Knowledge graph and stream computing based industrial equipment energy efficiency root cause analysis and optimization method and system

Abstract

The application proposes an industrial equipment energy efficiency root cause analysis and optimization method and system based on a knowledge graph and a stream computing, which comprises the following steps: extracting nodes with structural, running or mechanism relationship from the knowledge graph, and calculating energy efficiency causal constraint weights; obtaining running flow data, combining the weights and a jitter suppression term to calculate a trigger judgment quantity, and generating energy efficiency alignable events; within the allowed delay range, calculating the equivalent overlap length by shifting the time interval to obtain the event alignment strength, smoothing the stable alignment strength by exponential moving average to generate a candidate energy efficiency root cause alignment result; finally, calculating the relative optimization priority based on the stable alignment strength, matching and issuing the optimization action instruction. The application defines the causal boundary by using the knowledge graph, and solves the problems of inaccurate energy efficiency root cause positioning and repeated fluctuations of conclusions in the industrial scene by using the anti-jitter trigger, uncertain delay alignment and stability smoothing mechanism in the stream computing, and realizes the closed-loop control of energy efficiency optimization.

Description

Technical Field

[0001] This invention belongs to the field of industrial Internet of Things, and in particular relates to a method and system for root cause analysis and optimization of energy efficiency of industrial equipment based on knowledge graphs and streaming computing. Background Technology

[0002] In the industrial sector, equipment energy efficiency directly impacts energy utilization efficiency, operating costs, and system safety. This is especially true in scenarios such as power, energy, rail transportation, and large industrial parks, where the large number of devices, strong system coupling, and continuously changing operating conditions mean that energy efficiency anomalies often propagate across devices and manifest over time. Existing energy efficiency analysis technologies primarily rely on energy consumption monitoring systems, historical data statistical analysis, or rule-based threshold judgments to identify and assess anomalies in individual devices or local systems. Currently, some systems are incorporating streaming computing to achieve online data acquisition and real-time analysis, while also exploring the use of knowledge graphs to describe equipment structural relationships, operational dependencies, and process logic for auxiliary localization.

[0003] However, common bottlenecks still exist in engineering applications: On the one hand, streaming analysis is usually based on fixed time windows or correlations between nearby moments, assuming that changes in equipment operation and energy efficiency are approximately synchronous or delayed in time and can be covered by the fixed window. However, in actual operation, the impact of behaviors such as start-up, shutdown, adjustment, operating condition switching, and load distribution on energy efficiency often presents uncertain delays and gradual manifestations. The delay range varies with the operating conditions and equipment status, which makes it easy for correlation inferences under the fixed window to be mismatched, thus manifesting as root cause drift and repeated conclusions. On the other hand, the engineering implementation of knowledge graphs mostly stays at the level of static modeling or relation query. Even if there are structural connections and dependency links in the graph, it is difficult to transform "relationship exists" into "relationship is established at this time" in streaming time series. When there are sampling jitter, control swings, and short-term disturbances in the field data, the analysis results are more susceptible to transient fluctuations and lack continuous consistency, making it difficult for operation and maintenance personnel to perform effective operation adjustments or energy-saving controls. Therefore, the key challenge in the root cause analysis of energy efficiency in industrial equipment is not the lack of data or relationship descriptions, but how to combine the engineering causal boundaries provided by knowledge graphs with the correspondence of events on the time axis in a streaming computing environment, handle the misalignment caused by uncertain delays and transient disturbances, and stably transform the analysis conclusions into executable optimization actions. This is the core problem that existing technologies have not yet effectively solved. Summary of the Invention

[0004] The purpose of this invention is to design a method and system for root cause analysis and optimization of energy efficiency of industrial equipment based on knowledge graphs and streaming computing. This method can achieve the integration of causal boundaries of knowledge graphs and time-series alignment of streaming computing, so that the root cause analysis conclusions have continuous consistency and engineering operability.

[0005] To achieve the above objectives, a method for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing is provided in a first aspect of the present invention, the method comprising:

[0006] Candidate equipment nodes and energy efficiency-related nodes are extracted from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes.

[0007] The number of valid relationships is counted and normalized to obtain the energy efficiency causal constraint weights;

[0008] Acquire operational flow data of industrial equipment and calculate the magnitude of operational changes and persistence factors within a sliding time window;

[0009] Calculate the original event trigger strength, which is the product of the energy efficiency causal constraint weight, the operational change magnitude, and the persistence factor;

[0010] Calculate the jitter suppression term, which is the product of the absolute value of the difference in the trigger intensity of the original event at adjacent time moments and the preset jitter suppression coefficient;

[0011] Generate a trigger determination value, wherein the trigger determination value is the difference between the original event trigger intensity and the jitter suppression term;

[0012] If the trigger determination quantity reaches the trigger threshold, an industrial equipment energy efficiency alignment event carrying a time interval is generated.

[0013] Within the allowable delay range, the equivalent overlap interval length is calculated by shifting the time interval, and the equivalent overlap interval length is the maximum overlap length between the operation change event and the energy efficiency performance change event;

[0014] Generate event alignment strength, wherein the event alignment strength is the product of a first ratio and the energy efficiency causal constraint weight, and the first ratio is the ratio of the equivalent overlap interval length to the union interval length;

[0015] A stable alignment strength is obtained by cumulatively smoothing the event alignment strength using the exponential moving average recursive formula.

[0016] If the stable alignment strength reaches the candidate determination threshold, then candidate energy efficiency root cause alignment results are generated;

[0017] The normalized relative optimization priority is calculated based on the stable alignment strength.

[0018] Based on the relative optimization priority, the target optimization action is matched from the optimization action library and an execution instruction is issued.

[0019] Furthermore, the normalization calculation yields the energy efficiency causal constraint weights, including:

[0020] The number of valid relationships between candidate device nodes and energy efficiency-related nodes is used as the numerator;

[0021] The sum of the number of valid relationships between the candidate device nodes and all energy efficiency-related nodes is used as the denominator;

[0022] The quotient of the numerator and the denominator is calculated to obtain the dimensionless relative constraint weight, which is used as the energy efficiency causal constraint weight.

[0023] Furthermore, the calculation of the range of change includes:

[0024] If the data type of the point is continuous numerical, then the difference between adjacent sampled values ​​is calculated, and the difference is normalized using a preset allowable fluctuation range or historical stable segment statistics to obtain the running change range.

[0025] If the data type of the point is discrete state type, the number of state transitions within the sliding time window is counted, and the number of state transitions is normalized using a preset upper limit of allowed transition frequency to obtain the operational change range.

[0026] Further, the calculation of the equivalent overlapping interval length includes:

[0027] Determine whether there is a direct overlap between the time intervals of operational change events and energy efficiency performance change events;

[0028] If there is a direct overlap, the length of the direct overlap is determined as the equivalent overlap interval length;

[0029] If there is no direct overlap but the interval falls within the allowable delay range, then one of the time intervals is shifted within the allowable delay range until the maximum overlap is formed, and the length of the maximum overlap is determined as the equivalent overlap interval length.

[0030] Furthermore, obtaining the stable alignment strength includes:

[0031] Get the preset smoothing coefficient;

[0032] Calculate the first product, which is the product of the preset smoothing coefficient and the stable alignment strength at the previous moment;

[0033] Calculate the second value, which is the difference between 1 and the preset smoothing coefficient;

[0034] Calculate the second product, which is the product of the second value and the event alignment strength at the current moment;

[0035] The sum of the first product and the second product is used as the stable alignment strength at the current moment.

[0036] Furthermore, the calculation of relative optimization priority includes:

[0037] The stable alignment strength of a specific candidate root cause relationship is selected as the molecule;

[0038] The sum of the stable alignment strengths of all candidate root cause relationships under the same energy efficiency node is used as the denominator;

[0039] Divide the numerator by the denominator to obtain a percentage value, which is used as the relative optimization priority.

[0040] Furthermore, the optimization action library includes parameter adjustment actions, operating mode switching actions, load distribution adjustment actions, and operation and maintenance handling actions;

[0041] The parameter adjustment action corresponds to performing a write point operation to the control system;

[0042] The operation mode switching action corresponds to issuing a mode switching command to the control system;

[0043] The load allocation adjustment action corresponds to issuing a load allocation instruction to the scheduling system;

[0044] The operation and maintenance actions correspond to generating operation and maintenance work orders or pushing alarm information.

[0045] Furthermore, the method also includes:

[0046] Generate an operation adjustment result record, which includes at least the device node identifier, associated energy efficiency node identifier, action type identifier, action issuance timestamp, and action parameter identifier corresponding to the executed action.

[0047] Furthermore, the union interval length is the length of the union of the time interval of the operational change event and the time interval of the energy efficiency performance change event.

[0048] A second aspect of the present invention provides an industrial equipment energy efficiency root cause analysis and optimization system based on knowledge graphs and streaming computing, the system comprising:

[0049] The constraint information generation module is configured to extract candidate equipment nodes and energy efficiency-related nodes from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes. The number of valid relationships is counted and normalized to obtain the energy efficiency causal constraint weights.

[0050] The event generation module is configured to acquire industrial equipment operation flow data, calculate the operation change amplitude and persistence factor within a sliding time window; calculate the original event trigger intensity, which is the product of the energy efficiency causal constraint weight, the operation change amplitude, and the persistence factor; calculate a jitter suppression term, which is the product of the absolute value of the difference between the original event trigger intensities at adjacent times and a preset jitter suppression coefficient; generate a trigger determination quantity, which is the difference between the original event trigger intensity and the jitter suppression term; if the trigger determination quantity reaches a trigger threshold, an industrial equipment energy efficiency alignable event carrying the time interval is generated.

[0051] The root cause alignment analysis module is configured to calculate the equivalent overlap interval length by shifting time intervals within an allowable delay range. The equivalent overlap interval length is the maximum overlap length between operational change events and energy efficiency performance change events. It generates an event alignment strength, which is the product of a first ratio and the energy efficiency causal constraint weight. The first ratio is the ratio of the equivalent overlap interval length to the union interval length. It then uses an exponential moving average recursive formula to cumulatively smooth the event alignment strength to obtain a stable alignment strength. If the stable alignment strength reaches a candidate determination threshold, it generates candidate energy efficiency root cause alignment results.

[0052] The optimization execution module is configured to calculate a normalized relative optimization priority based on the stable alignment strength; and to match a target optimization action from the optimization action library and issue an execution instruction based on the relative optimization priority.

[0053] The beneficial technical effects of the present invention are at least as follows:

[0054] To address the aforementioned issues, this invention provides a method and system for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing. First, the structural connections, operational dependencies, and mechanistic action links in the industrial equipment knowledge graph are transformed into computable energy efficiency causal constraint information to define the engineering boundaries of energy efficiency impact relationships. Then, during the arrival of streaming data, operational changes are selectively eventified based on these causal constraints, ensuring that the generated events naturally carry causal semantics related to energy efficiency. Furthermore, the stability of events is enhanced by suppressing event saliency and short-cycle fluctuations. Further, a temporal correspondence between operational change events and energy efficiency performance events is established at the event level. Through interval alignment metrics within an allowable delay range and recursive stable accumulation, "occasional overlap" and "continuous correspondence" are distinguished, thus forming candidate root cause alignment results that can be used for decision-making. Finally, the stable alignment strength is mapped to optimization priority, driving the selection and issuance of actions for specific equipment control interfaces and operation and maintenance processes. This allows the root cause identification results to be implemented on-site in the form of parameter adjustments, mode switching, load allocation adjustments, or operation and maintenance handling instructions. Through the above combination of technologies, this invention achieves an effective fusion of knowledge graph causal boundaries and streaming time-series alignment without relying on the assumption of fixed-time synchronization. This enables the root cause analysis conclusions to have continuous consistency and engineering operability, and can directly support the actual implementation of energy efficiency optimization for industrial equipment. Attached Figure Description

[0055] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0056] Figure 1 This is a flowchart of the method for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing, as proposed in this invention.

[0057] Figure 2 This is a system framework diagram for the root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing, as presented in this invention. Detailed Implementation

[0058] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0059] In one or more embodiments, such as Figure 1 As shown, a method for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing is disclosed. The method includes the following:

[0060] S1: Extract candidate equipment nodes and energy efficiency-related nodes from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes. Count the number of valid relationships and normalize them to obtain the energy efficiency causal constraint weights.

[0061] Specifically, this step, centered on the engineering practice of industrial equipment energy efficiency analysis, transforms the industrial equipment knowledge graph into energy efficiency causal constraint information that can directly participate in subsequent calculations. Industrial equipment knowledge graphs typically consist of structured data generated during planning and design, system integration, and long-term operation and maintenance. This includes equipment ledgers, physical or logical connections between equipment, operational dependencies, and mechanistic action chains verified by engineering practice. In engineering systems, this information often exists in the form of configuration files, relational tables, or graph-structured data, clearly describing "how equipment is connected," "how it operates collaboratively," and "under what conditions it will influence each other." In energy efficiency analysis scenarios, directly using a complete knowledge graph would introduce a large number of relationships unrelated to energy efficiency. Therefore, it is necessary to reorganize and filter the relationships in the knowledge graph around the target industrial equipment and its energy efficiency performance, focusing on the energy efficiency scenario.

[0062] In the specific implementation process, the target equipment for energy efficiency analysis and its corresponding energy efficiency performance nodes are first identified. The target equipment can be a single critical device or a functional unit composed of multiple devices, whose energy efficiency performance typically exists in the knowledge graph as independent nodes or attribute nodes. Subsequently, device nodes and their associated edges that have structural connections, operational dependencies, or mechanistic relationships with the target equipment are extracted from the knowledge graph. This extraction process combines existing relationship types in the graph, rather than simply expanding based on topological distance, thus ensuring that the retained relationships have the potential to explain energy efficiency changes from an engineering logic perspective. For example, in a compressor system, the motor, drive unit, and cooling system may all affect energy efficiency performance through different paths, while devices with only spatial proximity relationships but no operational or mechanistic connections will not be included in the candidate scope. Through this process, a set of candidate device-energy efficiency relationships can be obtained to describe "which device operational changes may affect the target energy efficiency performance."

[0063] To enable these candidate relationships to be used as constraints in subsequent steps, their relative importance needs to be further quantified. This quantification method originates from the mathematical concept of normalization, which reflects the relative strength of each relationship in the engineering structure by comparing the number of relationships between the same device node and different energy-efficiency-related nodes. Let the device node be... Energy efficiency related nodes are In the industrial equipment knowledge graph, all structural connections, operational dependencies, or mechanistic relationships marked as valid are statistically analyzed to obtain equipment nodes. Energy efficiency related nodes Count of valid relationships between Based on this count, define the device node. To Energy Efficiency Node The energy efficiency causal constraint weights are:

[0064] ;

[0065] in, Represents device node To energy efficiency related nodes The relative constraint weights are used to characterize the relative importance of this relationship in energy efficiency analysis; In the knowledge graph of industrial equipment, nodes With nodes The count values ​​that are marked as valid structural, operational, or mechanistic relationships between them are directly derived from existing relationship records in the graph; Represents device node The sum of valid relation counts pointing to all energy efficiency-related nodes is used to normalize different candidate relations, ensuring that the resulting weights reflect only relative strength and not depend on the absolute number of relations. Through this calculation, different candidate relations are mapped to a unified dimensionless weight space, facilitating threshold filtering and search space pruning in subsequent steps.

[0066] In engineering practice, the above calculation process can be directly applied to aggregate calculations on graph databases or relational tables. For example, when a device node has valid relationship records with three energy efficiency-related nodes in a knowledge graph, and their counts show significant differences, the normalized constraint weights can intuitively reflect the device's emphasis on different energy efficiency impact paths. This allows for prioritizing more engineering-significant energy efficiency impact relationships in subsequent streaming processing. In this way, the structural and dependency information originally scattered throughout the knowledge graph is compressed into a set of energy efficiency causal constraints with constraint weights. This information is output in a structured form and directly used in subsequent steps to limit the scope of event generation objects and the set of causal relationships allowed during root cause alignment analysis. This ensures that the entire energy efficiency analysis process remains within clear and computable engineering causal boundaries.

[0067] S2: Acquire industrial equipment operation flow data, calculate the operation change amplitude and persistence factor within a sliding time window; calculate the original event trigger intensity, which is the product of the energy efficiency causal constraint weight, the operation change amplitude, and the persistence factor; calculate the jitter suppression term, which is the product of the absolute value of the difference between the original event trigger intensities at adjacent times and a preset jitter suppression coefficient; generate a trigger determination quantity, which is the difference between the original event trigger intensity and the jitter suppression term; if the trigger determination quantity reaches a trigger threshold, generate an industrial equipment energy efficiency alignable event carrying the time interval;

[0068] Specifically, based on the energy efficiency causal constraint information formed in Step 1, this step transforms the continuous operational flow data generated by industrial equipment during actual operation into energy efficiency-alignable events that can be used for subsequent time-aligned analysis. Operational data in industrial settings is typically generated in real time by control and monitoring systems, such as through DCS, PLC, or edge acquisition gateways that continuously output equipment status, control variables, and process variables. This data is high-frequency and continuous in time, and its content is a mixture of real operational changes and numerous subtle disturbances. To ensure that the event generation process directly serves energy efficiency root cause analysis, this step uses the candidate energy efficiency impact relationships output in Step 1 as constraint boundaries, performing calculations only around the equipment nodes and energy efficiency nodes corresponding to these relationships, thus ensuring that the events themselves carry explicit energy efficiency analysis semantics.

[0069] In practical implementation, the system first loads the energy efficiency causal constraint information output from step one, and then performs point-level mapping of the industrial equipment operation flow data based on the candidate relationships within it. Each equipment node is identified. This corresponds to a set of operational element points, which are derived from the control system engineering configuration or asset ledger, such as operating mode identifiers, start / stop status, key control variables, and key process variables; each energy efficiency-related node identifier. This corresponds to a set of energy efficiency performance data points, such as energy efficiency observations or their derivatives. After mapping, the streaming computing module only processes candidate relationships. The covered device nodes and their operational elements are continuously analyzed. Although the data of other devices are still collected, they do not enter the event generation chain, thus completing a structural tailoring for energy efficiency goals before the operational data enters the analysis.

[0070] The characterization of operational changes employs a three-element combination of "amplitude, duration, and causal weighting." Firstly, within the sliding time period, the device nodes... The operational element locations are aggregated and calculated online to obtain the operational change range. For continuous numerical data points, It is obtained by normalizing the differences between adjacent sampled values ​​through the allowable fluctuation range or historical stable segment statistics of the engineering configuration; for discrete state points... The persistence factor is obtained by normalizing the number of state transitions within the sliding time period after applying the upper limit of allowed transition frequency. Subsequently, the percentage of samples where the "change magnitude exceeds the minimum effective change threshold" is calculated within the same sliding time period. This factor reflects the degree of continuity of operational changes over time. The above... and All of these are window aggregation results derived from running stream data, which can be implemented in a streaming computing framework with a fixed window length and sliding step size.

[0071] After obtaining the operational variation range and persistence factor, these are compared with the energy efficiency causal constraint weights output from step one. Combine to construct candidate relationships In time location The event trigger strength. This construction originates from the mathematical concept of weighted scoring and the combinatorial logic of "prior constraints × observation strength × stability" in engineering signal processing. Its specific calculation method is as follows:

[0072] ;

[0073] in, Represents device node Pointing to energy efficiency nodes by the influence relationship of candidate energy efficiency. The intensity of the original event trigger; The energy efficiency causal constraint weights obtained in step one are derived from the normalized result of the knowledge graph relation count; For device nodes The magnitude of operational changes originates from the normalized differences or state transition aggregations of the operational flow data; The persistence factor is derived from the proportion of valid changing samples within the sliding time period. Through multiplicative combination, when any factor is low, the overall trigger strength decreases accordingly, thus conforming to the engineering intuition in industrial scenarios that "the impact must simultaneously possess prior rationality, significant change, and persistence."

[0074] Considering the short-cycle jitter and back-and-forth oscillation of control quantities that are common in industrial control scenarios, relying solely on Events with semantically unstable characteristics are easily generated. Therefore, a jitter suppression term based on the difference between adjacent time steps is introduced in the trigger determination stage. This concept originates from the total variation penalty in signal processing, penalizing transient jumps in trigger strength to enhance the event's sensitivity to continuous changes. The jitter-suppressed trigger determination quantity is defined as follows:

[0075] ;

[0076] in, This indicates the trigger determination value after jitter suppression; The original trigger strength from the previous moment is obtained from the same computational link cache; The jitter suppression coefficient, derived from engineering configuration parameters, is used to adjust the penalty intensity for short-period, sharp fluctuations. This calculation is performed in a single step within a streaming environment, without involving global optimization, but it effectively suppresses trigger intensity fluctuations caused by sampling glitches or control jitter.

[0077] In the event generation decision, Compare with the trigger threshold configured in the project. When When the triggering condition is met, the system generates an event instance and sets the duration factor within the sliding time period. The changing trend determines the start and end time intervals of the event: the start point corresponds to the moment when the persistence factor begins to rise steadily and reaches the effective change condition, and the end point corresponds to the moment when the persistence factor falls back to the end condition. Event instances are divided into two categories according to their location: operational change events and energy efficiency performance change events. Operational change events are caused by equipment nodes. The corresponding operational element points trigger generation, and energy efficiency performance change events are generated by energy efficiency nodes. The corresponding energy efficiency performance points are triggered and generated, and both are represented by the same structured field and enter the same event set.

[0078] By continuously executing the above process, the industrial equipment operation flow data is transformed in real time into an energy efficiency-alignable event set for the industrial equipment. This event set, as the output of this step, is continuously added in a streaming manner. Each event instance carries causal constraint semantics from step one and evidence of changes from the field operation data, enabling subsequent steps to directly perform time alignment and delay analysis around the events. This provides stable support for energy efficiency root cause analysis and optimization in complex industrial scenarios.

[0079] S3: Within the allowable delay range, calculate the equivalent overlap interval length by shifting the time interval. The equivalent overlap interval length is the maximum overlap length between the operational change event and the energy efficiency performance change event. Generate an event alignment strength, which is the product of a first ratio and the energy efficiency causal constraint weight. The first ratio is the ratio of the equivalent overlap interval length to the union interval length. Use the exponential moving average recursive formula to perform cumulative smoothing on the event alignment strength to obtain a stable alignment strength. If the stable alignment strength reaches the candidate determination threshold, generate candidate energy efficiency root cause alignment results.

[0080] Specifically, this step takes the industrial equipment energy efficiency alignable event set output from step two as input, performs energy efficiency root cause alignment analysis in a streaming computing environment, and outputs candidate energy efficiency root cause alignment results. The event instances in the event set already carry device node identifiers. Energy efficiency node label Along with the start and end time interval of the event, it also carries the trigger determination value calculated in step two. The representative value is thus used to define the structured semantics of "which device, which energy efficiency, and within what time period the significant change occurred" when an event enters this step. This step categorizes these events by energy efficiency node. The system organizes events into time-ordered queues, and maintains a sliding buffer of recent events within each queue; when a new event enters the buffer, the system selects events based on their candidate relationships. Form event pairs within the buffer: [device node] The corresponding runtime change event is denoted as Energy efficiency nodes The corresponding energy efficiency performance change event is denoted as ;in and Both are derived from the event instance fields generated in step two, the difference being the source of the event instance locations. By device node The mapping is triggered by the location of the running element. Energy efficiency nodes The mapping of energy efficiency performance points triggers generation. The system generates each pair... The alignment strength is calculated to assess the degree of correspondence of the candidate relationship over time. This approach corresponds to a typical industrial scenario: operational changes and energy efficiency changes often exhibit a pattern of "operational changes occurring first, followed by energy efficiency changes, with the delay magnitude varying with operating conditions." Therefore, the alignment analysis focuses on the relative position and overlap of event intervals, rather than on synchronous correlations within a fixed window.

[0081] The construction of event alignment strength originates from mathematical set similarity measures, particularly the "intersection-union ratio" form of the Jaccard similarity coefficient; the original form of the Jaccard similarity coefficient is based on two sets... definition To measure similarity, this step replaces "set" with "time interval" and "set size" with "interval length" to obtain the intersection-union ratio (IU) of the time intervals; it also introduces the energy efficiency causal constraint weights output from step one. As a priori modulation term, it enables more engineering-reasonable candidate relationships to achieve higher alignment strength under the same time similarity. To adapt to the non-fixed delays commonly found in industrial energy efficiency, an allowable delay range is adopted for the "intersection". Equivalent overlap within: When a running event and an energy efficiency event do not directly overlap but the interval falls within... At that time, through Within the allowed range, translate one interval to form the maximum reachable overlap with the other interval, and use this maximum overlap as the equivalent overlap. This yields the event alignment strength:

[0082] ;

[0083] in, Indicates candidate relationship Next pair of running events Energy Efficiency Issues Alignment strength; The energy efficiency causal constraint weights formed in step one are used to reflect the engineering prior importance of this relationship; Represents device node The corresponding runtime time range, Indicates energy efficiency node The corresponding energy efficiency event time intervals are both directly derived from the event instance field in step two; This indicates the allowable delay range, which is an engineering configuration parameter used to reflect the typical time scale of energy efficiency response under this type of equipment or this type of operating condition; This represents the interval length operator, used to calculate the length of a time interval; Indicates in The equivalent overlapping interval obtained within the allowable range, This represents the union of two intervals. Both sides of the expression maintain dimensionality: the numerator and denominator are both quantities of the same kind calculated using length operators, their ratio is dimensionless, and the product is... It is also a dimensionless weight, therefore This is a dimensionless strength quantity, consistent with the engineering principle that "alignment strength is a relative quantity." The connection between this formula and step two lies in: the event interval and... Derived from the event instance field, Since the event originates from step one and forms the same semantic link during the event generation process in step two, the alignment strength calculation simultaneously reflects both engineering priors and on-site event evidence.

[0084] Alignment strength This describes "the degree of alignment of a pair of events at the current moment," while root cause analysis in industrial settings emphasizes "correspondences that repeatedly hold true during continuous operation." To extend single-event alignment to streaming stable alignment, the recursive form of the Exponential Moving Average (EMA) from classical digital signal processing is adopted: the original form of EMA is... This step is used to smooth sequences and emphasize persistent trends without storing the entire history. Replace with the alignment strength at the current moment. ,Will Replace with stable alignment strength This allows for the streaming accumulation and smoothing of alignment strength.

[0085] ;

[0086] in, Indicates candidate relationship In time location Stable alignment strength; The stable alignment strength of the previous moment is obtained from the streaming state storage cache; The event alignment strength is calculated using the previous formula at the current moment; is the smoothing coefficient, and is the engineering configuration parameter used to adjust the proportion of influence of historical alignment strength on the current result. The logical relationship and derivation chain of this formula are as follows: first, the single alignment strength is obtained by combining the intersection-union ratio of time intervals with the engineering weight. Then The stable alignment strength is obtained by recursion using the EMA input. This distinguishes between "occasional overlap" and "continuous correspondence." Dimensional consistency also applies: and All are dimensionless intensity quantities. For dimensionless coefficients, add the same quantities at both ends of the recursion, and the result remains dimensionless.

[0087] Numerical implementations can directly substitute the operator based on the length of the event interval for calculation. Assume the weight of a candidate relation given in step one is... The interval of running events reached at a certain moment Energy efficiency event interval Within the allowable delay range The equivalent overlapping interval length formed within is The length of the union interval is Then the intersection and union ratio is Substituting into the first equation, we get Let's assume the stable alignment strength of the cache at the previous time step was... Smoothing coefficient is taken Substituting into the second equation, we get In streaming operation, as new events continuously enter the buffer queue, the computation is performed according to the candidate relationship. Continuous updates ;when When the candidate criteria for engineering configuration are met, the corresponding Write the candidate energy efficiency root cause alignment result set and synchronously record its current status. This serves as a characterization of the alignment strength of the candidate relation. Therefore, the candidate energy efficiency root cause alignment result set uses the candidate relation... The output of the basic unit is used directly for the next step of stability verification and optimization.

[0088] S4: Calculate the normalized relative optimization priority based on the stable alignment strength; match the target optimization action from the optimization action library according to the relative optimization priority and issue the execution instruction;

[0089] Specifically, this step uses the candidate energy efficiency root cause alignment result set output from step three as the decision-making basis, transforming the stable aligned root cause relationships into executable energy efficiency optimization actions for industrial equipment and forming operational adjustment results. The candidate energy efficiency root cause alignment result set is based on candidate relationships. As the basic unit, it contains the device node identifier. Energy efficiency node label and stable alignment strength .in The prior engineering weights are derived recursively from step three in the streaming alignment analysis. Alignment strength with events The time consistency is integrated into a single quantity, which can be directly used as a measure of the "strength of the root cause relationship in the current operational phase." Optimization execution in industrial settings typically requires prioritizing and implementing each item sequentially to avoid operational uncertainties caused by simultaneous interventions at multiple points. Therefore, this step first... Converted to the same energy efficiency node The optimization priorities are compared and sorted, and then the selection and issuance of actions are driven by these priorities.

[0090] The optimization priority calculation employs the normalization concept from mathematics, which originates from the "proportion normalization" in probability theory and statistics (dividing a set of non-negative weights by their sum to obtain a relative proportion). This method is often used to uniformly map the intensity of multiple candidate factors to a rankable relative scale. This step applies this concept to the same energy efficiency node. The set of candidate root causes, and the stable alignment strength of each candidate root cause. Normalization to relative priority This is to provide a clear order of execution when addressing the same energy efficiency issue. The specific calculation is as follows:

[0091] ;

[0092] in, Indicates at the energy efficiency node In the corresponding candidate root cause set, candidate relations The relative optimization priority; The stable alignment strength is the output of step three; Indicates the current energy efficiency node The corresponding set of all candidate root cause relationships, the set of elements is Candidate relations in the form of [formula missing]. The derivation chain of this formula is: [formula missing] "strength of root cause validity" [formula missing]. This indicates that, again for the same The total strength is obtained by summing all candidate relations, and the relative priority is obtained in proportion form, thus satisfying the requirement. And achieve stable sorting. Dimensionality consistency holds naturally in this formula: The dimensionless intensity quantity obtained in step three, with the denominator being the sum of similar quantities, and the ratio between the two being dimensionless, representing the relative priority. As a dimensionless quantity, it conforms to the common sense that "priority is relative quantity".

[0093] In obtaining Then, the system is configured according to energy efficiency nodes. The candidate relationships are sorted by granularity, and then selected in order of their relationship with the device node. Matching optimized actions. Action selection is achieved through a pre-configured optimized action library, established during the system integration phase. Data sources include: equipment type and control interface definitions (from DCS / PLC project configuration), operating modes and adjustable parameter lists (from equipment manuals and control strategy configurations), and maintenance handling rules (from maintenance procedures and work order templates). Each action item in the action library includes an executable carrier and a delivery method. For example, parameter adjustment actions correspond to control system write operations (writing target values ​​or setpoints), operating mode switching actions correspond to control command delivery (triggering start / stop, switching operating modes, or switching control loops), load allocation adjustment corresponds to scheduling command delivery (changing the allocation coefficient of parallel systems or switching between primary and backup), and maintenance handling corresponds to work order generation and alarm push. During system execution, actions are delivered item by item according to the sorting results, and the applicable conditions of the actions are matched and verified before delivery. For example, whether the current operating mode of the equipment allows the parameter adjustment, whether the current interlocking status allows mode switching, and whether the action execution conflicts with the established scheduling strategy. The action issuance interface can be implemented by the write interface provided by the industrial protocol gateway or control system, while operation and maintenance are handled through the work order interface or message push interface of the operation and maintenance platform.

[0094] The following numerical calculation illustrates the complete computational chain from candidate root cause alignment results to optimization priority ranking and action execution, for illustrative purposes only. Let's assume a certain energy efficiency node... Corresponding to three candidate root cause relationships , , The stable alignment strengths output in step three are respectively , , Then the sum of the denominators is Substituting into the above equation, we get... , , Sort according to this. Priority, then again for The system then searches for device nodes in the optimized action library. For the corresponding action items, such as when the device is a drive unit and the corresponding energy efficiency issue is increased energy consumption, the action templates for "adjust key control quantity settings" or "switch to energy-saving operation mode" should be selected first, and issued through the control system write point or command interface; after completion, the actions should be processed sequentially. , Execute the corresponding action. The numerical chain in this example demonstrates the operable computational process from the result of step three to the actual execution order, and the computation depends only on the candidate result set. With candidate set This step does not introduce new root cause calculation variables. The output of this step is the industrial equipment operation adjustment result, generated as a structured record, which includes at least the equipment node identifier corresponding to the executed action, the associated energy efficiency node identifier, the action type, the action issuance time, and the action parameter identifier used for backtracking. This output comes directly from the analysis of… The process of matching and issuing actions after sorting enables energy efficiency root cause analysis and optimization to be completed in the same logical link, and provides a consistent record carrier for subsequent operation effect evaluation or operation and maintenance audit.

[0095] In one or more embodiments, such as Figure 2 As shown, a system for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing is disclosed. The system includes:

[0096] The constraint information generation module is configured to extract candidate equipment nodes and energy efficiency-related nodes from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes. The number of valid relationships is counted and normalized to obtain the energy efficiency causal constraint weights.

[0097] The event generation module is configured to acquire industrial equipment operation flow data, calculate the operation change amplitude and persistence factor within a sliding time window; calculate the original event trigger intensity, which is the product of the energy efficiency causal constraint weight, the operation change amplitude, and the persistence factor; calculate a jitter suppression term, which is the product of the absolute value of the difference between the original event trigger intensities at adjacent times and a preset jitter suppression coefficient; generate a trigger determination quantity, which is the difference between the original event trigger intensity and the jitter suppression term; if the trigger determination quantity reaches a trigger threshold, an industrial equipment energy efficiency alignable event carrying the time interval is generated.

[0098] The root cause alignment analysis module is configured to calculate the equivalent overlap interval length by shifting time intervals within an allowable delay range. The equivalent overlap interval length is the maximum overlap length between operational change events and energy efficiency performance change events. It generates an event alignment strength, which is the product of a first ratio and the energy efficiency causal constraint weight. The first ratio is the ratio of the equivalent overlap interval length to the union interval length. It then uses an exponential moving average recursive formula to cumulatively smooth the event alignment strength to obtain a stable alignment strength. If the stable alignment strength reaches a candidate determination threshold, it generates candidate energy efficiency root cause alignment results.

[0099] The optimization execution module is configured to calculate a normalized relative optimization priority based on the stable alignment strength; and to match a target optimization action from the optimization action library and issue an execution instruction based on the relative optimization priority.

[0100] It is worth noting that the specific workflow of the industrial equipment energy efficiency root cause analysis and optimization system based on knowledge graph and streaming computing provided in this embodiment of the invention is the same as that of the industrial equipment energy efficiency root cause analysis and optimization method based on knowledge graph and streaming computing described in the above embodiment, and will not be repeated here.

[0101] This invention also provides an industrial equipment energy efficiency root cause analysis and optimization device based on knowledge graphs and streaming computing, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the industrial equipment energy efficiency root cause analysis and optimization method based on knowledge graphs and streaming computing, for example... Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above system embodiments.

[0102] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization device.

[0103] The knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization device can be a desktop computer, laptop, handheld computer, or cloud server, etc. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization device may also include input / output devices, network access devices, buses, etc.

[0104] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization device, connecting all parts of the device via various interfaces and lines.

[0105] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization equipment. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the operation of the air conditioner controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0106] The module integrating the knowledge graph and streaming computing-based industrial equipment energy efficiency root cause analysis and optimization equipment, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0107] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0108] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing, characterized in that, The method includes: Candidate equipment nodes and energy efficiency-related nodes are extracted from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes. The number of valid relationships is counted and normalized to obtain the energy efficiency causal constraint weights; Acquire operational flow data of industrial equipment and calculate the magnitude of operational changes and persistence factors within a sliding time window; Calculate the original event trigger strength, which is the product of the energy efficiency causal constraint weight, the operational change magnitude, and the persistence factor; Calculate the jitter suppression term, which is the product of the absolute value of the difference in the trigger intensity of the original event at adjacent time moments and the preset jitter suppression coefficient; Generate a trigger determination value, wherein the trigger determination value is the difference between the original event trigger intensity and the jitter suppression term; If the trigger determination quantity reaches the trigger threshold, an industrial equipment energy efficiency alignment event carrying a time interval is generated. Within the allowable delay range, the equivalent overlap interval length is calculated by shifting the time interval, and the equivalent overlap interval length is the maximum overlap length between the operation change event and the energy efficiency performance change event; Generate event alignment strength, wherein the event alignment strength is the product of a first ratio and the energy efficiency causal constraint weight, and the first ratio is the ratio of the equivalent overlap interval length to the union interval length; A stable alignment strength is obtained by cumulatively smoothing the event alignment strength using the exponential moving average recursive formula. If the stable alignment strength reaches the candidate determination threshold, then candidate energy efficiency root cause alignment results are generated; The normalized relative optimization priority is calculated based on the stable alignment strength. Based on the relative optimization priority, the target optimization action is matched from the optimization action library and an execution instruction is issued.

2. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The normalization calculation yields the energy efficiency causality constraint weights, including: The number of valid relationships between candidate device nodes and energy efficiency-related nodes is used as the numerator; The sum of the number of valid relationships between the candidate device nodes and all energy efficiency-related nodes is used as the denominator; The quotient of the numerator and the denominator is calculated to obtain the dimensionless relative constraint weight, which is used as the energy efficiency causal constraint weight.

3. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The calculation of the change range includes: If the data type of the point is continuous numerical, then the difference between adjacent sampled values ​​is calculated, and the difference is normalized using a preset allowable fluctuation range or historical stable segment statistics to obtain the running change range. If the data type of the point is discrete state type, the number of state transitions within the sliding time window is counted, and the number of state transitions is normalized using a preset upper limit of allowed transition frequency to obtain the operational change range.

4. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The calculation of the equivalent overlapping interval length includes: Determine whether there is a direct overlap between the time intervals of operational change events and energy efficiency performance change events; If there is a direct overlap, the length of the direct overlap is determined as the equivalent overlap interval length; If there is no direct overlap but the interval falls within the allowable delay range, then one of the time intervals is shifted within the allowable delay range until the maximum overlap is formed, and the length of the maximum overlap is determined as the equivalent overlap interval length.

5. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The process of obtaining stable alignment strength includes: Get the preset smoothing coefficient; Calculate the first product, which is the product of the preset smoothing coefficient and the stable alignment strength at the previous moment; Calculate the second value, which is the difference between 1 and the preset smoothing coefficient; Calculate the second product, which is the product of the second value and the event alignment strength at the current moment; The sum of the first product and the second product is used as the stable alignment strength at the current moment.

6. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The calculation of relative optimization priority includes: The stable alignment strength of a specific candidate root cause relationship is selected as the molecule; The sum of the stable alignment strengths of all candidate root cause relationships under the same energy efficiency node is used as the denominator; Divide the numerator by the denominator to obtain a percentage value, which is used as the relative optimization priority.

7. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The optimization action library includes parameter adjustment actions, operating mode switching actions, load distribution adjustment actions, and operation and maintenance actions; The parameter adjustment action corresponds to performing a write point operation to the control system; The operation mode switching action corresponds to issuing a mode switching command to the control system; The load allocation adjustment action corresponds to issuing a load allocation instruction to the scheduling system; The operation and maintenance actions correspond to generating operation and maintenance work orders or pushing alarm information.

8. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The method further includes: Generate an operation adjustment result record, which includes at least the device node identifier, associated energy efficiency node identifier, action type identifier, action issuance timestamp, and action parameter identifier corresponding to the executed action.

9. The method for root cause analysis and optimization of industrial equipment energy efficiency based on knowledge graphs and streaming computing as described in claim 1, characterized in that, The union interval length is the length of the union of the time interval of the operational change event and the time interval of the energy efficiency performance change event.

10. A system for root cause analysis and optimization of energy efficiency in industrial equipment based on knowledge graphs and streaming computing, characterized in that, The system includes: The constraint information generation module is configured to extract candidate equipment nodes and energy efficiency-related nodes from the industrial equipment knowledge graph. The candidate equipment nodes have structural connections, operational dependencies, or mechanistic relationships with the target equipment nodes. The number of valid relationships is counted and normalized to obtain the energy efficiency causal constraint weights. The event generation module is configured to acquire industrial equipment operation flow data, calculate the operation change amplitude and persistence factor within a sliding time window; calculate the original event trigger intensity, which is the product of the energy efficiency causal constraint weight, the operation change amplitude, and the persistence factor; calculate a jitter suppression term, which is the product of the absolute value of the difference between the original event trigger intensities at adjacent times and a preset jitter suppression coefficient; generate a trigger determination quantity, which is the difference between the original event trigger intensity and the jitter suppression term; if the trigger determination quantity reaches a trigger threshold, an industrial equipment energy efficiency alignable event carrying the time interval is generated. The root cause alignment analysis module is configured to calculate the equivalent overlap interval length by shifting time intervals within an allowable delay range. The equivalent overlap interval length is the maximum overlap length between operational change events and energy efficiency performance change events. It generates an event alignment strength, which is the product of a first ratio and the energy efficiency causal constraint weight. The first ratio is the ratio of the equivalent overlap interval length to the union interval length. It then uses an exponential moving average recursive formula to cumulatively smooth the event alignment strength to obtain a stable alignment strength. If the stable alignment strength reaches a candidate determination threshold, it generates candidate energy efficiency root cause alignment results. The optimization execution module is configured to calculate a normalized relative optimization priority based on the stable alignment strength; and to match a target optimization action from the optimization action library and issue an execution instruction based on the relative optimization priority.

Images