Knowledge graph-based port ship loader energy safety accident tracing method

Abstract

The application discloses a knowledge graph-based port ship loader energy safety accident tracing method and relates to the technical field of port equipment safety monitoring. The total time difference folding chain pieces in each time difference folding chain cluster are traversed and counted, the total number of bus forced top event nodes, the total accumulated continuous duration of brake extension event nodes, the total accumulated continuous duration of charge blocking event nodes, the total number of charge permission flip event nodes, the total number of repeated point operation event nodes, the total duration of adjacent event node net intervals and the number of time difference folding chain pieces are extracted, and a first evaluation value is further calculated. Through this processing, the chain piece structure that can only reflect local abnormalities is converted into a quantified result that can be compared uniformly, so that the abnormal chains formed in different operation cycles have the same caliber evaluation basis, and the problem that abnormal records are difficult to compare horizontally and difficult to distinguish between primary and secondary in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of port equipment safety monitoring technology, specifically a method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs. Background Technology

[0002] In existing technologies, the handling of the energy safety status of ship loaders and unloaders mainly revolves around the DC bus voltage level, battery charging permit status, supercapacitor acceptance status, braking unit conduction status, and protection action results. The focus is on determining over-limit conditions at a specific moment, assessing the status of a particular component, or recording a single protection action. While this approach can reflect local conditions, it struggles to capture the continuous evolution of anomalies across multiple work cycles, control units, and event nodes.

[0003] Especially during the descent of the ship loader / unloader, regenerative energy flows continuously between the DC bus, braking unit, supercapacitor, battery, and energy management system. Anomalies are often not caused by a single parameter instantaneously exceeding its limit, but rather by the accumulation of multiple small anomalies, such as bus overshoot, continuous braking, recharge obstruction, permitted reversal, strategy switching, and repeated jogging, which gradually develop into protective actions. Existing technologies lack a unified way of expressing the relationships between these anomalies, lack structured extraction methods for the preceding chains of anomalies, and lack the ability to continuously remember and trace back similar historical structures. This leads to accident analysis easily remaining at the surface results level, making it difficult to obtain the true causal structure that leads to the closure of the accident.

[0004] The aforementioned shortcomings are directly related to the inherent characteristics of the ship loader / unloader's energy consumption process, which involves dispersed multi-source data, inconsistent sampling cycles, frequent control state switching, and dynamic changes in energy destination. During the descent operation, changes in DC bus voltage, continuous engagement of the braking unit, changes in supercapacitor acceptance, switching of battery charging permission, adjustments to the energy management system strategy, and jogging operations are typically recorded separately in different control units. These records exhibit differences in fields, time stamps, and semantics. Without a unified timeline and structure, the same chain of anomalies can be broken down into multiple isolated records, making it impossible to continuously identify preceding anomalies within the same chain. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a knowledge graph-based method for tracing the source of energy safety accidents in port ship loading and unloading machinery, thus solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs, comprising the following steps: S1. Collect the operating data from the existing control system of the ship loading and unloading machine, and perform cyclic slicing, unified time scale alignment and synchronization processing on the operating data to generate the same frequency sequence matrix; S2. Extract event nodes based on the same frequency sequence matrix, and establish directed connection edges according to the time sequence relationship, control succession relationship and energy path obstruction relationship between event nodes. Cluster and fold the event nodes and directed connection edges to generate time difference folded chain clusters. S3. Evaluate the obtained time difference folded chain clusters, obtain the first evaluation value Q1, and sort all time difference folded chain clusters to obtain the main preceding chain. S4. Perform counter-evidence convergence processing on the main preorder chain to form a counter-evidence convergence kernel. Then, construct a second evaluation based on the counter-evidence convergence kernel, obtain the second evaluation value Q2, and obtain the causal core structure result package of the dominant accident closure. S5. Analyze the causal core structure result package, and write the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster and the counter-evidence convergence kernel into the historical memory map, and provide feedback based on the historical memory map.

[0007] Preferably, S1 includes S11 and S12; S11. Read the operating data from the existing control system of the ship loading and unloading machine. The operating data includes DC bus voltage data output by the frequency converter, conduction status data of the braking unit, acceptance status data of the supercapacitor, battery charging permission status data, energy management system strategy bit data, descent command and jog command data output by PLC or HMI, and protection action data output by the driver or PLC. The system performs field mapping, timestamp extraction, status coding standardization, duplicate record merging, and invalid record removal on the running data to generate a raw time-series dataset in a unified format. S12. Based on the original time series dataset, take the moment when the descent command first takes effect as the start time of the work cycle and take the moment when the DC bus voltage enters the steady state range and is continuously maintained for at least three sampling cycles as the end time of the work cycle, and perform loop slicing on the original time series dataset. The steady-state range is the DC bus voltage range constructed with the reference steady-state voltage as the center and a fixed voltage deviation range. The reference steady-state voltage is the average value of the DC bus voltage within the continuous sampling time window before the descent command takes effect. Within each cyclic slice job, the running data is aligned according to a unified time base to generate a synchronous data sequence. Feature data is calculated based on the synchronous data sequence. The synchronous data sequence and feature data are arranged in ascending order of sampling time, using the sampling time on the unified time axis as the matrix row index. The synchronous data sequence field and feature data field are used as the matrix column index. The data value, state value or feature value corresponding to the sampling time is written into the corresponding matrix cell to generate the same frequency order matrix. The characteristic data includes busbar top-pressing characteristics, braking continuity characteristics, permissible reversal characteristics, jogging discrete characteristics, and recharge blockage characteristics; The bus approach characteristic is obtained by subtracting the current bus voltage from the protection upper limit voltage of the DC bus voltage data, and is used to characterize the degree to which the bus approaches the upper limit. The braking duration characteristic is obtained by accumulating the conduction time of the braking unit conduction state data, and is used to characterize the degree of energy consumption duration. The permit flipping feature is obtained from the number of times the battery charging permit state data is switched within a single cycle, and is used to characterize the degree of change in the energy storage acceptance state. The discrete characteristics of jogging are calculated from the time interval between adjacent jogging command data and are used to characterize the degree of discreteness of the operation rhythm; The recharge blockage feature is obtained by the overlapping judgment results of the supercapacitor acceptance status data, battery charging permission status data and braking unit conduction status data on the same time axis within the same working cycle. When the supercapacitor is not accepted, the battery is not in the charging permission state and the braking unit is in the conduction state, the corresponding time period is determined as the recharge blockage section, which is used to characterize the state of obstructed regenerative energy destination.

[0008] Preferably, S2 includes S21 and S22; S21. Read the same frequency sequence matrix and scan the synchronous data sequence and feature data line by line on the same time axis; When the descent command data indicates that the descent action is in the execution state, the corresponding time period is extracted and a descent event node is generated. When the bus overshoot feature corresponding to the DC bus voltage data is in an effective state, extract the corresponding time period and generate the bus overshoot event node. When the braking continuity feature corresponding to the braking unit conduction state data is in an effective state, the corresponding time period is extracted and a braking extension event node is generated. When the permit flip feature corresponding to the battery charging permit status data is in a valid state, extract the corresponding time period and generate a charging permit flip event node. When the energy management system policy bit data changes state, the corresponding time period is extracted and an energy management system policy change event node is generated. When the jog discrete feature corresponding to the jog command data is in a valid state, extract the corresponding time period and generate a repeating jog event node; When the recharge blockage feature formed by the supercapacitor acceptance status data, battery charging permission status data, and braking unit conduction status data is in a valid state, the corresponding time period is extracted and a recharge obstruction event node is generated. For each event node, record the node type, node start time, node end time, node duration, job cycle number to which the node belongs, and the data field from which the node originates, and generate an event node set.

[0009] Preferably, in step S22, based on the set of event nodes, directed connections are established between event nodes within the same job cycle; When the start time of the subsequent event node is after the end time of the previous event node, and there are no other inserted event nodes of the same priority between them, a time sequence connection edge representing the time sequence relationship is established. The current event node originates from the energy management system strategy bit data or battery charging permission status data, and the subsequent event node originates from the braking unit conduction status data, DC bus voltage data, or recharge blockage characteristics. Furthermore, when the subsequent event node begins within a fixed continuous sampling period window after the end of the previous event node, a control receiving connection edge representing the control state transmission process is established. When the previous event node or the next event node is a recharge obstruction event node, and it overlaps or occurs adjacent to the bus top-pressure event node and the braking extension event node in time, an energy path obstruction connection edge representing the process of restricted regenerative energy destination is established; After establishing directed edges, the node sequence formed by connecting adjacent directed edges within the same job cycle is folded into a single chain. Then, multiple chains are clustered according to the node type arrangement, edge type combination, chain duration, and time difference distribution between nodes to generate time difference folded chain clusters. The number of chains, start and end times, number of nodes, edge type combination, and job cycle number corresponding to each chain cluster are recorded.

[0010] Preferably, S3 includes S31 and S32; S31. Read each time difference folded chain cluster and perform traversal statistics on all time difference folded chain segments contained within the time difference folded chain cluster to obtain the total number of bus top-pressing event nodes zM, the cumulative total duration of braking extension event nodes zT, the cumulative total duration of recharge obstruction event nodes sT, the total number of charging permission flip event nodes fM, the total number of repeated jogging event nodes cM, the total duration of net interval between adjacent event nodes xT, and the number of time difference folded chain segments zN. Calculate and obtain the first evaluation value Q1 of the time difference folded chain cluster. The first evaluation value Q1 is obtained by the following formula: Q1=(zM+fM+cM)×(zT+sT+zN) / (xT+1); Among them, the total number of bus overrun event nodes zM represents the total number of all bus overrun event nodes in the same time-difference folded chain cluster, which is derived from the bus overrun event nodes extracted based on DC bus voltage data in S2; The cumulative total duration of the braking extension event nodes zT represents the sum of the durations of all braking extension event nodes within the same time-difference folded chain cluster, which is derived from the braking extension event nodes extracted from the braking unit conduction state data in S2. The total duration of the recharge obstruction event nodes sT represents the sum of the durations of all recharge obstruction event nodes within the same time-difference folded chain cluster, which is derived from the recharge obstruction event nodes extracted in S2 based on supercapacitor acceptance status data, battery charging permission status data, and braking unit conduction status data. The total number of charging license flip event nodes fM represents the total number of all charging license flip event nodes within the same time-difference folded chain cluster, which is derived from the charging license flip event nodes extracted from battery charging license status data in S2. The total number of repeated jog event nodes cM represents the total number of all repeated jog event nodes within the same time-difference folded chain cluster, which is derived from the repeated jog event nodes extracted from the jog command data in S2; The total net interval between adjacent event nodes xT represents the cumulative time interval between adjacent event nodes within the same time difference folded chain cluster without overlap, which is derived from the start and end times of each event node in S2; The number of time difference folded chain segments zN represents the total number of time difference folded chain segments contained in the same time difference folded chain segment cluster, which is derived from the clustering folding results of S2; After obtaining the first evaluation value Q1, the central chain piece inside each time difference folded chain cluster is obtained: Specifically, the time difference folded chain piece selected from the time difference folded chain cluster is selected from the cluster whose node type arrangement order is consistent with the main sequence of the time difference folded chain cluster, and whose duration of each event node and the time difference between nodes deviate the least from the average value of the time difference folded chain cluster. This is taken as the central chain piece of the time difference folded chain cluster.

[0011] Preferably, based on the first evaluation value Q1 and the central chain of the time difference folded chain cluster, all time difference folded chain clusters are sorted from largest to smallest according to the first evaluation value Q1; When the first evaluation value Q1 of two time difference folded chain clusters is the same, the time difference folded chain cluster with more connecting edges is sorted first. When the number of control receiving connection edges is the same, the folded chain clusters with the smaller time difference between the end time of the central chain segment's end node and the corresponding time of the protection action data are sorted first. The time difference folded chain cluster that ranks first in the sorting is identified as the target time difference folded chain cluster, and the central chain corresponding to the target time difference folded chain cluster is identified as the main preorder chain.

[0012] Preferably, S4 includes S41 and S42; S41. Read the main prequence chain, establish the main prequence chain node table according to the event node order in the main prequence chain, and determine the protection action time corresponding to the protection action data at the end of the main prequence chain as the closing endpoint. The single-node removal calculation is performed sequentially on each event node in the main preorder chain. The single-node removal calculation process is as follows: each event node in the main preorder chain is sequentially treated as the current node to be removed and deleted, and the directed connection edges directly connected to the current node to be removed are deleted simultaneously; on the pruning chain structure formed by the remaining event nodes and the remaining directed connection edges, the directed arrival path is reconstructed according to the original time order; the protection action time corresponding to the protection action data is used as the closure determination endpoint, and a directed search is performed starting from the first event node; when the pruning chain structure still maintains the closure relationship up to the protection action time, the current node to be removed is determined as an accompanying node; when the pruning chain structure cannot maintain the closure relationship up to the protection action time, the current node to be removed is determined as a necessary node. When an event node is removed, if the first event node can still reach the protection action event node through the remaining time sequence connection edge, the control receiving connection edge, and the energy path blocked connection edge, the event node is identified as an accompanying node. When an event node is removed, and the first event node cannot reach the protection action event node, the event node is determined to be a necessary node. After completing the single-node removal calculation, a two-node combination removal calculation is performed on the event nodes in the main preorder chain. The two-node combination removal calculation includes: combining the event nodes in pairs according to the time sequence to form a node pair to be calculated; for each node pair to be calculated, simultaneously deleting the two event nodes in the node pair and the directed connection edges directly connected to the two event nodes respectively, reconstructing the directed arrival path on the deletion chain structure formed by the remaining event nodes and the remaining directed connection edges, and using the protection action time corresponding to the protection action data as the closure determination endpoint; when the main preorder chain remains closed when the two event nodes are removed separately, but the main preorder chain loses closure after the two event nodes are removed simultaneously, the two event nodes are determined to be a coupled node pair. If the two event nodes are removed individually, the protection action event node remains reachable; however, if the two event nodes are removed simultaneously, the protection action event node becomes unreachable. In this case, the two event nodes are identified as a coupled node pair. Subsequently, all blocked energy path edges connected to the recharge blocked event node in the main preorder chain are extracted to form a blocked edge set; Then, based on the necessary nodes, the pairs of coupled nodes, and the set of blocking edges, the proof-of-contrast convergence kernel is determined.

[0013] The specific process for determining the counter-evidence convergence kernel is as follows: Based on the main preorder chain, firstly, the set of necessary nodes is determined according to the calculation results of single-node removal; the set of coupled node pairs is determined according to the calculation results of double-node combination removal; and the set of blocked edges is determined according to the correspondence between the recharge obstruction event nodes and the energy path obstruction connection edges in the main preorder chain. Then, the set of necessary nodes, the set of coupled node pairs, and the set of blocked edges are mapped together onto the main preorder chain topology to form the counter-evidence convergence candidate set. Subsequently, a closure retention judgment is performed on the nodes and connection edges in the counter-evidence convergence candidate set, retaining only the nodes and connection edges that would cause the main preorder chain to lose its closure relationship up to the protection action time after deletion. Finally, the retained nodes and connection edges are minimized according to the time sequence and connection relationship to obtain the counter-evidence convergence kernel. Among them, the convergence kernel of the proof by contradiction includes at least the set of necessary nodes, the set of coupled node pairs, the set of blocking edges, the earliest cause time, and the latest cut-off time; The earliest causal moment is the starting moment of the earliest event node in the necessary path or coupled node pair that is the first in the time sequence of the proof by contradiction convergence kernel. The latest cutoff time is the end time of the last necessary node or the later event node in the coupled node pair that is located in the time sequence of the counter-evidence convergence core before the protection action event node and whose removal would make the protection action event node unreachable.

[0014] Preferably, in step S42, feature extraction is performed on the counter-evidence convergence kernel, extracting the chain closure span Hs, the number of blocked edges Ns, the cumulative total duration Ts of the recharge obstruction event nodes, and the minimum number of broken chain nodes Ks; Based on the chain closure span Hs, the number of blocked edges Ns, the total duration of the recharge obstruction event nodes Ts, and the minimum number of broken chain nodes Ks, the second evaluation value Q2 is calculated and obtained. The second evaluation value Q2 is obtained by the following formula: Q2=(Hs×Ns×Ts) / (Ks+1) Among them, the chain closure span Hs represents the net duration between the earliest cause time and the corresponding time of the protection action event node; The number of blocked edges Ns represents the total number of blocked connection edges in the convergence kernel of the proof by contradiction; The cumulative total duration Ts of the recharge blocked event nodes represents the cumulative value of the duration of all recharge blocked event nodes in the proof-of-confusion kernel; The minimum number of broken nodes Ks represents the minimum number of event nodes that need to be removed to change the main preceding chain from a reachable state to an unreachable state. After obtaining the second evaluation value Q2, the second evaluation value Q2 is fitted with the proof of contradiction convergence kernel to form the causal core structure result package.

[0015] Preferably, S5 includes S51 and S52; S51, Read the time difference folded chain cluster, the first evaluation value Q1, and the causal core structure result package; Perform structural analysis on the proof-of-contrast kernel to extract the set of necessary nodes, the set of coupled node pairs, the set of blocked edges, the earliest cause time, the latest cut-off time, the job cycle number to which it belongs, and the corresponding event node order and directed connection edge order; Using the current job cycle number as the main index, a current job cycle memory unit is established in the historical memory graph, and the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster, the main preorder chain, and the proof of contradiction convergence kernel are written into the current job cycle memory unit respectively. Among them, the first evaluation value Q1 is written into the chain evaluation field of the current job cycle memory unit, the second evaluation value Q2 is written into the causal evaluation field of the current job cycle memory unit, the time difference folding chain cluster is written into the chain structure field of the current job cycle memory unit, the main preorder chain is written into the main chain field of the current job cycle memory unit, and the proof by contradiction convergence kernel is written into the causal core field of the current job cycle memory unit. After the writing is completed, the current job cyclic memory unit is structurally matched with the existing job cyclic memory units in the historical memory graph to obtain the historical recurrence statistics.

[0016] The structure matching process is as follows: Extract the node type order, connection edge type order, required node set, coupled node pair set, and blocking edge set from the current job's cyclic memory unit and each historical job's cyclic memory unit; perform a bit-by-bit comparison on the node type order to obtain the number of matching node order sequences; perform a bit-by-bit comparison on the connection edge type order to obtain the number of matching connection edge order sequences; perform set comparison on the required node set, coupled node pair set, and blocking edge set to obtain the number of repeated required nodes, repeated coupled node pairs, and repeated blocking edges, respectively; then calculate the structure matching value based on the node order matching value, connection edge order matching value, repeated required nodes, repeated coupled node pairs, and repeated blocking edges; and obtain the number of identical time-difference folded chain clusters, repeated required nodes, repeated coupled node pairs, and repeated blocking edges based on the structure matching value and the number of repeated results. Preferably, in step S52, the feedback value Q3 is calculated based on historical recurrence statistics and in combination with the first evaluation value Q1 and the second evaluation value Q2. The calculation formula is as follows: ; In the formula, yM represents the number of time difference folded chain clusters of the same type, specifically the number of time difference folded chain clusters in the historical memory graph that are consistent with the current time difference folded chain cluster in terms of node type arrangement order and connection edge type combination; yN represents the number of repeated mandatory nodes, specifically the number of repeated event nodes in the historical memory graph that are consistent with the set of mandatory nodes in the current proof-of-confusion kernel; yP represents the number of repeated coupled node pairs, specifically the number of repeated node pairs in the historical memory graph that are consistent with the set of coupled node pairs in the current proof-of-confusion kernel; yL represents the number of repeated blocking edges, specifically the number of repeated directed connection edges in the historical memory graph that are consistent with the set of blocking edges in the current proof-of-confusion kernel; RL represents the difference between the current job cycle number and the most recent matched job cycle number, which comes from the job cycle number retrieval result in the historical memory graph; After the feedback value calculation is completed, a feedback result package is generated based on the feedback value Q3, the number of identical time difference folded chain clusters yM, the number of repeated mandatory nodes yN, the number of repeated coupled node pairs yP, and the number of repeated blocking edges yL. The feedback result package is then written back to the current job cycle memory unit in the historical memory graph and output to the subsequent job cycle identification process. This is used to prioritize the retrieval of the chain structure, mandatory nodes, coupled node pairs, and blocking edges corresponding to the feedback result package after a new time difference folded chain cluster is generated.

[0017] This invention provides a knowledge graph-based method for tracing the source of energy safety accidents in port ship loading and unloading machinery, which has the following beneficial effects: (1) The time difference folded chain clusters are uniformly evaluated using the first evaluation value Q1, and then all time difference folded chain clusters are sorted to extract the main preceding chain. After this processing, the results of the chain fragments that originally appeared in a large number of parallel events are compressed into a main chain with the highest priority. Subsequent analysis no longer revolves around all anomalies at the same time, but focuses on the preceding chain that is most closely related to the closure of the incident, which helps to reduce the interference caused by an overly broad analysis scope. The main preceding chain is subjected to a proof-of-contradiction process. Through single-node removal calculation and double-node combination removal calculation, the necessary nodes, coupled node pairs and blocking edge sets are extracted and a proof-of-contradiction kernel is formed.

[0018] (2) The start time of the first effective moment of the descent command is taken as the start time of the operation cycle, and the end time of the operation cycle is taken as the moment when the DC bus voltage enters the steady-state range and remains continuously for at least three sampling periods. Cyclic slicing is performed on the original time series dataset, and alignment processing is performed according to a unified time reference within each operation cycle. Through this processing, a complete regenerative energy processing process under the descent condition of the ship loader / unloader is divided into independently analyzable operation cycles. The originally discrete data across devices and records is compressed into the same cycle, solving the problem in the prior art that only scattered records can be seen and it is difficult to restore the abnormal sequence relationship within the same operation cycle.

[0019] (3) The node sequence formed by connecting adjacent directed edges within the same work cycle is folded into a chain, and multiple chains are further clustered according to the node type arrangement, the combination of connecting edge types, the duration of the chain, and the time difference distribution between nodes to generate a time difference folded chain cluster. Through this process, the originally numerous and dispersed event sequences are compressed into chain clusters with common structural features. This not only preserves the temporal characteristics of the energy consumption anomalies of the ship loading and unloading machine, but also forms structural units that are easy to uniformly evaluate and screen later. This solves the problem that existing technologies have difficulty merging similar anomaly structures and identifying the main anomaly chains when faced with a large number of discrete anomaly records.

[0020] (4) By traversing and statistically analyzing all time-difference folded chain segments within each time-difference folded chain segment cluster, the total number of bus top-pressing event nodes, the cumulative total duration of braking extension event nodes, the cumulative total duration of recharge obstruction event nodes, the total number of charging permit reversal event nodes, the total number of repeated jogging event nodes, the total duration of net interval between adjacent event nodes, and the number of time-difference folded chain segments are extracted, and the first evaluation value is further calculated. Through this processing, the chain segment structure, which originally could only reflect local anomalies, is transformed into a quantitative result that can be uniformly compared, so that the abnormal chains formed in different work cycles have the same evaluation basis, solving the problem that it is difficult to compare abnormal records horizontally and distinguish between primary and secondary anomalies in the existing technology. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the steps of the knowledge graph-based method for tracing the source of energy safety accidents in port ship loading and unloading machinery according to the present invention; Figure 2 This is a schematic diagram of the necessary node determination process of the present invention; Figure 3 This is a schematic diagram illustrating the decision-making process for controlling the connection edge of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] Example 1 This invention provides a knowledge graph-based method for tracing the source of energy safety accidents in port ship loading and unloading machinery. Please refer to [link / reference]. Figures 1 to 3 This includes the following steps: S1. Collect the operating data from the existing control system of the ship loading and unloading machine, and perform cyclic slicing, unified time scale alignment and synchronization processing on the operating data to generate the same frequency sequence matrix; S2. Extract event nodes based on the same frequency sequence matrix, and establish directed connection edges according to the time sequence relationship, control succession relationship and energy path obstruction relationship between event nodes. Cluster and fold the event nodes and directed connection edges to generate time difference folded chain clusters. S3. Evaluate the obtained time difference folded chain clusters, obtain the first evaluation value Q1, and sort all time difference folded chain clusters to obtain the main preceding chain. S4. Perform counter-evidence convergence processing on the main preorder chain to form a counter-evidence convergence kernel. Then, construct a second evaluation based on the counter-evidence convergence kernel, obtain the second evaluation value Q2, and obtain the causal core structure result package of the dominant accident closure. S5. Analyze the causal core structure result package, and write the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster and the counter-evidence convergence kernel into the historical memory map, and provide feedback based on the historical memory map.

[0024] In this embodiment, the DC bus voltage, braking unit conduction status, supercapacitor acceptance status, battery charging permission status, energy management system strategy bits, descent commands and jogging commands, and protection action records are first cyclically sliced, unified in time scale alignment, and synchronized to generate a matrix with the same frequency sequence. After this processing, the data that was originally scattered in different control units is placed in the same work cycle and the same time axis, which makes it easier to continuously represent the related abnormal processes and helps to solve the problem of broken abnormal chains caused by scattered multi-source data and inconsistent time scales. The ship loader / unloader itself has a continuous energy transfer relationship between the DC bus, braking unit, supercapacitor, battery, and energy management system during the descent operation. This embodiment completes the unified organization on this basis.

[0025] This embodiment extracts bus overshoot event nodes, braking extension event nodes, charging permission reversal event nodes, energy management system strategy switching event nodes, repetitive jogging event nodes, and recharge obstruction event nodes based on the same frequency sequence matrix. It further establishes time-series connection edges, control takeover connection edges, and energy path obstruction connection edges, forming a time-difference folded chain cluster. In this way, the analysis object is no longer just "overvoltage at a certain moment" or "conduction at a certain moment," but a chain structure of "which events occur first, which events occur later, and which control changes bring out subsequent anomalies," which is beneficial for clearly representing the evolution process of preceding anomalies.

[0026] This embodiment sets up two levels of quantification results: a first evaluation value Q1 and a second evaluation value Q2. The first evaluation value Q1 corresponds to the degree of anomalous aggregation and continuous unfolding of the time difference folded chain clusters, while the second evaluation value Q2 corresponds to the degree of support for the accident closure by the counter-evidence convergence check. This processing separates the external manifestations of the preceding anomalous chain from the internal structure of the causal core, facilitating the distinction between "the chain appears complex" and "the chain truly supports closure," and promoting the tiered use of subsequent feedback results.

[0027] In this embodiment, the first evaluation value Q1, the second evaluation value Q2, the time difference folded chain clusters, the main preorder chain, and the proof of contradiction convergence kernel are written into the historical memory graph. Structural matching is then performed on the historical memory graph to obtain the number of identical time difference folded chain clusters, the number of repeated necessary nodes, the number of repeated coupled node pairs, and the number of repeated blocking edges. After this processing, the analysis results of the current job cycle are no longer one-off conclusions, but become historical samples for subsequent retrieval and comparison, which helps solve the problems of difficulty in reusing similar anomalies and difficulty in identifying historical recurrences.

[0028] Example 2 Please refer to Figure 1 Specifically: S1 includes S11 and S12; S11. Read the operating data from the existing control system of the ship loading and unloading machine. The operating data includes DC bus voltage data output by the frequency converter, conduction status data of the braking unit, acceptance status data of the supercapacitor, battery charging permission status data, energy management system strategy bit data, descent command and jog command data output by PLC or HMI, and protection action data output by the driver or PLC. The system performs field mapping, timestamp extraction, status coding standardization, duplicate record merging, and invalid record removal on the running data to generate a raw time-series dataset in a unified format. S12. Based on the original time series dataset, take the moment when the descent command first takes effect as the start time of the work cycle and take the moment when the DC bus voltage enters the steady state range and is continuously maintained for at least three sampling cycles as the end time of the work cycle, and perform loop slicing on the original time series dataset. The steady-state range is the DC bus voltage range constructed with the reference steady-state voltage as the center and a fixed voltage deviation range. The reference steady-state voltage is the average value of the DC bus voltage within the continuous sampling time window before the descent command takes effect. Within each cyclic slice job, the running data is aligned according to a unified time base to generate a synchronous data sequence. Feature data is calculated based on the synchronous data sequence. The synchronous data sequence and feature data are arranged in ascending order of sampling time, using the sampling time on the unified time axis as the matrix row index. The synchronous data sequence field and feature data field are used as the matrix column index. The data value, state value or feature value corresponding to the sampling time is written into the corresponding matrix cell to generate the same frequency order matrix. The characteristic data includes busbar top-pressing characteristics, braking continuity characteristics, permissible reversal characteristics, jogging discrete characteristics, and recharge blockage characteristics; The bus over-limit characteristic is obtained by subtracting the current bus voltage from the protection upper limit voltage of the DC bus voltage data; The braking duration characteristic is obtained by accumulating the conduction time of the braking unit conduction state data; The permission flip feature is derived from the number of times the battery charging permission state data switches within a single cycle; The discrete characteristics of jogging are calculated from the time interval between adjacent jogging command data; The recharge congestion characteristics are obtained by the overlapping judgment results of the supercapacitor acceptance status data, battery charging permission status data and braking unit conduction status data on the same time axis within the same operating cycle. When the supercapacitor is not accepted, the battery is not in the charging permission state, and the braking unit is in the conduction state, the corresponding time period is determined as the recharge congestion section.

[0029] In this embodiment, step S1 first reads DC bus voltage data, braking unit conduction status data, supercapacitor acceptance status data, battery charging permission status data, energy management system strategy bit data, descent command and jog command data, and protection action data from the existing control system of the ship loader / unloader. Then, it performs field mapping, timestamp extraction, status encoding unification, duplicate record merging, and invalid record removal on the above data to generate a unified format of raw time-series dataset. Through this processing, the heterogeneous data originally scattered across the frequency converter, energy storage control unit, energy management system, and control terminal is organized into a single data entry point, solving the problems of scattered data sources, inconsistent field expressions, and inconsistent time records in the prior art, which make continuous analysis difficult. This provides a unified data foundation for subsequent steps.

[0030] Step S1 further generates bus top-approaching features, braking continuity features, permitted reversal features, jogging discrete features, and recharge blockage features based on the synchronous data sequence, and writes the synchronous data sequence and feature data together into the same frequency sequence matrix. Through this processing, the original numerical values ​​and state records are further transformed into feature results that can be directly used for event extraction and chain analysis. In particular, the recharge blockage feature can directly express the overlapping relationship of the three states of supercapacitor not being accepted, battery not being in a charging permitted state, and braking unit being on within the same time period, which is beneficial to reveal the process of regenerative energy obstruction that is not easily identified in the prior art.

[0031] In this embodiment, the process goes beyond simply collecting data from the ship loader / unloader's operations. It involves continuous processing of "data unification—cyclic partitioning—time alignment—feature generation—matrix support," enabling subsequent steps to extract event nodes, establish directed connections, and form a preceding anomaly chain within the same timeline and work cycle. Compared to existing technologies described in the background section, which often rely on single-point recording, localized judgment, and lack of a continuous preceding structure, this embodiment organizes multi-source operational data from the ship loader / unloader's descent into a foundational structure capable of directly conducting accident tracing analysis using a co-frequency preceding matrix. This provides a unified input for subsequent main preceding chain extraction, counter-evidence convergence kernel generation, and historical memory feedback.

[0032] Example 3 Please refer to Figure 1 Specifically: S2 includes S21 and S22; S21. Read the same frequency sequence matrix and scan the synchronous data sequence and feature data line by line on the same time axis; When the descent command data indicates that the descent action is in the execution state, the corresponding time period is extracted and a descent event node is generated. When the bus overshoot feature corresponding to the DC bus voltage data is in an effective state, extract the corresponding time period and generate the bus overshoot event node. When the braking continuity feature corresponding to the braking unit conduction state data is in an effective state, the corresponding time period is extracted and a braking extension event node is generated. When the permit flip feature corresponding to the battery charging permit status data is in a valid state, extract the corresponding time period and generate a charging permit flip event node. When the energy management system policy bit data changes state, the corresponding time period is extracted and an energy management system policy change event node is generated. When the jog discrete feature corresponding to the jog command data is in a valid state, extract the corresponding time period and generate a repeating jog event node; When the recharge blockage feature formed by the supercapacitor acceptance status data, battery charging permission status data, and braking unit conduction status data is in a valid state, the corresponding time period is extracted and a recharge obstruction event node is generated. For each event node, record the node type, node start time, node end time, node duration, job cycle number to which the node belongs, and the data field from which the node originates, and generate an event node set.

[0033] S22. Based on the set of event nodes, establish directed connections between event nodes within the same job cycle; When the start time of the subsequent event node is after the end time of the previous event node, and there are no other inserted event nodes of the same priority between them, a time sequence connection edge representing the time sequence relationship is established. The current event node originates from the energy management system strategy bit data or battery charging permission status data, and the subsequent event node originates from the braking unit conduction status data, DC bus voltage data, or recharge blockage characteristics. Furthermore, when the subsequent event node begins within a fixed continuous sampling period window after the end of the previous event node, a control receiving connection edge representing the control state transmission process is established. When the previous event node or the next event node is a recharge obstruction event node, and it overlaps or occurs adjacent to the bus top-pressure event node and the braking extension event node in time, an energy path obstruction connection edge representing the process of restricted regenerative energy destination is established; After establishing directed edges, the node sequence formed by connecting adjacent directed edges within the same job cycle is folded into a single chain. Then, multiple chains are clustered according to the node type arrangement, edge type combination, chain duration, and time difference distribution between nodes to generate time difference folded chain clusters. The number of chains, start and end times, number of nodes, edge type combination, and job cycle number corresponding to each chain cluster are recorded.

[0034] In this embodiment, the synchronous data sequence and feature data are first scanned line by line on a unified time axis, and the operating states such as descent, bus top approach, braking extension, charging permission reversal, energy management system strategy switching, repeated jogging, and charging obstruction are extracted as event nodes. Through this processing, the change process that was originally scattered in the continuous sampled data is converted into a structured event object with fields such as start time, end time, duration, operation cycle number, and source data. This solves the problem in the prior art that only changes in a single data point can be seen and it is difficult to clearly represent continuous abnormal processes. It gives the abnormal state in the descent operation of the ship loader / unloader a unified expression form that is traceable and comparable.

[0035] After the event node is generated, this embodiment further establishes directed connections based on temporal sequence, control continuity, and energy path obstruction, so that various abnormal states are no longer isolated but form a directed relationship structure with sequential logic. This processing not only reflects the occurrence of an anomaly but also indicates which state the anomaly originated from and to which state it propagated. In particular, it connects changes in energy management system strategies and battery charging permits with subsequent braking extensions, bus overshooting, and recharge obstruction processes, solving the problem described in the background section where existing technologies only focus on local state judgments and struggle to reconstruct the anomaly evolution path.

[0036] This embodiment further folds the node sequence formed by connecting adjacent directed edges within the same work cycle into chain pieces, and then clusters multiple chain pieces according to the node type arrangement order, connection edge type combination, chain piece duration, and time difference distribution between nodes to generate time difference folded chain piece clusters. Through this processing, the originally numerous and morphologically dispersed event sequences are compressed into chain piece clusters with common structural characteristics. This not only preserves the temporal characteristics of the energy consumption anomalies of ship loading and unloading machines, but also forms structural units that facilitate subsequent unified evaluation and screening. This solves the problem in existing technologies that it is difficult to merge similar anomaly structures and identify the main anomaly chains when faced with a large number of discrete anomaly records.

[0037] Example 4 Please refer to Figure 3 Specifically: S3 includes S31 and S32; S31. Read each time difference folded chain cluster and perform traversal statistics on all time difference folded chain segments contained within the time difference folded chain cluster to obtain the total number of bus top-pressing event nodes zM, the cumulative total duration of braking extension event nodes zT, the cumulative total duration of recharge obstruction event nodes sT, the total number of charging permission flip event nodes fM, the total number of repeated jogging event nodes cM, the total duration of net interval between adjacent event nodes xT, and the number of time difference folded chain segments zN. Calculate and obtain the first evaluation value Q1 of the time difference folded chain cluster. The first evaluation value Q1 is obtained by the following formula: Q1=(zM+fM+cM)×(zT+sT+zN) / (xT+1); After obtaining the first evaluation value Q1, the central chain piece inside each time difference folded chain cluster is obtained: Specifically, the time difference folded chain piece selected from the time difference folded chain cluster is selected from the cluster whose node type arrangement order is consistent with the main sequence of the time difference folded chain cluster, and whose duration of each event node and the time difference between nodes deviate the least from the average value of the time difference folded chain cluster. This is taken as the central chain piece of the time difference folded chain cluster.

[0038] Based on the first evaluation value Q1 and the central chain of the time difference folded chain cluster, all time difference folded chain clusters are sorted from largest to smallest according to the first evaluation value Q1; When the first evaluation value Q1 of two time difference folded chain clusters is the same, the time difference folded chain cluster with more connecting edges is sorted first. When the number of control receiving connection edges is the same, the folded chain clusters with the smaller time difference between the end time of the central chain segment's end node and the corresponding time of the protection action data are sorted first. The time difference folded chain cluster that ranks first in the sorting is identified as the target time difference folded chain cluster, and the central chain corresponding to the target time difference folded chain cluster is identified as the main preorder chain.

[0039] In this embodiment, by traversing and statistically analyzing all time-difference folded chain segments within each time-difference folded chain segment cluster, the total number of bus top-pressing event nodes, the cumulative total duration of braking extension event nodes, the cumulative total duration of recharge obstruction event nodes, the total number of charging permit reversal event nodes, the total number of repeated jogging event nodes, the total duration of the net interval between adjacent event nodes, and the number of time-difference folded chain segments are extracted, and a first evaluation value is further calculated. Through this processing, the chain segment structure, which originally only reflected local anomalies, is transformed into a quantitative result that can be uniformly compared, enabling the abnormal chains formed in different work cycles to have a consistent evaluation basis, solving the problem in the prior art that it is difficult to compare abnormal records horizontally and distinguish between primary and secondary anomalies.

[0040] This embodiment uses a first evaluation value to sort the time difference folded chain clusters and extracts the central chain fragment within each time difference folded chain cluster. This allows the analysis to move beyond simply identifying "a certain chain fragment has appeared" and instead converge to "the most representative chain fragment structure in this type of chain fragment." This process preserves the common characteristics of similar anomalous chains while extracting representative chain fragments that best fit the evolution path of that type of anomalous chain. It solves the problems of existing technologies where there are many anomalous chains, their structures are discrete, and it is difficult to determine typical links.

[0041] In this embodiment, the first evaluation value is used as the primary ranking criterion during the ranking process. When the first evaluation values ​​are the same, the number of control connection edges and the time interval between the end time of the central chain segment's final node and the protection action time are introduced as further judgment conditions. This multi-level ranking method ensures that the determination of the main preceding chain considers not only the degree and duration of anomaly aggregation but also the control state transmission relationship and the proximity of accident closure, avoiding ranking bias caused by relying on only a single indicator. This results in the selected main preceding chain being closer to the actual accident formation process.

[0042] Example 5 Please refer to Figure 2 Specifically: S4 includes S41 and S42; S41. Read the main prequence chain, establish the main prequence chain node table according to the event node order in the main prequence chain, and determine the protection action time corresponding to the protection action data at the end of the main prequence chain as the closing endpoint. Perform single-node removal calculations on each event node in the main preorder chain in sequence; When an event node is removed, if the first event node can still reach the protection action event node through the remaining time sequence connection edge, the control receiving connection edge, and the energy path blocked connection edge, the event node is identified as an accompanying node. When an event node is removed, and the first event node cannot reach the protection action event node, the event node is determined to be a necessary node. After completing the single-node removal calculation, perform a two-node combined removal calculation on the event nodes in the main preorder chain. If the two event nodes are removed individually, the protection action event node remains reachable; however, if the two event nodes are removed simultaneously, the protection action event node becomes unreachable. In this case, the two event nodes are identified as a coupled node pair. Subsequently, all blocked energy path edges connected to the recharge blocked event node in the main preorder chain are extracted to form a blocked edge set; Then, based on the necessary nodes, the pairs of coupled nodes, and the set of blocking edges, the proof-of-contrast convergence kernel is determined.

[0043] S42. Extract features from the counter-evidence convergence kernel, including chain closure span Hs, number of blocked edges Ns, total duration of refill obstruction event nodes Ts, and minimum number of broken chain nodes Ks. Based on the chain closure span Hs, the number of blocked edges Ns, the total duration of the recharge obstruction event nodes Ts, and the minimum number of broken chain nodes Ks, the second evaluation value Q2 is calculated and obtained. The second evaluation value Q2 is obtained by the following formula: Q2=(Hs×Ns×Ts) / (Ks+1) After obtaining the second evaluation value Q2, the second evaluation value Q2 is fitted with the proof of contradiction convergence kernel to form the causal core structure result package.

[0044] S5 includes S51 and S52; S51, read time difference folded chain cluster, first evaluation value Q1, and causal core structure result package; Perform structural analysis on the proof-of-contrast kernel to extract the set of necessary nodes, the set of coupled node pairs, the set of blocked edges, the earliest cause time, the latest cut-off time, the cycle number of the job to which it belongs, and the order of the corresponding event nodes and the order of the directed connection edges; Using the current job cycle number as the main index, a current job cycle memory unit is established in the historical memory graph, and the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster, the main preorder chain, and the proof of contradiction convergence kernel are written into the current job cycle memory unit respectively. Among them, the first evaluation value Q1 is written into the chain evaluation field of the current job cycle memory unit, the second evaluation value Q2 is written into the causal evaluation field of the current job cycle memory unit, the time difference folding chain cluster is written into the chain structure field of the current job cycle memory unit, the main preorder chain is written into the main chain field of the current job cycle memory unit, and the proof by contradiction convergence kernel is written into the causal core field of the current job cycle memory unit. After the writing is completed, the current job circular memory unit is structurally matched with the existing job circular memory units in the historical memory graph to obtain the historical recurrence statistics.

[0045] S52. Based on historical recurrence statistics and combined with the first evaluation value Q1 and the second evaluation value Q2, calculate the feedback value Q3. The calculation formula is as follows: ; In the formula, yM represents the number of identical time-difference folded chain clusters, yN represents the number of repeated mandatory nodes, yP represents the number of repeated coupled node pairs, yL represents the number of repeated blocking edges, and RL represents the difference between the current job cycle number and the most recent matched job cycle number. After completing the feedback value calculation, a feedback result packet is generated based on the feedback value Q3, the number of identical time difference folded chain clusters yM, the number of repeated necessary nodes yN, the number of repeated coupled node pairs yP, and the number of repeated blocking edges yL. The feedback result packet is then written back to the current job cycle memory unit in the historical memory graph and output to the subsequent job cycle identification process.

[0046] In this embodiment, by performing single-node removal calculations and dual-node combination removal calculations on the main preceding chain, the originally long preceding anomaly chain is further compressed into a convergence kernel composed of necessary nodes, coupled node pairs, and blocking edge sets. This processing goes beyond simply identifying "which anomalies occur sequentially," and instead clarifies "which event nodes, if missing, cannot maintain incident closure," "which event nodes need to appear in combination to support incident closure," and "which types of energy paths are blocked and whose connecting edges are in critical positions," thereby distinguishing the dominant part of the anomaly chain from the general accompanying parts. Compared to the prior art described in the background section, which mostly focuses on single-point alarms and surface phenomenon analysis, this embodiment can converge incident analysis to a more specific causal core location.

[0047] This embodiment further extracts the chain closure span, number of blocked edges, cumulative total duration of recharge obstruction event nodes, and minimum number of broken chain nodes from the counter-evidence convergence kernel, and calculates the second evaluation value Q2 accordingly. Through this processing, the causal structure most closely related to the incident closure in the main preorder chain is quantitatively expressed, forming a causal core structure result package that includes both topological structure and numerical representation. In this way, the analysis results are no longer just a list of anomalies, but can provide a unified evaluation around "closure duration range," "degree of energy path obstruction," and "degree of dependence on key nodes," which helps to solve the problem described in the background section that although the anomalous chains are identified, they are difficult to further compress and it is difficult to determine the strength of their dominant role.

[0048] In this embodiment, the first evaluation value Q1, the second evaluation value Q2, the time difference folded chain clusters, the main preorder chain, and the proof of contradiction convergence kernel are jointly written into the historical memory graph. Memory units are established according to the current job cycle, and then the number of identical time difference folded chain clusters, the number of repeated necessary nodes, the number of repeated coupled node pairs, and the number of repeated blocking edges are extracted through structural matching. Through this processing, the analysis results of the current job cycle are no longer a one-time output, but form a historical structural sample that can be stored for a long time, reviewed, and compared.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs, characterized by: Includes the following steps: S1. Collect the operating data from the existing control system of the ship loading and unloading machine, and perform cyclic slicing, unified time scale alignment and synchronization processing on the operating data to generate the same frequency sequence matrix; S2. Extract event nodes based on the same frequency sequence matrix, and establish directed connection edges according to the time sequence relationship, control succession relationship and energy path obstruction relationship between event nodes. Cluster and fold the event nodes and directed connection edges to generate time difference folded chain clusters. S3. Evaluate the obtained time difference folded chain clusters, obtain the first evaluation value Q1, and sort all time difference folded chain clusters to obtain the main preceding chain. S3 includes S31; S31. Read each time difference folded chain cluster and perform traversal statistics on all time difference folded chain segments contained within the time difference folded chain cluster to obtain the total number of bus top-pressing event nodes zM, the cumulative total duration of braking extension event nodes zT, the cumulative total duration of recharge obstruction event nodes sT, the total number of charging permission flip event nodes fM, the total number of repeated jogging event nodes cM, the total duration of net interval between adjacent event nodes xT, and the number of time difference folded chain segments zN. Calculate and obtain the first evaluation value Q1 of the time difference folded chain cluster. The first evaluation value Q1 is obtained by the following formula: Q1=(zM+fM+cM)×(zT+sT+zN) / (xT+1); S4. Perform counter-evidence convergence processing on the main preorder chain to form a counter-evidence convergence kernel. Then, construct a second evaluation based on the counter-evidence convergence kernel, obtain the second evaluation value Q2, and obtain the causal core structure result package of the dominant accident closure. S4 includes S41 and S42; S41. Read the main prequence chain, establish the main prequence chain node table according to the event node order in the main prequence chain, and determine the protection action time corresponding to the protection action data at the end of the main prequence chain as the closing endpoint. Perform single-node removal calculations on each event node in the main preorder chain in sequence; When an event node is removed, if the first event node can still reach the protection action event node through the remaining time sequence connection edge, the control receiving connection edge, and the energy path blocked connection edge, the event node is identified as an accompanying node. When an event node is removed, and the first event node cannot reach the protection action event node, the event node is determined to be a necessary node. After completing the single-node removal calculation, perform a two-node combined removal calculation on the event nodes in the main preorder chain. If the two event nodes are removed individually, the protection action event node remains reachable; however, if the two event nodes are removed simultaneously, the protection action event node becomes unreachable. In this case, the two event nodes are identified as a coupled node pair. Subsequently, all blocked energy path edges connected to the recharge blocked event node in the main preorder chain are extracted to form a blocked edge set; Then, based on the necessary nodes, pairs of coupled nodes, and the set of blocking edges, determine the proof-of-contrast convergence kernel; S42. Extract features from the counter-evidence convergence kernel, including chain closure span Hs, number of blocked edges Ns, total duration of refill obstruction event nodes Ts, and minimum number of broken chain nodes Ks. Based on the chain closure span Hs, the number of blocked edges Ns, the total duration of the recharge obstruction event nodes Ts, and the minimum number of broken chain nodes Ks, the second evaluation value Q2 is calculated and obtained. The second evaluation value Q2 is obtained by the following formula: Q2=(Hs×Ns×Ts) / (Ks+1) After obtaining the second evaluation value Q2, the second evaluation value Q2 is fitted with the proof of contradiction convergence kernel to form the causal core structure result package. S5. Analyze the causal core structure result package, and write the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster and the counter-evidence convergence kernel into the historical memory map, and provide feedback based on the historical memory map.

2. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 1, characterized in that: S1 includes S11 and S12; S11. Read the operating data from the existing control system of the ship loading and unloading machine. The operating data includes DC bus voltage data output by the frequency converter, conduction status data of the braking unit, acceptance status data of the supercapacitor, charging permission status data of the battery, strategy bit data of the energy management system, descent command and jog command data output by the PLC or HMI, and protection action data output by the driver or PLC. The system performs field mapping, timestamp extraction, status coding standardization, duplicate record merging, and invalid record removal on the running data to generate a raw time-series dataset in a unified format. S12. Based on the original time series dataset, take the moment when the descent command first takes effect as the start time of the work cycle and take the moment when the DC bus voltage enters the steady state range and is continuously maintained for at least three sampling cycles as the end time of the work cycle, and perform loop slicing on the original time series dataset. The steady-state range is the DC bus voltage range constructed with the reference steady-state voltage as the center and a fixed voltage deviation range. The reference steady-state voltage is the average value of the DC bus voltage within the continuous sampling time window before the descent command takes effect. Within each cyclic slice job, the running data is aligned according to a unified time base to generate a synchronous data sequence. Feature data is calculated based on the synchronous data sequence. The synchronous data sequence and feature data are arranged in ascending order of sampling time, using the sampling time on the unified time axis as the matrix row index. The synchronous data sequence field and feature data field are used as the matrix column index. The data value, state value or feature value corresponding to the sampling time is written into the corresponding matrix cell to generate the same frequency order matrix. The characteristic data includes busbar top-pressing characteristics, braking continuity characteristics, permissible reversal characteristics, jogging discrete characteristics, and recharge blockage characteristics; The bus overshoot characteristic is obtained by subtracting the current bus voltage from the protection upper limit voltage of the DC bus voltage data; The braking duration characteristic is obtained by accumulating the conduction time of the braking unit conduction state data; The permission flip feature is derived from the number of times the battery charging permission state data switches within a single cycle; The discrete characteristics of jogging are calculated from the time interval between adjacent jogging command data; The recharge congestion characteristics are obtained by the overlapping judgment results of the supercapacitor acceptance status data, battery charging permission status data and braking unit conduction status data on the same time axis within the same operating cycle. When the supercapacitor is not accepted, the battery is not in the charging permission state, and the braking unit is in the conduction state, the corresponding time period is determined as the recharge congestion section.

3. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 2, characterized in that: S2 includes S21 and S22; S21. Read the same frequency sequence matrix and scan the synchronous data sequence and feature data line by line on the same time axis; When the descent command data indicates that the descent action is in the execution state, the corresponding time period is extracted and a descent event node is generated. When the bus overshoot feature corresponding to the DC bus voltage data is in an effective state, extract the corresponding time period and generate the bus overshoot event node. When the braking continuity feature corresponding to the braking unit conduction state data is in an effective state, the corresponding time period is extracted and a braking extension event node is generated. When the permit flip feature corresponding to the battery charging permit status data is in a valid state, extract the corresponding time period and generate a charging permit flip event node. When the energy management system policy bit data changes state, the corresponding time period is extracted and an energy management system policy change event node is generated. When the jog discrete feature corresponding to the jog command data is in a valid state, extract the corresponding time period and generate a repeating jog event node; When the recharge blockage feature formed by the supercapacitor acceptance status data, battery charging permission status data, and braking unit conduction status data is in a valid state, the corresponding time period is extracted and a recharge obstruction event node is generated. For each event node, record the node type, node start time, node end time, node duration, job cycle number to which the node belongs, and the data field from which the node originates, and generate an event node set.

4. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 3, characterized in that: S22. Based on the set of event nodes, establish directed connections between event nodes within the same job cycle; When the start time of the subsequent event node is after the end time of the previous event node, and there are no other inserted event nodes of the same priority between them, a time sequence connection edge representing the time sequence relationship is established. The current event node originates from the energy management system strategy bit data or battery charging permission status data, and the subsequent event node originates from the braking unit conduction status data, DC bus voltage data, or recharge blockage characteristics. Furthermore, when the subsequent event node begins within a fixed continuous sampling period window after the end of the previous event node, a control receiving connection edge representing the control state transmission process is established. When the previous event node or the next event node is a recharge obstruction event node, and it overlaps or occurs adjacent to the bus top-pressure event node and the braking extension event node in time, an energy path obstruction connection edge representing the process of restricted regenerative energy destination is established; After establishing directed edges, the node sequence formed by connecting adjacent directed edges within the same job cycle is folded into a single chain. Then, multiple chains are clustered according to the node type arrangement, edge type combination, chain duration, and time difference distribution between nodes to generate time difference folded chain clusters. The number of chains, start and end times, number of nodes, edge type combination, and job cycle number corresponding to each chain cluster are recorded.

5. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 4, characterized in that: After obtaining the first evaluation value Q1, the central chain piece inside each time difference folded chain cluster is obtained: Specifically, the time difference folded chain piece selected from the time difference folded chain cluster is selected from the cluster whose node type arrangement order is consistent with the main sequence of the time difference folded chain cluster, and whose duration of each event node and the time difference between nodes deviate the least from the average value of the time difference folded chain cluster. This is taken as the central chain piece of the time difference folded chain cluster.

6. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 5, characterized in that: S3 also includes S32; S32, based on the first evaluation value Q1 and the center chain of the time difference folded chain cluster, sort all time difference folded chain clusters in descending order according to the first evaluation value Q1; When the first evaluation value Q1 of two time difference folded chain clusters is the same, the time difference folded chain cluster with more connecting edges is sorted first. When the number of control receiving connection edges is the same, the folded chain clusters with the smaller time difference between the end time of the central chain segment's end node and the corresponding time of the protection action data are sorted first. The time difference folded chain cluster that ranks first in the sorting is identified as the target time difference folded chain cluster, and the central chain corresponding to the target time difference folded chain cluster is identified as the main preorder chain.

7. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 6, characterized in that: S5 includes S51 and S52; S51, Read the time difference folded chain cluster, the first evaluation value Q1, and the causal core structure result package; Perform structural analysis on the proof-of-contrast kernel to extract the set of necessary nodes, the set of coupled node pairs, the set of blocked edges, the earliest cause time, the latest cut-off time, the cycle number of the job to which it belongs, and the order of the corresponding event nodes and the order of the directed connection edges; Using the current job cycle number as the main index, a current job cycle memory unit is established in the historical memory graph, and the first evaluation value Q1, the second evaluation value Q2, the time difference fold chain cluster, the main preorder chain, and the proof of contradiction convergence kernel are written into the current job cycle memory unit respectively. Among them, the first evaluation value Q1 is written into the chain evaluation field of the current job cycle memory unit, the second evaluation value Q2 is written into the causal evaluation field of the current job cycle memory unit, the time difference folding chain cluster is written into the chain structure field of the current job cycle memory unit, the main preorder chain is written into the main chain field of the current job cycle memory unit, and the proof by contradiction convergence kernel is written into the causal core field of the current job cycle memory unit. After the writing is completed, the current job cyclic memory unit is structurally matched with the existing job cyclic memory units in the historical memory graph to obtain the historical recurrence statistics.

8. The method for tracing the source of energy safety accidents in port ship loading and unloading machinery based on knowledge graphs according to claim 7, characterized in that: S52. Based on historical recurrence statistics and combined with the first evaluation value Q1 and the second evaluation value Q2, calculate the feedback value Q3. The calculation formula is as follows: ； In the formula, yM represents the number of identical time-difference folded chain clusters, yN represents the number of repeated mandatory nodes, yP represents the number of repeated coupled node pairs, yL represents the number of repeated blocking edges, and RL represents the difference between the current job cycle number and the most recent matched job cycle number. After completing the feedback value calculation, a feedback result packet is generated based on the feedback value Q3, the number of identical time difference folded chain clusters yM, the number of repeated necessary nodes yN, the number of repeated coupled node pairs yP, and the number of repeated blocking edges yL. The feedback result packet is then written back to the current job cycle memory unit in the historical memory graph and output to the subsequent job cycle identification process.

Images