a future, definite point in time—the planned start time of the task—the conversion from relative time to absolute time is achieved. This generates a predictive benchmark: it explicitly predicts how theoretical energy consumption will evolve over time at a specific future moment (the planned start time). This makes the personalized dynamic energy efficiency benchmark curve not just a theoretical curve, but a predicted trajectory anchored to the real-world timeline for specific future events. When actual operations begin, the collected actual energy efficiency data can be compared point-by-point and moment-by-moment with this benchmark curve, which is already positioned on the absolute time axis, thereby achieving accurate and real-time detection of energy efficiency anomalies.

[0071] S5. During the actual execution of the work task, collect actual energy efficiency data in real time.

[0072] S5.1 When a task is started, activate the real-time data acquisition channel of all equipment on the corresponding process path to capture the energy meter pulse signal and operating status word of the equipment.

[0073] Furthermore, when a task is started, based on the process path parameters of the task, all the equipment involved in the path are identified, including ship unloaders, ship loaders, stacker-reclaimers, belt conveyors and their drive stations. Instructions are sent to the monitoring network to which these equipment belong to activate the real-time pulse output channel of the smart energy meter associated with each piece of equipment and the real-time status register reading channel of the equipment controller. After activation, pulse signals representing energy accumulation are continuously captured from the energy meter channel, and operating status words containing information such as running, stopping, fault, and load are captured from the equipment controller channel.

[0074] Specifically, this involves a precise data acquisition strategy based on process paths and activated on demand. Unlike traditional continuous 24 / 7 data acquisition across all equipment, this strategy employs event-triggered acquisition, binding data acquisition initiation to specific job tasks. Only when a job task begins are the acquisition channels of all relevant equipment along its process path activated. This optimizes the allocation of acquisition resources, avoids collecting large amounts of irrelevant data, improves processing efficiency, and ensures that the acquired data is strictly related to the current task in terms of business semantics. It simultaneously captures energy and state signals—specifically, power consumption signals—and equipment operating status. The combination of these two is crucial for understanding whether energy consumption is abnormal. For example, associating high energy consumption with idle operation can immediately identify a typical energy efficiency anomaly. This task-path-equipment-acquisition linkage mechanism is the core of achieving correlated data acquisition.

[0075] S5.2 The pulse signal of the electricity meter is translated into a cumulative electrical energy reading through the pulse-energy conversion relationship and encapsulated into a time-stamped data frame. The encapsulated time-stamped data frames are sorted and cached according to the process flow order.

[0076] Furthermore, the captured energy meter pulse signals are translated into cumulative energy readings through a pulse-energy conversion relationship. This relationship means that each pulse represents a fixed energy value. By accumulating the number of pulses received within a fixed time window and multiplying by this fixed energy value, the cumulative energy reading within that time window is obtained. Subsequently, the cumulative energy reading is packaged with the operating status word captured from the device controller within the same time window, and this data packet is given a precise timestamp based on a unified clock source, encapsulating it into a time-stamped data frame. The encapsulated time-stamped data frame is placed into a first-in-first-out (FIFO) buffer queue for sorting and caching according to the order of its source equipment in the process flow. For example, data frames from the upstream ship unloader are placed before data frames from the downstream stacker-reclaimer.

[0077] Specifically, the raw signals are encapsulated with business semantics and ordered according to process timing. Simple pulse counting is merely a physical signal; it is converted into a cumulative electrical energy reading with clear physical meaning, combined with an operating status word characterizing equipment behavior, encapsulated in a unified data frame structure, and given an authoritative timestamp. This is equivalent to giving each segment of energy consumption data a complete description of who consumed it, when it occurred, how much was consumed, and what state it is in. Ordering and caching according to the process flow sequence is a more insightful design. In continuous transport dock processes, the flow of materials and the transfer of energy have a clear order and slight delays. Ordering data frames according to the process sequence logically reconstructs the sequence of energy transfer along the process path. For example, an energy efficiency anomaly event in upstream equipment will have its data frame processed before the data frames of the downstream equipment affected by it. This facilitates causal correlation analysis based on event sequence. Encapsulation and ordering elevate the raw data stream into a structured information stream with spatiotemporal and business logic.

[0078] S5.3. After sorting and caching, the time-stamped data frames are output as actual energy efficiency data at fixed time intervals.

[0079] Furthermore, the sorted and cached time-stamped data frames are output as actual energy efficiency data at fixed time intervals; a periodic time trigger is set, for example, triggered once every fixed time window; when the trigger is triggered, all time-stamped data frames received since the last trigger and which have been sorted are retrieved from the cache queue; these data frames are processed, and the difference between multiple cumulative energy readings of the same device within the time window is calculated as the energy consumption of the device within the window, and combined with the corresponding operating status word to form a structured energy consumption and status record; such records of all operating devices within the fixed time interval are collected and packaged into an actual energy efficiency data packet containing the time interval, energy consumption of each device, and status.

[0080] Specifically, through periodic triggering and aggregation, the transformation from continuous event streams to discrete management data is achieved while ensuring time synchronization. Data frames arrive asynchronously and continuously, but energy efficiency management and comparison require synchronous, periodic snapshots. This problem is solved by a fixed-interval triggering output mechanism. It defines a unified reporting rhythm, under which the energy consumption and status of all devices are aligned and aggregated. The advantages of this are: first, it generates data points with the same time resolution as the personalized dynamic energy efficiency benchmark curve, enabling real-time comparison; second, by acquiring the energy consumption difference within a fixed window, the cumulative amount is converted into an increment, more intuitively reflecting the energy consumption rate; third, it ensures that the output actual energy efficiency data is complete and synchronous in time, covering the performance of all monitored devices within the same time period.

[0081] S6. Compare the actual energy efficiency data with the personalized dynamic energy efficiency benchmark curve to detect abnormal energy efficiency events.

[0082] S6.1 Match the baseline energy efficiency value at the same time point from the personalized dynamic energy efficiency baseline curve, extract the current energy consumption reading from the actual energy efficiency data, and form a data pair with the matched baseline energy efficiency value at the same time point. Judge the current energy consumption reading and the baseline energy efficiency value, and record the duration of the excess. Furthermore, the system matches the benchmark energy efficiency value corresponding to the moment that is exactly the same as the current actual energy efficiency data timestamp from the personalized dynamic energy efficiency benchmark curve; extracts the total energy consumption reading of each relevant device or process at the current moment from the actual energy efficiency data; pairs the current energy consumption reading of each device or process with the corresponding benchmark energy efficiency value at the same time point matched from the personalized dynamic energy efficiency benchmark curve to form a series of data pairs for comparison; for each data pair, it determines whether the current energy consumption reading continuously exceeds the benchmark energy efficiency value; and records the continuous time length for which the current energy consumption reading exceeds the benchmark energy efficiency value by continuously monitoring the data pairs.

[0083] Specifically, traditional energy efficiency comparisons may use fixed thresholds or moving averages, lacking dynamism and contextual sensitivity. First, it utilizes personalized dynamic energy efficiency benchmark curves as dynamic and contextualized comparison standards. The benchmark value at each time point is theoretically derived for that specific operation and time condition. Second, it not only compares instantaneous values ​​but, more importantly, records the duration of exceedances, effectively filtering noise and instantaneous fluctuations. For example, a brief power spike caused by uneven material particle size (short-term exceedance) is a normal operating condition, while continuous high energy consumption caused by equipment bearing wear (long-term exceedance) is a true anomaly. By recording the duration, the judgment criteria are upgraded from whether an instantaneous exceedance occurs to how long the exceedance lasts, avoiding false alarms for brief disturbances.

[0084] S6.2 When the duration of the excess exceeds the preset tolerance time window, a primary anomaly marker is triggered. The triggered primary anomaly markers are collected. If the density of primary anomaly markers exceeds the threshold in a specific process segment, an energy efficiency anomaly event is confirmed.

[0085] Furthermore, when the duration of a data pair exceeding the limit exceeds a preset tolerance time window, a primary anomaly marker is triggered for the corresponding equipment or process location. During operation, all triggered primary anomaly markers are continuously collected. For specific, continuous process segments, such as the entire conveyor belt process from the ship unloader to a certain stockyard, the number of primary anomaly markers appearing within a set observation time range is counted. If the number density of primary anomaly markers in the specific process segment, i.e., the frequency of marker appearance per unit time, exceeds a preset judgment threshold, then an energy efficiency anomaly event is determined to have occurred in the specific process segment.

[0086] Specifically, a point-line-surface progressive anomaly confirmation logic is adopted, starting with the continuous exceedance of a single data pair (point) and progressing to the aggregation of exceedance events within a specific process segment (line / surface). Two layers of filtering mechanisms are implemented to confirm anomalies. The first layer is time-based persistence filtering, eliminating transient interference. The second layer is spatial / logical aggregation filtering (marker density threshold), filtering out isolated, potentially accidental, minor equipment-level anomalies. Only when multiple primary anomaly markers continuously exceed limits within a short period within a specific process segment is it confirmed as a significant energy efficiency anomaly. This reflects a profound understanding of the systemic nature of the terminal's process flow: genuine energy efficiency problems (such as abnormally increased resistance in a certain section of a conveyor belt) often affect multiple devices in the process or persist on a single device, rather than being an isolated, transient phenomenon. By detecting the aggregation of anomaly markers, energy efficiency anomalies reflecting systemic problems can be more reliably identified, rather than responding to occasional noise or minor deviations from individual sensors, improving the accuracy of anomaly confirmation and its focus on the real problem.

[0087] S7. Using the energy efficiency causal knowledge graph of dry bulk cargo terminals as a reasoning map, intelligent backtracking analysis is performed on energy efficiency anomalies to locate and quantify the root cause variables.

[0088] S7.1 Based on the energy efficiency causal knowledge graph of dry bulk cargo terminals, locate the terminal energy consumption nodes associated with energy efficiency anomalies, retrieve the primary cause nodes that directly affect the terminal energy consumption nodes, and obtain the actual observation sequence of the primary cause nodes during the energy efficiency anomaly event period.

[0089] Furthermore, based on the energy efficiency causal knowledge graph of the dry bulk terminal, the terminal energy consumption nodes directly associated with the specific process segment and time period in which the energy efficiency anomaly occurred are first located, such as the BC5 belt conveyor drive power node. In the energy efficiency causal knowledge graph of the dry bulk terminal, all primary cause nodes that are direct results of the terminal energy consumption node and have arrows pointing to it are retrieved. For example, multiple primary cause nodes such as drive motor bearing vibration, belt conveyor drum temperature, and ambient wind speed may be retrieved. From the standardized multi-source data event stream or real-time data cache, the actual observation value sequence of the primary cause nodes in the same time period when the energy efficiency anomaly occurred is obtained.

[0090] Specifically, by utilizing a pre-constructed causal knowledge graph as a navigation map, rapid and targeted root cause tracing of anomalies is achieved. Compared to blindly checking all possible variables, graph-based retrieval is a knowledge-driven and structured search. The graph defines a network of energy flow directions. When an anomaly is confirmed in a terminal energy consumption node (such as the power of a conveyor belt), the tracing is not random but proceeds backward along the causal directed edges. The retrieved primary cause nodes are all variables that have a direct and statistical impact on terminal energy consumption. This narrows the scope of investigation, focusing from hundreds or thousands of variables to a few key candidate causes. By utilizing existing relational network knowledge, the most likely culprit's direct associates are located, ensuring that root cause analysis starts with a high level of information from the outset, avoiding the inefficiency of enumerating a large number of possible causes in traditional fault tree analysis.

[0091] S7.2 Extract the baseline expected value sequence of primary cause nodes on the personalized dynamic energy efficiency baseline curve within the same time period.

[0092] Furthermore, the baseline expected value sequence of primary cause nodes on the personalized dynamic energy efficiency baseline curve is extracted within the same time period. Since the generation process of the personalized dynamic energy efficiency baseline curve also includes the contextualized expectations for each primary cause node, the theoretical expected value sequence corresponding to each primary cause node during the energy efficiency anomaly event period is derived from the simulation configuration and causal logic used when generating the curve. This sequence is the baseline expected value sequence of primary and secondary cause nodes.

[0093] Specifically, a normative baseline is established for each possible causal variable to obtain the theoretical value of that causal variable under the same operating conditions. This is achieved through reverse mapping of the generation logic of the personalized dynamic energy efficiency baseline curve. It recognizes that the premise of comparison is alignment. When analyzing the anomaly of the cause, it is not possible to simply use historical averages or fixed values ​​for comparison, because the cause itself is also affected by conditions such as the task and environment. For example, when operating in strong winds, the actual wind speed is already high, but this cannot be directly identified as an abnormal cause. It is necessary to see whether the actual wind speed exceeds the expected wind speed value at that time and during that operation. The expected value comes from the environmental prediction of the simulation model that integrates causal knowledge when generating the personalized dynamic energy efficiency baseline curve, ensuring that the root cause analysis focuses on the real and unexpected deviations after stripping away the influence of normal operating conditions.

[0094] S7.3. By comparing the actual observation sequence of the primary cause node with its baseline expected value sequence, the primary cause node that causes the persistent deviation is identified.

[0095] Furthermore, by comparing the actual observation sequence of each primary cause node with its baseline expected value sequence, the primary cause nodes that cause persistent deviations are identified. For each primary cause node, its actual observation sequence during the abnormal period is compared with the baseline expected value sequence at each time point. By analyzing the sign, magnitude, and persistence of the deviation, those primary cause nodes whose actual observations consistently deviate from the baseline expected values ​​are identified. Those primary cause nodes that do not show persistent deviations are considered to be performing normally in the current abnormal event and are excluded.

[0096] Specifically, by comparing actual and expected sequences, nodes exhibiting abnormal behavior are filtered from directly related causal nodes to examine whether their behavior is erratic. The difference comparison principle is applied: the actual value of a variable may be high, but if the expected value is also high (consistent with current operating conditions), then it is not abnormal; conversely, the actual value of a variable may be only moderate, but if the expected value is low, then this moderate value is abnormal. By comparing sequences, unexpected and persistent deviations are identified. For example, the actual value of a motor bearing temperature may not exceed the absolute alarm threshold, but if it consistently exceeds the expected temperature under that operating condition, it indicates an abnormal friction or cooling problem. Filtering based on deviation rather than absolute value can uncover early, hidden causes that do not exceed absolute limits but whose trends are unreasonable, improving the sensitivity and depth of root cause analysis.

[0097] S7.4. Based on the causal influence intensity indicated in the causal knowledge graph of dry bulk terminal energy efficiency, evaluate the contribution score of each primary cause node of the persistent deviation.

[0098] Furthermore, based on the causal influence intensity indicated in the dry bulk terminal energy efficiency causal knowledge graph, the contribution score of each primary cause node of a persistent deviation is evaluated. The quantified causal strength coefficient attached to the directed edges connecting each primary cause node of a persistent deviation to the terminal energy consumption node is retrieved from the dry bulk terminal energy efficiency causal knowledge graph. This causal strength coefficient describes the influence weight of the cause node on the result node in the statistical history. The deviation degree (e.g., average deviation amplitude, deviation duration) of each primary cause node of a persistent deviation is combined with the corresponding causal strength coefficient to calculate and evaluate the contribution score of each node to the current terminal energy consumption anomaly event. The contribution score comprehensively reflects the degree of deviation of the cause node itself and the magnitude of its inherent influence.

[0099] Specifically, by combining the current degree of deviation with the inherent historical influence, the contribution of causes can be quantitatively assessed. Simply looking at the degree of deviation, a variable with a small impact may have a large deviation, while a key variable may have only a slight deviation. By introducing a causal strength coefficient as a weight, which comes from the statistics of historical data, the causal strength coefficient reflects the inherent influence or sensitivity of the cause on the result. For example, the graph may show that the causal strength coefficient of the conveyor belt roller temperature on the drive power is high (sensitive), while the coefficient of the ambient humidity is low (insensitive). When assessing the contribution, a moderate deviation in roller temperature, due to its high weight, may contribute more to the final energy consumption anomaly than a large deviation in ambient humidity. The assessment method is more in line with reality: a variable that has a large impact on the system output may lead to output changes even if its change is small. The contribution scoring mechanism that combines current evidence (degree of deviation) and prior knowledge (causal strength) makes the quantitative ranking of root causes more scientific and reasonable, and can more accurately identify the main driving factors leading to anomalies.

[0100] S7.5 Sort the primary cause nodes that cause persistent bias according to their contribution scores, and output the sorted list of primary cause nodes as the root cause variable results.

[0101] Furthermore, the primary cause nodes that cause persistent deviations are sorted according to their contribution scores, and the sorted list of primary cause nodes is output as the root cause variable result. All identified primary cause nodes that cause persistent deviations are sorted from high to low according to their contribution scores. An ordered list is generated, in which each entry contains the name of the primary cause node, its contribution score, and a brief description of the deviation. This sorted list of primary cause nodes is the root cause variable result located and quantified after intelligent backtracking analysis of the current energy efficiency anomaly event.

[0102] Specifically, the analysis results are presented in the form of a priority ranking list, transforming complex analysis into clear and actionable insights. Traditional root cause analysis may provide a set of possible causes, but lacks prioritization, which can be confusing for decision-makers. Based on quantitative contribution scores, the ranking list intuitively tells maintenance personnel which causes are primary and which are secondary. The node with the highest contribution score is the most likely and most important root cause to be addressed or checked first. For example, the list may show that the vibration of the drive motor bearing has the highest contribution score, followed by the instantaneous flow fluctuation of the material, guiding maintenance personnel to prioritize checking the bearing condition.

[0103] S8. Integrate and output a visual diagnostic report on the contribution of energy efficiency anomalies and root cause variables.

[0104] S8.1 Collect information on the time, location, and deviation descriptions of energy efficiency anomalies, and correlate and collect contribution scores of root cause variables.

[0105] Furthermore, the occurrence time, location, and deviation description information of energy efficiency anomalies are collected. The occurrence time refers to the start and end time of the anomaly, the location refers to the specific process section or equipment where the anomaly occurs, and the deviation description information includes the duration of the anomaly, the average deviation amplitude, the peak deviation, etc. The contribution scores of the root cause variable are associated and collected. The root cause variable refers to the sorted list of primary cause nodes, and the contribution score corresponding to each node is collected.

[0106] Specifically, metadata is collected from the multi-dimensional and fragmented information generated during the diagnostic process to establish a structured association between events and root causes. A complete diagnosis not only knows where and when something happened, but also identifies the most likely cause. It goes beyond simply collecting information on events (abnormal events); it proactively associates and collects information on causes (root cause variables and their quantitative scores). This collection is not a simple data packaging process, but rather the establishment of event-root cause pairs based on the logic of previous analysis. For example, it precisely associates the event of the BC5 belt conveyor process exceeding the energy consumption standard by 15% from time X to time Y with two root causes, bearing vibration (contribution 0.6) and drum temperature (contribution 0.3), and their scores. This structured association binds the diagnosed phenomena with the analytical conclusions, ensuring the completeness and logical consistency of subsequent reports and avoiding misalignment between causes and events.

[0107] S8.2 Align and integrate the collected event and cause information according to the timeline. Based on the aligned and integrated information, generate a structured report draft and render the structured report draft into a visual diagnostic report.

[0108] Furthermore, the collected event and causal information is aligned and integrated according to the timeline, with the occurrence time of energy efficiency anomalies as the main line, and the root cause variable information is attached as an auxiliary analysis result under this timeline. Based on the aligned and integrated information, a structured report draft is generated. The report draft includes chapters, such as an event overview, a timeline graph, a comparison graph of the abnormal energy efficiency curve and the baseline curve, a bar chart ranking the contribution of root causes, and a summary of possible countermeasures for high-contribution causes. The structured report draft containing charts and text is rendered into a final visual diagnostic report containing visual charts, color coding, and interactive elements (such as a drill-down timeline) through a graphics rendering engine and report template.

[0109] Specifically, the process transforms structured diagnostic data into knowledge products with strong narrative logic and visual appeal. Time-based information alignment and integration ensures that the report is not merely a list of facts, but rather a narrative woven around a timeline, telling the story of when an anomaly occurred, how long it lasted, and what possible causes were identified. This aligns with human cognitive patterns. From structured drafts to visualized reports, the draft defines the logical framework and key data points, while visualization adds flesh and blood and expressiveness. For example, a contribution ranking bar chart clearly identifies primary and secondary causes; a curve comparison chart visually displays the degree and pattern of anomaly deviations. Visualization is not just for aesthetics, but also for efficient communication of complex information. It presents quantitative analysis results in a way that managers and operations personnel can quickly understand and grasp the key points without requiring specialized background knowledge. The delivered product is a diagnostic instruction manual, lowering the application threshold of the technological achievement.

[0110] This embodiment also provides an energy efficiency diagnosis system for dry bulk cargo terminals based on multi-source data, including: a data processing module that collects and integrates multi-source heterogeneous data to form a standardized multi-source data event stream; The analysis module uses causal discovery algorithms to automatically analyze standardized multi-source data event streams and construct a causal knowledge graph of energy efficiency for dry bulk terminals. The module constructs a basic digital twin based on the terminal's physical layout, equipment parameters, and process logic. It then injects the quantified causal logic from the dry bulk terminal's energy efficiency causal knowledge graph into the basic digital twin, forming a terminal process digital twin enhanced with causal knowledge. The simulation module inputs detailed parameters of the operation task and real-time environmental data into the digital twin of the dock process enhanced with causal knowledge for simulation and simulation, and generates personalized dynamic energy efficiency benchmark curves. The inspection module collects actual energy efficiency data in real time during the actual execution of the task, compares the actual energy efficiency data with the personalized dynamic energy efficiency benchmark curve, and detects abnormal energy efficiency events. The quantification module uses the energy efficiency causal knowledge graph of dry bulk cargo terminals as the reasoning map to conduct intelligent backtracking analysis on energy efficiency anomalies, locate and quantify root cause variables, and integrate and output a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables.

[0111] This embodiment also provides a computer device applicable to the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data as proposed in the above embodiment.

[0112] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0113] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0114] In summary, this invention integrates heterogeneous data from multiple sources, including equipment, production, and the environment, into a standardized event stream. It then utilizes a causal discovery algorithm to automatically construct a dynamic knowledge graph that quantifies the causal relationships between equipment, environment, operation, and energy efficiency. The causal logic from this graph is injected into a physical model-based digital twin, forming a causal knowledge-enhanced twin that integrates data-driven real-world patterns. Real-time parameters and environmental data are injected into each specific task, and high-fidelity, context-adaptive, personalized dynamic energy efficiency benchmark curves are generated through simulation and deduction by the enhanced twin. During operation, real-time data is compared with the personalized benchmark to sensitively detect actual energy efficiency anomalies. Using the causal knowledge graph as a reasoning map, anomalies are intelligently traced back to pinpoint and quantify the contribution of root cause variables, and a visual diagnostic report is output. This achieves a leap from static, universal assessment to dynamic, personalized diagnosis, and from phenomenon monitoring to causal tracing, improving the sensitivity, accuracy, and effectiveness of energy efficiency anomaly detection under complex operating conditions.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data, characterized in that: This includes collecting and fusing heterogeneous data from multiple sources to form a standardized multi-source data event stream; By using causal discovery algorithms to automatically analyze standardized multi-source data event streams, a causal knowledge graph of energy efficiency at dry bulk terminals is constructed. Based on the physical layout of the terminal, equipment parameters and process logic, a basic digital twin is constructed. The quantitative causal logic in the energy efficiency causal knowledge graph of the dry bulk terminal is injected into the basic digital twin to form a terminal process digital twin enhanced with causal knowledge. Detailed parameters of the work tasks and real-time environmental data are input into a digital twin of the dock process enhanced with causal knowledge for simulation and deduction, generating personalized dynamic energy efficiency benchmark curves. During the actual execution of the task, real-time energy efficiency data is collected and compared with the personalized dynamic energy efficiency benchmark curve to detect abnormal energy efficiency events. Using the energy efficiency causal knowledge graph of dry bulk cargo terminals as a reasoning map, intelligent backtracking analysis is performed on energy efficiency anomalies to locate and quantify root cause variables, and a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables is integrated and output.

2. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 1, characterized in that: Collecting and fusing heterogeneous data from multiple sources to form a standardized multi-source data event stream includes the following steps: Multi-source heterogeneous data is obtained from the equipment monitoring unit, production management unit, video monitoring unit, environmental monitoring unit, and energy metering unit of the dry bulk terminal, and the data is cleaned. After data cleaning, multi-source heterogeneous data is mapped and bound to entities in the digital twin model. The mapped and bound multi-source heterogeneous data is then associated and fused based on spatiotemporal labels and business semantics to form a standardized multi-source data event stream.

3. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 2, characterized in that: The causal discovery algorithm is used to automatically analyze standardized multi-source data event streams and construct a causal knowledge graph of energy efficiency for dry bulk cargo terminals, including the following steps: From standardized multi-source data event streams, fields describing equipment status, work instructions, environmental conditions, and energy consumption are extracted and reorganized into a sequence of causal variables; The stripped and recombined causal variable sequences are fed into a pre-screening layer with transit entropy as the correlation metric, and the information flow intensity between each pair of causal variable sequences is calculated. The pre-screened causal variable sequences enter the causal direction discrimination stage based on convergent cross mapping. The embedding theorem is used to verify the existence of a unidirectional driving relationship between the causal variable sequences in the reconstructed phase space. The verified unidirectional driving relationship is marked as a directed causal hypothesis. For each directed edge in the causal hypothesis network, a nonparametric Granger causality test combined with the Bayesian information criterion is used to quantify the predictive contribution of the causal variable sequence to the outcome causal variable sequence under the condition of controlling for confounding causal variable sequences, and this contribution is used as the causal strength coefficient. By combining the topology of the causal hypothesis network with the causal strength coefficients of each edge, a causal knowledge graph of energy efficiency for dry bulk cargo terminals is constructed.

4. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 3, characterized in that: Based on the terminal's physical layout, equipment parameters, and process logic, a basic digital twin is constructed. The quantified causal logic from the dry bulk terminal's energy efficiency causal knowledge graph is then injected into this basic digital twin, forming a terminal process digital twin enhanced with causal knowledge. This process includes the following steps: Based on the physical layout diagram of the wharf, the geometric and dynamic parameter library of the equipment, and the process logic rule library of material transportation, a basic digital twin is constructed. The causal knowledge graph of energy efficiency in dry bulk terminals is analyzed, and the directed edge relationships in the graph are mapped to the dynamic influence links between corresponding entities in the basic digital twin. The quantitative causal strength coefficient attached to each directed edge in the causal knowledge graph of energy efficiency in dry bulk terminals is extracted. The quantified causal strength coefficient is transformed into a real-time correction parameter for the corresponding dynamic influence link in the basic digital twin; Using the real-time correction parameters after transformation, the simulation operation parameters and interaction rules of the corresponding entities in the basic digital twin are dynamically adjusted to form a digital twin of the dock process enhanced with causal knowledge.

5. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 4, characterized in that: Detailed parameters of the operational tasks and real-time environmental data are input into a digital twin of the terminal process enhanced with causal knowledge for simulation and deduction, generating a personalized dynamic energy efficiency benchmark curve, including the following steps: Receive detailed parameters of the work tasks, including cargo type, flow rate, equipment assignment and process path, as well as real-time environmental data obtained from the environmental monitoring terminal; Configure the detailed parameters of the task and real-time environmental data as the initial conditions and boundary constraints for the simulation of the digital twin of the dock process enhanced with causal knowledge. Under the initial conditions and boundary constraints of the simulation, the digital twin of the dock process enhanced with causal knowledge is driven to run at the virtual clock synchronization rate to deduce the complete execution process of the operation task in the virtual environment. During the simulation, the instantaneous power data of each energy-consuming virtual entity in the digital twin of the dock process, which has been enhanced with causal knowledge, is recorded at each virtual moment. The instantaneous power data stream recorded in the digital twin simulation, arranged according to virtual timestamps, is resampled and windowed according to the time resolution of the actual work plan to generate a simulation energy efficiency curve describing the change of theoretical energy consumption of the work task over time. Based on the start time of the work task plan, the simulated energy efficiency curve is aligned and mapped on the time axis to generate a personalized dynamic energy efficiency benchmark curve.

6. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 5, characterized in that: During the actual execution of the task, real-time energy efficiency data is collected, including the following steps: When a task is started, the real-time data acquisition channel of all equipment on the corresponding process path is activated to capture the energy meter pulse signal and the running status word of the equipment. The pulse signal of the energy meter is translated into a cumulative energy reading through the pulse-energy conversion relationship and encapsulated into a time-stamped data frame. The encapsulated time-stamped data frames are sorted and cached according to the process flow order. The sorted and cached time-stamped data frames are output as actual energy efficiency data at fixed time intervals.

7. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 6, characterized in that: The actual energy efficiency data is compared with the personalized dynamic energy efficiency benchmark curve to detect energy efficiency anomalies, including the following steps: The system matches the baseline energy efficiency value at the same time point from the personalized dynamic energy efficiency baseline curve, extracts the current energy consumption reading from the actual energy efficiency data, and forms a data pair with the matched baseline energy efficiency value at the same time point. The system judges the current energy consumption reading and the baseline energy efficiency value, and records the duration of the exceedance. When the duration of the excess exceeds the preset tolerance time window, a primary anomaly marker is triggered. The primary anomaly markers that are triggered are aggregated. If the density of primary anomaly markers exceeds the threshold within a specific process segment, an energy efficiency anomaly event is confirmed.

8. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 7, characterized in that: Using the causal knowledge graph of energy efficiency at dry bulk terminals as a reasoning map, intelligent backtracking analysis is performed on energy efficiency anomalies to locate and quantify root cause variables, including the following steps: Based on the energy efficiency causal knowledge graph of dry bulk cargo terminals, the terminal energy consumption nodes associated with energy efficiency anomalies are located, the primary cause nodes that directly affect the terminal energy consumption nodes are retrieved, and the actual observation sequence of the primary cause nodes during the energy efficiency anomaly event period is obtained. Extract the baseline expected value sequence of primary cause nodes on the personalized dynamic energy efficiency baseline curve within the same time period; By comparing the actual observation sequence of the primary cause node with its baseline expected value sequence, the primary cause node that causes the persistent deviation can be identified. Based on the causal influence intensity indicated in the causal knowledge graph of dry bulk terminal energy efficiency, the contribution score of each primary cause node of persistent deviation is evaluated. The primary cause nodes that cause persistent bias are sorted according to their contribution scores, and the sorted list of primary cause nodes is output as the root cause variable result.

9. The method for energy efficiency diagnosis of dry bulk cargo terminals based on multi-source data as described in claim 8, characterized in that: Integrate and output a visual diagnostic report on the contribution of energy efficiency anomalies and root cause variables, including the following steps: Collect information on the timing, location, and deviation descriptions of energy efficiency anomalies, and correlate and aggregate the contribution scores of root cause variables; The collected event and cause information is aligned and integrated according to the timeline. Based on the aligned and integrated information, a structured report draft is generated, and the structured report draft is rendered into a visual diagnostic report.

10. A dry bulk cargo terminal energy efficiency diagnosis system based on multi-source data, based on the energy efficiency diagnosis method for dry bulk cargo terminals based on multi-source data according to any one of claims 1 to 9, characterized in that: This includes a data processing module that collects and merges heterogeneous data from multiple sources to form a standardized multi-source data event stream; The analysis module uses causal discovery algorithms to automatically analyze standardized multi-source data event streams and construct a causal knowledge graph of energy efficiency for dry bulk terminals. The module constructs a basic digital twin based on the terminal's physical layout, equipment parameters, and process logic. It then injects the quantified causal logic from the dry bulk terminal's energy efficiency causal knowledge graph into the basic digital twin, forming a terminal process digital twin enhanced with causal knowledge. The simulation module inputs detailed parameters of the operation task and real-time environmental data into the digital twin of the dock process enhanced with causal knowledge for simulation and simulation, and generates personalized dynamic energy efficiency benchmark curves. The inspection module collects actual energy efficiency data in real time during the actual execution of the task, compares the actual energy efficiency data with the personalized dynamic energy efficiency benchmark curve, and detects abnormal energy efficiency events. The quantification module uses the energy efficiency causal knowledge graph of dry bulk cargo terminals as the reasoning map to conduct intelligent backtracking analysis on energy efficiency anomalies, locate and quantify root cause variables, and integrate and output a visualized diagnostic report on the contribution of energy efficiency anomalies and root cause variables.

Images