A load scheduling method and device based on process dynamic analysis

Abstract

This invention discloses a load dispatching method and apparatus based on process dynamic analysis, belonging to the field of power dispatching. The method includes extracting load time-series data from real-time telemetry data, production plan data, and state time-series data, performing wavelet transform, matching the separated high-frequency transient components and low-frequency steady-state baseband components with real-time work order data according to preset process constraints, obtaining matching results; calculating the influence intensity value based on the process constraints, matching results, and identified target process links; searching within a decision space composed of production risk and production benefit indicators calculated based on the influence intensity value to generate a set of flexible dispatching plans; for each flexible dispatching plan, calculating the total entropy value of disturbance propagation based on the generated process influence chain, calculating the real-time safety entropy value based on the adjusted voltage amplitude and active power sequence, and determining the final flexible dispatching plan based on Pareto mapping. By implementing this invention, the problem of inaccurate dispatching in existing technologies is solved.

Description

Technical Field

[0001] This invention relates to the field of power dispatching technology, and in particular to a load dispatching method and apparatus based on process dynamic analysis. Background Technology

[0002] In the scenario of coordinated management and control of industrial production and power grid dispatch, real-time tracking and dynamic response to load demand are key to ensuring production continuity and the safe and stable operation of the power grid.

[0003] In existing technologies, load dispatching largely relies on response mechanisms based on fixed thresholds or static scheduling models for dispatching and control, lacking the dynamic perception capability of the correlation between transient load fluctuations and production processes. When unexpected situations such as equipment start-up and shutdown and process switching occur during actual production, they can cause rapid load fluctuations. Existing dispatching systems often can only passively respond to load fluctuations, making it difficult to establish a precise causal relationship between load fluctuations and production processes. The lack of dynamic analysis of load and production leads to inaccurate dispatching responses, affecting production efficiency and easily causing grid load imbalances, threatening grid security. Summary of the Invention

[0004] This invention provides a load scheduling method and apparatus based on process dynamic analysis, which can effectively solve the problem that existing technologies are unable to establish accurate causal relationships between load fluctuations and production processes, lack dynamic analysis of load and production, and thus lead to inaccurate scheduling responses.

[0005] An embodiment of the present invention provides a load scheduling method based on process dynamic analysis, comprising: Acquire real-time telemetry data, production plan data, status time-series data of equipment IoT sensors, and real-time work order data; Load time series data is extracted from real-time telemetry data, production plan data and status time series data. Wavelet transform is performed on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. Anchoring and matching are performed based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to the preset process constraint relationship to obtain the matching result; Based on the matching results, the target process step is identified from the real-time work order data. The impact intensity value used to characterize the process interruption is calculated according to the process constraint relationship, the target process step and the matching results. Based on the impact intensity value and the target process step, topology analysis is performed to generate the process impact chain. Based on the influence intensity value in the process influence chain and the process constraint relationship, the production risk index and production benefit index are calculated. A rolling path search is performed in the decision grid space composed of the production risk index and the production benefit index to generate a set of flexible scheduling plans. For each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. Pareto mapping is performed based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

[0006] Furthermore, load time-series data is extracted based on real-time telemetry data, production plan data, and status time-series data, including: The first data stream is obtained by aligning the timestamps of each data item in the real-time telemetry data, production plan data, and status time sequence data. The first data stream is converted into a data format according to a preset communication protocol to obtain a protocol data stream. The protocol data stream is then aggregated and encapsulated based on a preset data topic to generate a real-time status data stream. Active power time-series data is parsed and extracted from the real-time status data stream and used as load time-series data.

[0007] Furthermore, wavelet transform is performed on the load time series data to separate the high-frequency transient components and the low-frequency steady-state baseband components, including: Based on the load time series data, a discrete wavelet transform with a preset decomposition scale is performed to obtain the approximate coefficients and detail coefficients of the load time series data at each decomposition scale. Wavelet reconstruction is performed on the approximation coefficients at the maximum decomposition scale to obtain the low-frequency steady-state baseband components of the load time series data. Wavelet reconstruction is performed on the detail coefficients at all decomposition scales except the maximum decomposition scale. All detail components obtained from the wavelet reconstruction are superimposed to obtain the high-frequency transient components of the load time series data.

[0008] Furthermore, anchoring and matching are performed based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to preset process constraints to obtain matching results, including: The local extreme points of the low-frequency steady-state baseband components are used as power feature points, and the starting point of the high-frequency transient components exceeding the preset component threshold is used as transient event feature points. Based on real-time work order data, extract equipment status change signals and process code switching signals, and use these signals as work order event feature points. The initial association is constructed by matching timestamps based on power feature points, transient event feature points, and work order event feature points; wherein, if the absolute time difference between any pair of feature points is less than the allowable time tolerance of the corresponding process in the preset process constraint relationship, the initial association is constructed. Based on the equipment start-stop power logic and process flow sequence in the process constraint relationship, the consistency of the feature point pairs that have been initially associated is checked, and the feature point pairs that contradict the equipment start-stop power logic or process flow sequence are deleted to obtain the verified associated point pairs. Each verified associated point pair is encapsulated as a mapping entry, all mapping entries are aggregated into a mapping relationship set, and the mapping relationship set is used as the matching result.

[0009] Furthermore, based on process constraints, target process steps, and matching results, the impact intensity value used to characterize process interruption is calculated. Based on the impact intensity value and the target process step, topology analysis is performed to generate a process influence chain, including: Find all subsequent processes that have the target process as a direct predecessor from the process constraints, and construct a set of direct subsequent processes; Based on the dependencies between processes in the process constraints, starting from the set of directly dependent processes, iteratively query all indirectly dependent processes to construct the complete set of affected processes; Based on the matching results, determine the amplitude corresponding to the transient event feature point associated with the work order event feature point, and take the transient event feature point in the mapping relationship set corresponding to the amplitude as the load transient feature point; The baseline influence coefficient of the target process step in the process constraint relationship is multiplied with the amplitude of the load transient characteristic point to obtain the influence intensity value used to characterize the process interruption. Starting with the impact intensity value, based on the dependencies between processes, the impact intensity propagation calculation is performed on each process in the set of all affected processes. The dependency path is multiplied by the corresponding propagation attenuation factor. For each process in the set of all affected processes, the impact intensity value of all preceding processes of the current process is attenuated according to the impact propagation weight factor and then summed to obtain the cumulative impact intensity value of the current process. The impact propagation weight factor is preset according to the urgency level of the process, and the propagation attenuation factor is preset according to the dependency path between processes. The target process step and subsequent processes whose cumulative influence intensity value exceeds the preset intensity threshold are organized into a topological structure according to their dependencies to generate a process influence chain.

[0010] Furthermore, production risk indicators include process delay risk indicators and quality deviation risk indicators; production efficiency indicators include capacity utilization rate indicators and expected return indicators. Based on the influence intensity values ​​in the process influence chain and the process constraint relationships, production risk indicators and production benefit indicators are calculated, including: The cumulative influence intensity value of each process is determined based on the influence intensity value in the process influence chain; Multiply the cumulative impact strength value and the time conversion coefficient in the process constraint relationship to obtain the delay time of each process. The ratio of the delay time of each process to the preset standard working hours is used as the process delay risk indicator, and the product of the process quality sensitivity coefficient and the cumulative impact intensity value in the process constraint relationship is used as the quality deviation risk indicator. Add the standard working hours of each process to the delay time to obtain the estimated working hours of each process; The ratio of estimated working hours to the preset total planned working hours is used as the capacity utilization rate indicator. The expected revenue indicator is obtained by summing the product of the output unit price of each process and the preset output quantity.

[0011] Furthermore, a rolling path search is performed within the decision grid space comprised of production risk indicators and production efficiency indicators to generate a set of flexible scheduling plans, including: A two-dimensional decision grid space is constructed using the weighted sum of process delay risk indicators and quality deviation risk indicators as the first coordinate axis, and the weighted sum of capacity utilization rate indicators and expected return indicators as the second coordinate axis. Based on the process delay risk index, quality deviation risk index, capacity utilization rate index, and expected return index under the current production status, calculate the coordinates of the current status in the two-dimensional decision grid space and use them as the initial point. Based on the production time and output constraints in the production plan data, the target area is determined in the two-dimensional decision grid space. In the two-dimensional decision grid space, starting from the initial point and ending at the target area, a path search is performed according to the preset cluster search method to obtain several candidate decision paths and their corresponding costs. Candidate decision paths with a cost value less than a preset cost threshold are selected as the preferred decision paths. Based on the coordinates of the decision state point of each preferred decision path, the start and stop times, production order, and resource allocation parameters of the corresponding processes in the process influence chain are adjusted to generate a flexible scheduling plan. The flexible scheduling plans of all preferred decision paths are then integrated to obtain a flexible scheduling plan set.

[0012] Furthermore, for each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated, including: For each flexible scheduling plan in the flexible scheduling plan set, the estimated execution time sequence of each process is generated based on the adjustment instructions of the corresponding process in the process influence chain. For each process, the estimated execution time series is calculated using first-order difference to obtain the fluctuation value series of process execution time. Based on the pre-defined symbolic transfer entropy model, each fluctuation value in the fluctuation value sequence is symbolized in time series to obtain a symbolized fluctuation sequence. For each process pair in the process influence chain that has a direct dependency relationship, based on the symbolic fluctuation sequence of the preceding process and the symbolic fluctuation sequence of the following process, the symbolic transfer probability is calculated and substituted into the preset transfer entropy formula to calculate the symbolic transfer entropy from the preceding process to the following process. The total entropy of disturbance propagation is obtained by summing the symbolic propagation entropy of all process pairs with direct dependencies in the process influence chain. After each flexible scheduling plan is executed in the simulation, the voltage amplitude sequence of the grid nodes and the active power sequence of the grid lines are obtained after adjustment. The sum of the squares of the deviations of each data point from the rated voltage value in the voltage amplitude sequence is taken as the voltage stability entropy component; The variance of the ratio of each data point in the active power sequence to the transmission capacity limit of the line is used as the power flow load entropy component, and the variance of the sequence of the standard deviation of the active power values ​​of each critical line at the same time is used as the power flow equilibrium entropy component. The real-time safety entropy value is obtained by weighted summation of the voltage stability entropy component, the power flow load entropy component, and the power flow balance entropy component.

[0013] Furthermore, a Pareto mapping is performed based on the total entropy value of the disturbance propagation and the real-time safety entropy value to determine the final flexible scheduling plan, including: The total entropy value of disturbance propagation and the real-time safety entropy value corresponding to each flexible scheduling plan are used as a two-dimensional evaluation vector; A two-dimensional target space is constructed with the total entropy of disturbance propagation as the first dimension and the real-time security entropy as the second dimension, and all two-dimensional evaluation vectors are mapped to the two-dimensional target space respectively; wherein, each two-dimensional evaluation vector corresponds to a coordinate point in the two-dimensional target space; In the two-dimensional target space, identify all Pareto non-dominated coordinate points, construct the Pareto front solution set, and calculate the weighted Chebyshev distance from each coordinate point in the Pareto front solution set to the preset reference point. The flexible scheduling plan corresponding to the coordinate point with the smallest weighted Chebyshev distance is taken as the final flexible scheduling plan.

[0014] As an improvement to the above solution, another embodiment of the present invention provides a load scheduling device based on process dynamic analysis, comprising: The scheduling data acquisition module is used to acquire real-time telemetry data, production plan data, status time-series data of equipment IoT sensors, and real-time work order data. The load time series data transformation module is used to extract load time series data based on real-time telemetry data, production plan data and status time series data, and perform wavelet transform on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. The process constraint matching module is used to perform anchoring matching based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to the preset process constraint relationship, and obtain the matching result; The process influence chain generation module is used to identify the target process step from real-time work order data based on the matching results, calculate the influence intensity value to characterize the process interruption based on the process constraint relationship, the target process step and the matching results, and perform topology analysis based on the influence intensity value and the target process step to generate the process influence chain. The rolling search module for scheduling plans is used to calculate production risk indicators and production benefit indicators based on the influence intensity value in the process influence chain and the process constraint relationship. It then performs a rolling path search in the decision grid space composed of the production risk indicators and production benefit indicators to generate a set of flexible scheduling plans. The scheduling plan entropy calculation module is used to calculate the symbol transfer entropy for each flexible scheduling plan in the flexible scheduling plan set based on the fluctuation value sequence in the corresponding process influence chain, so as to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. The scheduling plan determination module is used to perform Pareto mapping based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

[0015] Another embodiment of the present invention provides a terminal device, 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 a load scheduling method based on process dynamic analysis as described in the above embodiments.

[0016] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a load scheduling method based on process dynamic analysis as described in the above embodiment.

[0017] By implementing this invention, at least the following beneficial effects are achieved: This invention provides a load scheduling method and apparatus based on dynamic process analysis. The method simultaneously acquires real-time telemetry data, production plan data, status time-series data from equipment IoT sensors, and real-time work order data, breaking the isolated processing mode of load data and production process data in existing technologies. This lays the data foundation for establishing the correlation between load fluctuations and the production process. Then, by using wavelet transform to separate the high-frequency transient components and low-frequency steady-state baseband components of the load time-series data, it can accurately capture the rapid load fluctuation characteristics caused by sudden events such as equipment start-up and shutdown, and process switching, solving the problem of existing technologies lacking dynamic perception capabilities for transient load fluctuations. Based on preset process constraints, this invention anchors and matches the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data, directly establishing a causal relationship between load fluctuation characteristics and production processes. This overcomes the bottleneck of existing technologies that struggle to build effective correlations between load fluctuations and production processes. By identifying target process processes, calculating the intensity of process interruption impacts, and generating process impact chains, it achieves dynamic analysis of the production source and impact range of load fluctuations, changing the current situation where existing technologies can only passively respond to load fluctuations. Based on the process influence chain, production risk and efficiency indicators are extracted, generating a set of flexible scheduling plans in the decision grid space. By calculating symbolic transfer entropy and assessing grid security entropy, combined with Pareto mapping, the optimal flexible scheduling plan is selected. This overcomes the limitations of existing technologies that rely on fixed thresholds or static scheduling, achieving dynamic optimization scheduling that balances production efficiency and grid security. It ensures targeted scheduling decisions through precise correlation and impact analysis, preventing a decline in production efficiency, while dynamically assessing grid voltage and active power sequences to ensure grid load balance, effectively reducing grid security risks and ultimately achieving synergistic protection of industrial production continuity and grid security and stability. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a load scheduling method based on process dynamic analysis according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a load scheduling device based on process dynamic analysis provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1To address the problem in existing technologies that struggle to establish precise causal relationships between load fluctuations and production processes, and the lack of dynamic analysis of load and production, leading to inaccurate scheduling responses, an embodiment of this invention provides a flowchart illustrating a load scheduling method based on process dynamic analysis, comprising: S1. Acquire real-time telemetry data, production plan data, status and timing data of equipment IoT sensors, and real-time work order data. Specifically, real-time telemetry data refers to the real-time operating data of power equipment acquired through SCADA (Supervisory Control and Data Acquisition) systems, including voltage, current, active power, and switch status. It is a core electrical quantity characterizing changes in grid load. Production planning data refers to the scheduling instructions issued by the production planning system, including process sequence, equipment allocation, planned start and stop times, and output targets, defining the preset logic of the production process. Equipment IoT sensor status time-series data refers to the equipment operating status data collected by intelligent sensors installed on production equipment and arranged in chronological order. Equipment IoT sensors, such as vibration, temperature, and speed sensors, reflect the actual operating conditions of the equipment. Real-time work order data refers to the work order execution status information collected in real-time through the production planning system API, including work order number, current process, actual start time, elapsed time, equipment status change signals, process code switching signals, abnormal flags, and completion progress, mapping the real-time execution status on the production floor.

[0021] To illustrate, data interface protocols such as OPC UA, Modbus TCP, or MQTT are used to connect to SCADA systems, production planning systems, and equipment IoT sensors to collect real-time telemetry data, production planning data, and status time-series data, forming a multi-source heterogeneous real-time dataset. At the same time, work order execution status information is collected through the RESTful API or database interface of the production planning system in a polling or subscription manner to obtain real-time work order data.

[0022] S2. Extract load time series data based on real-time telemetry data, production plan data and status time series data. Perform wavelet transform on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. Specifically, load time-series data refers to active power time-series data extracted from multi-source data, recording continuous load changes in chronological order, and is the core object of load analysis. High-frequency transient components are components in the load time-series data separated by wavelet transform that reflect rapid, short-term fluctuations caused by sudden events such as equipment start-up and shutdown, and process switching. Low-frequency steady-state baseband components are components in the load time-series data separated by wavelet transform that reflect the basic load level or planned, slow changes in the production process.

[0023] Schematic, firstly, timestamp alignment, format conversion, and physical dimension normalization are performed on multi-source heterogeneous data to generate a unified real-time status data stream. Active power time-series data is then extracted from this unified real-time status data stream as load time-series data. Multi-scale discrete wavelet transform is performed on the load time-series data to obtain approximation coefficients and detail coefficients at each scale. The approximation coefficients at the largest decomposition scale are reconstructed to obtain low-frequency steady-state baseband components, and the detail coefficients at other scales are reconstructed and superimposed to obtain high-frequency transient components. Based on a unified timestamp, the two types of components are combined to form load morphology data. By accurately characterizing the steady-state trend and transient fluctuations of the load, this addresses the problem that existing technologies cannot accurately capture sudden load changes.

[0024] Preferably, load time-series data is extracted based on real-time telemetry data, production plan data, and status time-series data, including: The first data stream is obtained by aligning the timestamps of each data item in the real-time telemetry data, production plan data, and status time sequence data. The first data stream is converted into a data format according to a preset communication protocol to obtain a protocol data stream. The protocol data stream is then aggregated and encapsulated based on a preset data topic to generate a real-time status data stream. Active power time-series data is parsed and extracted from the real-time status data stream and used as load time-series data.

[0025] Specifically, the first data stream is an intermediate data set synchronized with a time base after timestamp alignment of multi-source data. Communication protocols, such as OPC UA and Modbus TCP, define the syntax and semantic rules for data exchange between different systems, ensuring data format uniformity. The protocol data stream is a structurally sound and semantically consistent data set obtained after format conversion of the first data stream. Data topics are data categories categorized according to analytical needs, such as load data and equipment status, used for data aggregation and classification. The real-time status data stream is a set of self-describing data packets generated after aggregation and streaming encapsulation (such as Apache Avro encapsulation) of the protocol data stream based on data topics. Active power time-series data consists of time-series records of active power changes in electrical equipment, representing core data characterizing load demand.

[0026] Specifically, the first data stream is obtained by aligning the timestamps of each data item in real-time telemetry data, production plan data, and status time series data. Since data from different sources may generate timestamps based on their own independent clock sources, time base consistency must be ensured during cross-system fusion. The timestamp information of each data item is extracted and uniformly converted to Coordinated Universal Time (UTC) or a local high-precision time base. Time interpolation or nearest neighbor matching methods are used to synchronize and align the timestamps of all data items within the same time interval, ensuring accurate correspondence between power load data, production plan data, and status time series data at the same moment. More specifically, an intermediate data stream with a fully synchronized time base, i.e., the first data stream, is output, where each record carries a unified timestamp, laying the foundation for subsequent time series correlation analysis. Ensuring consistency in time, format, and semantics among data from power monitoring, production scheduling, and equipment sensing dimensions provides a high-quality, highly consistent input data foundation for subsequent load time series analysis, process correlation matching, and dynamic scheduling decisions.

[0027] Specifically, the first data stream is converted into a protocol data stream according to a preset communication protocol. This protocol data stream is then aggregated and encapsulated based on a preset data theme to generate a real-time status data stream. Data format conversion and physical dimension normalization are performed based on a predefined equipment information model and the communication protocol. In detail, the equipment information model refers to a structured description of various types of equipment in the production line and their electrical and mechanical attributes, including equipment identification, rated parameters, units of measurement, and data semantics.

[0028] Based on the dimension lookup table defined in the device information model, data values ​​from different systems in the first data stream are converted into unified physical units, such as power units being standardized to kilowatts and temperature units to degrees Celsius. Simultaneously, according to the data format agreed upon in the communication protocol, the original diverse data structures are converted into consistent key-value pairs or tabular forms, eliminating semantic ambiguity. More specifically, after the format and dimension unification processing, a standardized data stream that can be directly used for calculations is output.

[0029] The specification data stream undergoes aggregation and streaming encapsulation based on data themes. Specifically, a data theme refers to a data category categorized according to analytical needs, such as load data, equipment status, and planned instructions. Specifically, data records belonging to the same theme within the specification data stream are merged; for example, all records related to active power are aggregated into load time-series theme data blocks. Streaming data encapsulation technologies, such as Apache Avro or Protocol Buffers, are used to serialize and compress the aggregated data blocks, adding theme identifiers, time ranges, and data volume metadata to form self-describing data packets. This step integrates the originally scattered multi-source data into a unified real-time status data stream that is structurally clear, efficiently transmitted, and easily invoked by subsequent modules. This unified real-time status data stream carries continuous load changes and equipment operation information, while real-time work order data reflects the real-time status of the production process. The combination of these two allows load fluctuations to be interpreted within the specific context of production activities, supporting accurate load attribution and the construction of influence chains, ultimately serving the dynamic tracking and optimized scheduling of load demand.

[0030] Specifically, active power time-series data is parsed and extracted from the real-time status data stream as load time-series data. The real-time status data stream is a structured data sequence containing various thematic information, formed after the aforementioned data reduction processing. Active power time-series data specifically refers to the sequence of active power measurements arranged in chronological order from this data stream. It directly reflects the actual power demand drawn by electrical equipment or production units from the power grid and is the most core electrical quantity characterizing load demand. Performing the parsing operation means identifying and locating data packets belonging to the load time-series or active power theme based on predefined data theme identifiers or field labels in the data stream; performing the extraction operation means reading the power values ​​from these data packets along with their precise timestamps to form a clean load time-series data set containing only time and active power value pairs, providing accurate input for subsequent frequency domain analysis.

[0031] Preferably, wavelet transform is performed on the load time series data to separate the high-frequency transient components and the low-frequency steady-state baseband components, including: Based on the load time series data, a discrete wavelet transform with a preset decomposition scale is performed to obtain the approximate coefficients and detail coefficients of the load time series data at each decomposition scale. Wavelet reconstruction is performed on the approximation coefficients at the maximum decomposition scale to obtain the low-frequency steady-state baseband components of the load time series data. Wavelet reconstruction is performed on the detail coefficients at all decomposition scales except the maximum decomposition scale. All detail components obtained from the wavelet reconstruction are superimposed to obtain the high-frequency transient components of the load time series data.

[0032] Specifically, the preset decomposition scale is the maximum decomposition level set before the wavelet transform, such as level 5, determined based on the load fluctuation cycle and analysis accuracy. Discrete wavelet transform is a signal processing method that decomposes continuous load time-series data into discrete approximation coefficients and detail coefficients, suitable for non-stationary signal analysis. Approximation coefficients are coefficients in the wavelet transform that characterize the low-frequency trend information of the signal. Detail coefficients are coefficients in the wavelet transform that characterize the high-frequency details and abrupt changes of the signal. Wavelet reconstruction is the inverse transform process of restoring specific components of the signal based on wavelet coefficients.

[0033] Specifically, a discrete wavelet transform with a preset decomposition scale is performed on the load time series data to obtain the approximate coefficients and detail coefficients of the load time series data at each decomposition scale. Multi-scale discrete wavelet transform is a signal processing method whose core is to decompose a time series signal into a series of different resolutions or scales to simultaneously reveal the overall trend and local detailed features of the signal. Specifically, the load time series data is used as the input signal, and a wavelet basis function suitable for non-stationary signal analysis is selected, such as the Daubechies wavelet or Symlets wavelet.

[0034] Then, a multi-level decomposition operation is performed: the first-level decomposition breaks down the original load time-series data into first-level approximation coefficients and first-level detail coefficients; next, the first-level approximation coefficients are used as new input for the second-level decomposition to obtain second-level approximation coefficients and second-level detail coefficients; this process is iterated until the preset maximum decomposition scale is reached. More specifically, at each scale, the approximation coefficients characterize the general outline or low-frequency trend information of the signal at that scale, while the detail coefficients characterize the high-frequency details or abrupt changes of the signal at that scale. Through this transformation operation, the original load time-series data is converted into a set of multi-scale coefficient sequences, laying the foundation for separating load components of different frequency bands.

[0035] Specifically, wavelet reconstruction is performed on the approximate coefficients at the maximum decomposition scale to obtain the low-frequency steady-state baseband components of the load time series data. More specifically, the maximum decomposition scale refers to the pre-defined final decomposition level in the multi-scale discrete wavelet transform, its selection based on the main period of load fluctuations and the required level of analysis. Wavelet reconstruction refers to the process of restoring the original signal or its specific components using wavelet coefficients. From the coefficient set obtained after multi-scale decomposition, only the approximate coefficient sequence corresponding to the maximum decomposition scale is selected, while the approximate coefficients and detail coefficients at all other scales are set to zero. Then, the wavelet reconstruction algorithm is used to perform an inverse transform operation on this processed coefficient set. More specifically, this reconstruction operation essentially uses only the approximate information representing the coarsest-scale trend to restore the signal. Therefore, the output signal components filter out most of the short-term, rapid fluctuations, retaining the slowest and most stable long-term trend part of load changes, namely the low-frequency steady-state baseband components, which reflect the basic load level or planned slow changes in the production process.

[0036] Specifically, wavelet reconstruction is performed on the detail coefficients at all decomposition scales except the maximum decomposition scale. All detail components obtained from the wavelet reconstruction are then superimposed to obtain the high-frequency transient components of the load time series data. The other scales refer to all decomposition levels from the first layer to the maximum decomposition scale. Detail components refer to the signal components reconstructed from the detail coefficients at a single scale. Specifically, multiple independent single-scale reconstruction operations are performed: for each scale except the maximum decomposition scale, only the sequence of detail coefficients at that scale is retained, while all other coefficients are set to zero, and wavelet reconstruction is performed to obtain a signal component reflecting the high-frequency details at that specific scale.

[0037] Then, a linear superposition operation is performed: the detail components reconstructed from all the above scales are numerically accumulated according to their respective time points. More specifically, since the detail coefficients capture the rapid changes and abrupt changes in the signal at different levels of detail, the composite component formed after superposition embodies all the rapid fluctuations, spikes, and short-term fluctuations in the load sequence, i.e., high-frequency transient components, which are usually associated with random start-ups and shutdowns of equipment, instantaneous switching of processes, or sudden interference events.

[0038] S3. Based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data, anchor matching is performed according to the preset process constraint relationship to obtain the matching result; Specifically, the process constraints are a predefined knowledge base of physical and logical rules for the production line, including parameters such as process dependencies, equipment start-up and shutdown power logic, process flow sequence, allowable time tolerance, baseline influence coefficient, and time conversion coefficient. Anchoring matching refers to the process of aligning high-frequency transient components and low-frequency steady-state baseband components with real-time work order data through timestamps and logical verification to establish a causal relationship between load fluctuations and production processes.

[0039] Schematic, local extrema of the low-frequency steady-state baseband component are identified as power feature points, and the starting point of the high-frequency transient component exceeding a preset threshold is identified as a transient event feature point. Equipment status change signals and process code switching signals are analyzed from real-time work order data as work order event feature points. The three types of feature points are matched by timestamps, and an initial association is established if the absolute time difference is less than the allowable time tolerance of the corresponding process in the process constraint relationship. Consistency verification is performed based on the equipment start-stop power logic and the process flow sequence, and contradictory associations are deleted. The verified association points are encapsulated into a mapping relationship set as the matching result. By establishing a precise causal relationship between load fluctuations and production processes, the bottleneck of existing technologies in effectively linking load fluctuations with production processes is solved.

[0040] Preferably, anchoring and matching are performed based on the high-frequency transient component, the low-frequency steady-state baseband component, and real-time work order data according to a preset process constraint relationship to obtain the matching result, including: The local extreme points of the low-frequency steady-state baseband components are used as power feature points, and the starting point of the high-frequency transient components exceeding the preset component threshold is used as transient event feature points. Based on real-time work order data, extract equipment status change signals and process code switching signals, and use these signals as work order event feature points. The initial association is constructed by matching timestamps based on power feature points, transient event feature points, and work order event feature points; wherein, if the absolute time difference between any pair of feature points is less than the allowable time tolerance of the corresponding process in the preset process constraint relationship, the initial association is constructed. Based on the equipment start-stop power logic and process flow sequence in the process constraint relationship, the consistency of the feature point pairs that have been initially associated is checked, and the feature point pairs that contradict the equipment start-stop power logic or process flow sequence are deleted to obtain the verified associated point pairs. Each verified associated point pair is encapsulated as a mapping entry, all mapping entries are aggregated into a mapping relationship set, and the mapping relationship set is used as the matching result.

[0041] Specifically, power feature points are local maxima or minima in the low-frequency steady-state baseband components, marking the turning point in the steady-state load trend. Transient event feature points are the starting points in the high-frequency transient components that exceed a preset threshold, corresponding to the occurrence of sudden load fluctuations. Work order event feature points are event points extracted from real-time work order data that characterize changes in production status, such as equipment status changes and process code switching. Allowable time tolerance is the maximum time deviation between load feature points and work order event points preset in the process constraints. Consistency verification represents the process of verifying the validity of the initial association based on the equipment start / stop power logic and process flow sequence in the process constraints. The mapping relationship set is a collection of all verified feature point pairs and their associated information, serving as the final result of anchor matching.

[0042] Schematic illustration: Based on a unified timestamp, low-frequency steady-state baseband components and high-frequency transient components are time-aligned and combined to form load pattern data. In detail, the unified timestamp refers to the time coordinate axis inherited from the original load time series data and remaining unchanged throughout the wavelet transform and reconstruction process. Specifically, performing time alignment ensures that each data point in the low-frequency steady-state baseband component sequence and the high-frequency transient component sequence strictly corresponds to the same original time point, checking and eliminating minor time offsets that may be introduced by data processing. Performing data combination pairs the low-frequency steady-state baseband component value and the high-frequency transient component value at the same time point, forming a new data record containing two dimensions. More specifically, organizing the aligned and combined data in chronological order, the final output load pattern data is a multi-dimensional time series data representation. The state at each moment is described by a steady-state baseband value and a transient fluctuation value, thus achieving a complete and structured expression of the original load demand from long-term trends to instantaneous details, providing a refined data foundation for subsequently accurately anchoring load changes to specific production stages.

[0043] Specifically, power characteristic points and transient event characteristic points are identified in the load pattern data. The load pattern data is time-series data composed of low-frequency steady-state baseband components and high-frequency transient components. The low-frequency steady-state baseband components are scanned to find all local maxima and minima in the sequence. These points mark trend reversals during stable load changes, such as stepwise changes in base power levels caused by planned production phase transitions. These points are collectively referred to as power characteristic points.

[0044] Simultaneously, high-frequency transient components are scanned. When the absolute value of a component exceeds a threshold pre-set based on historical data or equipment characteristics, this point of excess is recorded as the starting point. These starting points correspond to rapid and severe pulse-like fluctuations in the load, typically triggered by events such as direct motor starting or instantaneous switching of high-power equipment. These starting points are collectively referred to as transient event characteristic points. More specifically, the output consists of two sets of characteristic points: power characteristic points and transient event characteristic points, which respectively characterize significant events on the load curve from the perspectives of trend and detail.

[0045] Specifically, equipment status change signals and process code switching signals are parsed from real-time work order data as work order event feature points. In detail, real-time work order data is a structured information stream continuously acquired from the production planning system interface, reflecting the real-time execution status of production tasks. Specifically, performing the parsing operation involves traversing each record in the real-time work order data, identifying and extracting event signals representing changes in the physical equipment status, such as status change signals from stopped to running, or from standby to processing.

[0046] Simultaneously, the system identifies and extracts coded switching signals representing the transition from one process step to the next in the production process. More specifically, each extracted equipment status change signal and process code switching signal, along with its precise timestamp, is defined as a work order event point. These points directly map to the specific operations and process evolution on the production floor.

[0047] Specifically, power characteristic points, transient event characteristic points, and work order event characteristic points are timestamped and initial associations are established. More specifically, timestamping is an alignment operation based on the absolute time information carried by all characteristic points and event points. Specifically, from the power characteristic point set, transient event characteristic point set, and work order event point set, one point from the load side and one point from the work order side are arbitrarily selected, and the absolute difference between their timestamps is calculated. This difference is compared with the maximum allowable time deviation (i.e., allowable time tolerance) for the corresponding process step in the predefined process constraints. If the calculated absolute time difference is less than the allowable time tolerance, the two points are considered to be temporally adjacent and likely caused by the same production activity, thus establishing an initial association between them. More specifically, this step, based on the principle of temporal proximity, initially links load fluctuation events with production execution events, forming a series of initial association pairs that have not yet been logically verified.

[0048] Specifically, based on the equipment start-stop power logic and process flow sequence in the process constraints, consistency checks are performed on the initially associated point pairs. In detail, the process constraints are a predefined knowledge base that encodes the physical and logical rules of the production line. Specifically, the equipment start-stop power logic defines the typical pattern of power change that a particular piece of equipment should exhibit when starting or stopping; for example, the start-up of a certain type of motor should be accompanied by a current surge lasting several seconds, manifesting as a positive transient pulse in the load pattern. The process flow sequence defines the strict sequential dependencies between various production processes. Performing the consistency check operation means that for each initially associated point pair, the associated load characteristic type and work order event type are checked to see if they conform to the power change logic specified for that type of event in the process constraints; simultaneously, the occurrence of the work order event is checked to ensure logical coherence with its preceding and following work order events in the process flow sequence. More specifically, this check process utilizes domain knowledge to filter the initial time matching results, eliminating associations that are impossible in terms of physical mechanisms or production logic.

[0049] Simultaneously, pairs of related points that contradict the logic of equipment start-up and shutdown power or the sequence of work processes are deleted. Specifically, after performing consistency checks, if the causal relationship implied by an initial pair of related points is found to contradict the explicitly defined logic of equipment start-up and shutdown power in the process constraints—for example, a work order event feature point identifying equipment startup is associated with a power feature point indicating a sudden load drop—then the association is determined to be contradictory. Similarly, if an association causes a time reversal or logical break in the sequence of work processes, it is also determined to be contradictory. More specifically, a deletion operation is performed to permanently remove these marked contradictory pairs of related points from the current association set, ensuring that the remaining associations all satisfy the basic physical laws and scheduling logic of the production process.

[0050] Specifically, all verified associated point pairs are constructed into a mapping relationship set, which serves as the matching result. In detail, verified associated point pairs refer to those that successfully pass timestamp matching and show no contradictions after consistency checks. Specifically, the construction operation involves encapsulating each valid associated point pair into a structured mapping entry. This entry must at least include the type, timestamp, and feature value of the load-side feature point, and the type, timestamp, and event description of the work order-side event point. All such entries are then aggregated to form a complete mapping relationship set.

[0051] More specifically, this mapping set constitutes a reliable correspondence network between dynamic fluctuations in load demand and specific production execution links, which has been doubly verified. The matching result not only indicates the time when the load change occurs, but more importantly, it accurately locates the specific production source that caused the change, providing a direct causal input for subsequent calculation of the process impact chain.

[0052] S4. Identify the target process step from the real-time work order data based on the matching results. Calculate the impact intensity value to characterize the process interruption based on the process constraint relationship, the target process step, and the matching results. Perform topology analysis based on the impact intensity value and the target process step to generate the process impact chain. Specifically, the target process step refers to the process step corresponding to the work order event feature point successfully associated with load fluctuation in the matching results; it is the root production step that causes load fluctuation. The impact intensity value represents the quantitative value of the degree of impact of process interruption on subsequent processes, combining the inherent impact potential of the process with the actual load fluctuation intensity. The process impact chain is a chain-like or network structure formed by the target process step and subsequent processes whose cumulative impact intensity value exceeds a preset threshold, organized according to the dependency relationship topology, visually presenting the disturbance propagation path and impact range.

[0053] Schematic, the process involves identifying work order event feature points associated with load data from the matching results; the corresponding processes are then identified as the target process steps. Direct and indirect subsequent processes of the target process step are located, forming a complete set of affected processes. The amplitude of transient event feature points associated with the work order event feature points is obtained and multiplied by the baseline impact coefficient of the target process step to obtain the initial interruption impact intensity value. Impact propagation weighting factors are assigned to all affected processes, and propagation attenuation factors are assigned to dependency paths. Starting from the initial intensity value, the process propagates along the dependency path, summing the attenuated intensity values ​​of preceding processes to obtain the cumulative impact intensity value of each process. Processes with cumulative intensity values ​​exceeding a preset intensity threshold are selected, and process impact chains are generated based on dependency topology. By realizing production traceability and impact range analysis of load fluctuations, the current passive response approach of existing technologies is changed.

[0054] Specifically, based on the matching results, work order event feature points that are successfully associated with load pattern data are identified from the work order event points in the real-time work order data. Specifically, the matching results are a set of mapping relationships generated by the aforementioned anchoring matching step, which clearly records which load feature points have established verified associations with which work order event points. Performing the identification operation involves traversing this mapping relationship set and extracting all work order event point identifiers that are associated parties. Then, based on these identifiers, the complete information of these work order event points is located and read from the real-time work order data stream. More specifically, the output of this step is a filtered subset of work order event points that have been confirmed to have a causal relationship with observable load fluctuations; these event points constitute the direct input and starting point of fact for subsequent impact analysis.

[0055] The process steps corresponding to the identified work order event points are determined as active process steps, i.e., target process steps. Specifically, each work order event feature point contains the code or name of its corresponding specific process step within its data structure. The determination operation involves parsing the selected work order event feature point information to obtain its embedded process identifier. This process identifier points to a specific, executable step unit in the production process, such as raw material mixing, CNC machine tool processing, or finished product packaging. More specifically, since the occurrence of these process steps is directly confirmed by characteristic changes in the load data, they are defined as target process steps, meaning they are the initial origin or root cause of the currently observed load demand fluctuations in the production process.

[0056] Preferably, the impact intensity value for characterizing process interruption is calculated based on process constraints, target process steps, and matching results. Then, topology analysis is performed based on the impact intensity value and the target process steps to generate a process influence chain, including: Find all subsequent processes that have the target process as a direct predecessor from the process constraints, and construct a set of direct subsequent processes; Based on the dependencies between processes in the process constraints, starting from the set of directly dependent processes, iteratively query all indirectly dependent processes to construct the complete set of affected processes; Based on the matching results, determine the amplitude corresponding to the transient event feature point associated with the work order event feature point, and take the transient event feature point in the mapping relationship set corresponding to the amplitude as the load transient feature point; The baseline influence coefficient of the target process step in the process constraint relationship is multiplied with the amplitude of the load transient characteristic point to obtain the influence intensity value used to characterize the process interruption. Starting with the impact intensity value, based on the dependencies between processes, the impact intensity propagation calculation is performed on each process in the set of all affected processes. The dependency path is multiplied by the corresponding propagation attenuation factor. For each process in the set of all affected processes, the impact intensity value of all preceding processes of the current process is attenuated according to the impact propagation weight factor and then summed to obtain the cumulative impact intensity value of the current process. The impact propagation weight factor is preset according to the urgency level of the process, and the propagation attenuation factor is preset according to the dependency path between processes. The target process step and subsequent processes whose cumulative influence intensity value exceeds the preset intensity threshold are organized into a topological structure according to their dependencies to generate a process influence chain.

[0057] Specifically, the set of directly following processes is the set of all processes that have the target process as their direct predecessor and have no intermediate processes. The set of all processes affected is the collective term for the directly following processes and all indirectly following processes of the target process. The load transient characteristic point is a transient event characteristic point that is successfully associated with the work order event characteristic point, and its amplitude represents the intensity of the load change. The baseline influence coefficient is a parameter preset in the process constraint relationship that represents the inherent influence potential of each process on the subsequent process, with a value of 0-1, and the larger the value, the stronger the influence. The influence transmission weight factor is a parameter allocated according to the urgency level of the process production (urgent, general, lenient), used to adjust the transmission ratio of the influence intensity. The transmission attenuation factor is a parameter preset in the process constraint relationship that represents the degree of natural dissipation of the influence intensity on the dependent path, with a value of 0-1, and the longer the path, the greater the attenuation. The cumulative influence intensity value is the sum of the influence intensity received by a certain process through all preceding dependent paths, after attenuation and weight adjustment.

[0058] Specifically, the process constraint relationships are used to identify all subsequent processes that have the target process step as their direct predecessor, forming a set of directly preceding processes. In detail, the process constraint relationships are a predefined knowledge model that encodes the dependencies and sequences between all processes in the production line in the form of a diagram or table. Specifically, performing a lookup operation means searching this knowledge model, using the identified target process step as the query key, for all process nodes that list it as their immediate predecessor. These found process nodes are logically immediately following the target step, and their initiation or execution typically depends on the completion of the initiating process. More specifically, all such process nodes found are aggregated to form a set of directly preceding processes, representing the first wave of processes affected by the root cause disturbance.

[0059] Specifically, based on the dependencies between processes within the process constraints, starting from the set of directly dependent processes, all indirectly dependent processes are iteratively searched to form the complete set of affected processes. More specifically, the dependencies between processes include not only direct sequential connections but also indirect dependencies spanning multiple stages; for example, a process may depend on the predecessor of its predecessor process. Performing the iterative search operation means first using the set of directly dependent processes as the current search basis; then, for each process in this set, its respective directly dependent processes are searched again within the process constraints; these newly found processes are added to the set, and this process is repeated until no new dependent processes can be found in the dependency network. More specifically, the final complete set of affected processes includes every downstream process reachable from the target process stage along all possible dependent paths, fully characterizing the potential scope of the impact.

[0060] Specifically, based on the matching results, the amplitude recorded by the transient event feature point successfully associated with the work order event feature point is obtained. More specifically, in the matching results, a successful association not only establishes the temporal and logical correspondence between the work order event feature point and the load transient feature point, but also records the specific attributes of the load-side feature point. Specifically, performing the acquisition operation means, for the work order event feature point corresponding to the identified target process step, finding the paired transient event feature point in the mapping relationship set, and reading its amplitude attribute from the data of that feature point. This amplitude characterizes the specific intensity of the load transient at the corresponding moment, such as the magnitude of a sudden increase or decrease in power. More specifically, this amplitude, as a key physical quantity for quantifying the initial intensity of the disturbance, provides an objective measurement basis for subsequent calculations of the degree of impact.

[0061] Specifically, the product of the baseline influence coefficient of the target process step in the process constraint relationship and the amplitude of the load transient characteristic point is used as the influence intensity value of the target process step to characterize the process interruption. More specifically, the baseline influence coefficient is a predefined parameter for each process step in the process constraint relationship knowledge base. It reflects the inherent potential or sensitivity of the process itself to the stability of the production process and is usually set based on factors such as process type and resource utilization. Performing the calculation operation means first retrieving the baseline influence coefficient of the target process step from the process constraint relationship, and then multiplying it by the amplitude of the load transient characteristic point obtained in the previous step. More specifically, this product result is the initial influence intensity value, which integrates the inherent attributes of the root process and the intensity of the current actual disturbance, constituting the initial energy value for calculating the propagation of the influence in the process chain.

[0062] Specifically, based on process constraints, an impact propagation weighting factor is assigned to each process in the set of all affected processes according to its production urgency level. In detail, the production urgency level is a graded evaluation of the importance or time pressure of a process within the process constraints, such as urgent, moderate, or lenient. Specifically, performing the assignment operation means that for each process in the set of all affected processes, an impact propagation weighting factor is retrieved from a predefined mapping table or calculated using rules, based on its urgency level. This factor is a numerical value used to adjust the proportion by which the intensity of the impact transmitted to that process should be amplified or reduced during the impact propagation process; processes with higher urgency are typically assigned a larger weighting factor. More specifically, this step ensures that the impact propagation model can reflect the differences in importance of different processes in responding to disturbances, making the impact assessment more practically instructive.

[0063] Starting with an initial impact strength value, and based on dependencies, the impact strength propagation calculation is performed on all affected processes in the entire set of processes. Specifically, dependencies refer to the sequential links between processes defined in the process constraints. Performing the propagation calculation operation means, following the direction of the dependency network, recursively distributing and calculating the initial impact strength value from the source node (the target process step) along each dependency edge to subsequent process nodes. More specifically, this process simulates how an anomaly or interruption in one stage of the production process, and the resulting shockwaves, affect the dynamics of downstream processes sequentially through logical and temporal links between processes.

[0064] The process multiplies the traversed dependency paths by the corresponding transmission attenuation factor. For each process in the entire set of affected processes, the attenuated impact strength values ​​of all its preceding processes are summed to obtain the cumulative impact strength value for each process. Specifically, the transmission attenuation factor is a coefficient less than or equal to one defined for each type of dependency or path length in the process constraints, used to characterize the natural dissipation or weakening effect of the impact during transmission. Performing the calculation and summation involves two levels: First, when an impact value is transmitted from a parent process to a child process along a specific dependency path, the transmitted value is multiplied by the transmission attenuation factor corresponding to that path to obtain the attenuated value received by the child process. Second, for a specific child process, it may receive impact values ​​from multiple different preceding processes; these attenuated impact values ​​from different paths need to be arithmetically summed. More specifically, a cumulative impact strength value is finally calculated for each process in the entire set of affected processes, quantifying the overall impact pressure borne by that process due to the source disturbance.

[0065] Specifically, the target process step and all subsequent processes with a cumulative impact intensity value exceeding a preset threshold are organized according to a dependency topology to generate a process impact chain. In detail, the preset threshold is a numerical threshold set based on production tolerance or management needs, used to screen out critically affected processes worthy of attention. The organization and generation process involves first using the target process step as the starting node of the process impact chain; then, selecting all processes from the entire set of affected processes whose cumulative impact intensity value exceeds the preset threshold; finally, connecting these processes with directed edges based on the actual dependencies in the process constraints, forming a chain-like or network-like topology graph starting from the active process step and extending to the critically affected process steps. More specifically, the final generated process impact chain is a structured knowledge representation that intuitively reveals the production root causes of load fluctuations, the main impact paths, and the ranking of the degree of impact on each step, providing a clear decision-making object and scope for formulating precise scheduling intervention measures.

[0066] S5. Based on the influence intensity value in the process influence chain and the process constraint relationship, calculate the production risk index and production benefit index, and perform rolling path search in the decision grid space composed of the production risk index and production benefit index to generate a flexible scheduling plan set. Specifically, production risk indicators quantify parameters that cause production to deviate from expectations due to load fluctuations, including process delay risk indicators (schedule risk) and quality deviation risk indicators (quality risk). Production efficiency indicators quantify the efficiency of production resource utilization and economic benefits, including capacity utilization rate indicators (resource efficiency) and expected return indicators (economic benefits). The decision grid space represents a two-dimensional space constructed with production risk indicators as the first axis and production efficiency indicators as the second axis, used to visualize the relationship between production status and scheduling objectives. Rolling path search refers to the process of iteratively exploring and optimizing paths in the decision grid space, starting from the current production status, using a cluster search algorithm to generate a scheduling strategy adapted to the current status. The flexible scheduling plan set is a collection containing multiple executable scheduling schemes. Each scheme adapts to on-site uncertainties by adjusting process start and stop times, production order, and resource allocation parameters, possessing flexible adaptability.

[0067] Schematic diagram: Production risk indicators are calculated based on the process influence chain: Process delay risk indicator = Delayed period / Standard working hours; Quality deviation risk indicator = Quality sensitivity coefficient × Cumulative impact intensity value; Production efficiency indicators: Capacity utilization rate indicator = Total estimated working hours / Planned total working hours; Expected revenue indicator = (Output unit price × preset output quantity); a two-dimensional decision grid space is constructed with the weighted sum of risk indicators as the X-axis and the weighted sum of benefit indicators as the Y-axis to determine the initial point and target area; a cluster search algorithm is adopted, setting the maximum iteration step size L and the number of paths retained in each generation K, using the production risk indicator / production benefit indicator as the substitution value, K candidate paths are obtained, and the M paths with the smallest substitution value are selected as the preferred paths; based on the preferred paths, the start and stop times of processes, production order, and resource allocation parameters are adjusted to generate a set of flexible scheduling plans. By generating flexible plans that take into account multiple objectives, the problem of existing technical scheduling plans lacking multi-objective consideration is solved.

[0068] Preferably, the production risk indicators include process delay risk indicators and quality deviation risk indicators; the production efficiency indicators include capacity utilization rate indicators and expected return indicators. Based on the influence intensity values ​​in the process influence chain and the process constraint relationships, production risk indicators and production benefit indicators are calculated, including: The cumulative influence intensity value of each process is determined based on the influence intensity value in the process influence chain; Multiply the cumulative impact strength value and the time conversion coefficient in the process constraint relationship to obtain the delay time of each process. The ratio of the delay time of each process to the preset standard working hours is used as the process delay risk indicator, and the product of the process quality sensitivity coefficient and the cumulative impact intensity value in the process constraint relationship is used as the quality deviation risk indicator. Add the standard working hours of each process to the delay time to obtain the estimated working hours of each process; The ratio of estimated working hours to the preset total planned working hours is used as the capacity utilization rate indicator. The expected revenue indicator is obtained by summing the product of the output unit price of each process and the preset output quantity.

[0069] Specifically, the process delay risk index characterizes the relative degree of delay caused by disturbances in a process, and is the ratio of the delayed time to the standard working hours. The quality deviation risk index characterizes the potential risk of deterioration in the quality of process output due to disturbances, and is the product of the process quality sensitivity coefficient and the cumulative impact intensity value. The capacity utilization rate index characterizes the saturation level of production resources (time, equipment), and is the ratio of the total estimated working hours to the total planned working hours. The expected return index characterizes the total economic return that the production process can achieve after being affected by disturbances, and is the sum of the products of the unit price of each process output and the estimated completed quantity. The time conversion coefficient is a parameter (unit: h / kW) preset in the process constraint relationship that converts the abstract impact intensity value into the specific delay time. The estimated working hours are the sum of the standard working hours of the process and the delayed time, reflecting the actual time required after the process is disturbed.

[0070] Specifically, the cumulative impact intensity value of each process in the process influence chain is multiplied by the time conversion coefficient in the process constraint relationship to obtain the delay time for each process. Specifically, the cumulative impact intensity value is a quantitative value calculated for each process in the process influence chain, characterizing the degree of impact from the source disturbance. The time conversion coefficient is a parameter preset for each type of process in the process constraint relationship knowledge base. Its function is to convert the abstract impact intensity value into a specific time delay. This coefficient is usually derived based on the historical average processing time of the process and disturbance sensitivity analysis.

[0071] Performing a multiplication operation refers to multiplying the cumulative impact intensity value of each process node in the process impact chain with its corresponding time conversion coefficient. More specifically, the result is a delay time for each process, which predicts the additional time that the process may take compared to the original plan due to the current disturbance, providing basic time-dimensional data for subsequent risk assessment.

[0072] Specifically, the ratio of the delay time of each process to its predefined standard working hours is calculated as a process delay risk indicator. More specifically, the predefined standard working hours refer to the rated completion time for that process under undisturbed conditions, set in the process constraints or production plan. Specifically, calculating the ratio involves dividing the delay time of a particular process obtained in the previous step by the corresponding standard working hours to obtain a proportional value. More specifically, this proportional value is the process delay risk indicator. It eliminates the influence of differences in the duration of different processes and directly reflects the severity of the planned delay relative to its normal cycle time; the higher the value, the higher the risk of on-time completion of the process.

[0073] Simultaneously, the product of the process quality sensitivity coefficient and the cumulative influence intensity value in the process constraint relationship is used as a quality deviation risk indicator. Specifically, the process quality sensitivity coefficient is another parameter defined for each process in the process constraint relationship, used to characterize the sensitivity of the output quality of that process to fluctuations in the production process. For example, precision machining processes typically have a high quality sensitivity coefficient.

[0074] Performing a product operation refers to multiplying the quality sensitivity coefficient of each process in the process influence chain by its cumulative influence strength value. More specifically, the result is a quality deviation risk index, which combines the inherent quality vulnerability of the process with the actual intensity of the disturbance it is subjected to, and is used to quantify and predict the potential risk of quality degradation in the output of that process due to the propagation of the current influence chain.

[0075] Specifically, the standard working hours for each process are added to the delayed time to obtain the estimated working hours for that process. Specifically, the estimated working hours refer to the total time actually required for that process, re-predicted after considering the impact of the current disturbance. Performing the addition operation involves arithmetically summing the standard working hours reflecting the normal demand of the process with the delayed time reflecting the abnormal increase. More specifically, this step generates a revised time estimate for each process, which will serve as the basis for assessing the impact on resource utilization efficiency and overall production schedule.

[0076] The ratio of estimated working hours to total planned working hours is used as a capacity utilization indicator. Specifically, total planned working hours refer to the sum of standard working hours required for all processes according to the original schedule within the current production task or assessment period, representing the theoretical total capacity load. To calculate this ratio, the estimated working hours of all processes in the process influence chain are first summed, and then this sum is divided by the corresponding total planned working hours. More specifically, the resulting ratio is the capacity utilization indicator, reflecting the saturation level of overall production resources under the current disturbance. A ratio that is too high may indicate resource shortages and production bottlenecks, while a ratio that is too low may indicate idle resources.

[0077] The expected revenue index is obtained by summing the product of the output unit price and the estimated output quantity for each process. Specifically, the output unit price is the economic value defined for the semi-finished or finished products produced at each process stage within the process constraints or production plan. The preset output quantity, or estimated output quantity, is the output that can be completed at that stage, re-predicted based on corrected production parameters after considering the impact of current disturbances.

[0078] Performing the product and summation operations involves two levels: First, for each process, the product of its output unit price and the estimated completed quantity is calculated to obtain the expected contribution revenue of that process; then, the contribution revenues of all processes in the process influence chain are summed up. More specifically, the summation is the expected revenue indicator, which quantifies the overall economic benefits that the entire related production process may achieve under the current influence chain scenario from the perspective of economic output.

[0079] Preferably, a rolling path search is performed in the decision grid space composed of production risk indicators and production efficiency indicators to generate a set of flexible scheduling plans, including: A two-dimensional decision grid space is constructed using the weighted sum of process delay risk indicators and quality deviation risk indicators as the first coordinate axis, and the weighted sum of capacity utilization rate indicators and expected return indicators as the second coordinate axis. Based on the process delay risk index, quality deviation risk index, capacity utilization rate index, and expected return index under the current production status, calculate the coordinates of the current status in the two-dimensional decision grid space and use them as the initial point. Based on the production time and output constraints in the production plan data, the target area is determined in the two-dimensional decision grid space. In the two-dimensional decision grid space, starting from the initial point and ending at the target area, a path search is performed according to the preset cluster search method to obtain several candidate decision paths and their corresponding costs. Candidate decision paths with a cost value less than a preset cost threshold are selected as the preferred decision paths. Based on the coordinates of the decision state point of each preferred decision path, the start and stop times, production order, and resource allocation parameters of the corresponding processes in the process influence chain are adjusted to generate a flexible scheduling plan. The flexible scheduling plans of all preferred decision paths are then integrated to obtain a flexible scheduling plan set.

[0080] Specifically, the two-dimensional decision grid space is a two-dimensional space constructed with production risk indicators as the X-axis and production efficiency indicators as the Y-axis, used to visualize production status and scheduling objectives. The initial point is the coordinate of the current production status in the two-dimensional decision grid space, determined by the current comprehensive risk and efficiency indicators. The target region is the ideal state range defined in the two-dimensional decision grid space based on the time and output constraints of the production plan. The bundle search method searches for an optimal path from the initial point to the target region in the decision space by enumerating feasible states, calculating cost values, screening optimal paths, and iteratively expanding; this is the bundle search algorithm. Candidate decision paths are feasible paths extending from the initial point to the target region generated in each iteration, with each path corresponding to a set of production parameter adjustment strategies. The cost value is a quantitative indicator for evaluating the merits of candidate paths, calculated as production risk indicator ÷ production efficiency indicator. The preferred decision path is the path with the optimal cost value selected from the candidate paths.

[0081] Specifically, a two-dimensional decision-making grid space is constructed using the weighted sum of process delay risk and quality deviation risk indicators as the first coordinate axis, and the weighted sum of capacity utilization and expected return indicators as the second coordinate axis. In detail, the weighted summation involves assigning a weight coefficient to each indicator, reflecting the decision-maker's level of importance to that indicator, and then calculating the sum of the products of each indicator value and its corresponding weight. The construction operation defines the weighted sum of the process delay risk and quality deviation risk indicators as a comprehensive production risk dimension, serving as the horizontal axis in the two-dimensional space; and defines the weighted sum of the capacity utilization and expected return indicators as a comprehensive production benefit dimension, serving as the vertical axis in the two-dimensional space. More specifically, each point in the resulting two-dimensional decision-making grid space uniquely corresponds to a specific comprehensive state of production risk and benefit, providing a unified mathematical framework for weighing and optimizing multiple objectives.

[0082] Specifically, based on the process delay risk indicators, quality deviation risk indicators, capacity utilization rate indicators, and expected benefit indicators under the current production status, their coordinates in the two-dimensional decision grid space are calculated as the initial point. Specifically, the current production status refers to the values ​​of each indicator directly derived from the process influence chain without any scheduling adjustments. Performing the coordinate calculation operation means weighting and summing these real-time calculated indicator values ​​to obtain their scalar values ​​in the risk and benefit dimensions, thereby determining a unique coordinate position in the two-dimensional decision grid space. More specifically, this coordinate point is the initial point, which accurately depicts the true risk and benefit situation faced by the production system after the current disturbance occurs and before scheduling intervention, and serves as the starting point for subsequent optimization searches.

[0083] Simultaneously, based on the production time and output constraints of the production planning data in the production planning system, a target region is defined in the two-dimensional decision grid space. Specifically, the time and output constraints originate from higher-level production management objectives, such as requiring total delays not to exceed a certain upper limit or total output to reach a certain lower limit. Specifically, performing the definition operation means transforming these management objectives into value ranges for the production risk and production benefit dimensions in the two-dimensional decision grid space. For example, the risk dimension value must be below a certain threshold, and the benefit dimension value must be above a certain threshold. These thresholds enclose a region in the space. More specifically, the defined target region represents the ideal set of states that the scheduling decision needs to achieve or approach, conforming to the production plan requirements, providing a clear target orientation for the search algorithm.

[0084] Specifically, in the two-dimensional collaborative decision-making space, starting from the current state point and ending at entering the target region, a beam search algorithm is used to search for paths, resulting in K candidate decision paths. More specifically, the beam search algorithm is a heuristic graph search algorithm that retains only a few of the most promising paths in each generation of the search to avoid combinatorial explosion.

[0085] Specifically, the path search operation involves discretizing the two-dimensional decision grid space into state points, each representing a risk-benefit state corresponding to a possible production parameter setting. Starting from the initial point representing the current state, the algorithm enumerates all feasible adjustment actions in the next decision cycle (e.g., within the next few minutes or hours). Each action moves the current state point to a new state point, thus forming a path. After each generation of expansion, the algorithm scores all new paths according to a preset evaluation function, retaining only the K paths with the best scores for the next round of expansion. This process iterates until a path's terminal state point enters the target region or the maximum number of search steps is reached. More specifically, the final output of K candidate decision paths represents multiple optimized trajectories from the initial state to the target region. Each path consists of a series of consecutive state points, representing a step-by-step strategy sequence for adjusting production parameters.

[0086] Specifically, from the K candidate decision paths, the M paths with the lowest substitution values ​​are selected as the preferred decision paths. In detail, the substitution value is the evaluation score calculated for each path during the beam search process; generally, the better the overall performance of the path's terminal state and the smoother the path adjustment, the lower the substitution value. Specifically, the selection operation means that from the set of K candidate decision paths obtained at the end of the algorithm, the paths are sorted in ascending order according to their substitution values, and then the top M paths are selected. More specifically, these M preferred decision paths are the few solutions with the best overall performance among all feasible solutions obtained in the search; they balance the speed of convergence to the target region with the robustness of the path.

[0087] Specifically, based on the coordinates of the decision state points of each optimal decision path, the start-up and shutdown times, production sequences, and resource allocation parameters of relevant processes in the process influence chain are adjusted to generate flexible scheduling plans. More specifically, the coordinates of each state point on the decision path map back to a set of specific production risk and benefit index values, which in turn originate from a series of specific process-level parameters. Specifically, performing the adjustment and generation operations means, for an optimal decision path, reverse-analyzing the implicit process delay risks, quality risks, capacity utilization rates, and expected return levels of its key state points (especially the endpoint).

[0088] To achieve these target levels, production parameters need to be reset for the relevant processes in the process impact chain that caused the current disturbance and those affected. This includes adjusting the planned start and end times of these processes, changing the execution order of several processes, or reallocating resources such as equipment and manpower. These specific adjustment instructions are then organized into a complete and executable set of production scheduling instructions, thus generating a flexible scheduling plan. More specifically, the core characteristic of this plan is flexibility; it is not a rigid solution but allows for adjustments within a certain parameter range to adapt to uncertainties in the production environment.

[0089] Simultaneously, the M flexible scheduling plans corresponding to all M preferred decision paths are combined to form a flexible scheduling plan set. Specifically, the composition operation refers to collecting and summarizing the flexible scheduling plan generated based on each preferred decision path. More specifically, the final flexible scheduling plan set is a collection of solutions containing various optimization approaches and adjustment strategies. These solutions were all generated within the framework of preliminary indicator extraction and multi-objective search, collectively providing a diverse and selectable decision-making pool to address current load disturbances and process impacts, laying a rich foundation for subsequent entropy value evaluation and final solution selection.

[0090] In a preferred embodiment of the present invention, in a two-dimensional decision grid space, starting from an initial point and ending at entering the target region, a path search is performed according to a preset cluster search method to obtain several candidate decision paths, including: Starting from the current state point, the initial candidate path set is defined as the path containing only this starting point; and the maximum iteration step size L and the number of paths retained in each generation K are set. For each path in the candidate path set, enumerate all feasible next state points of its terminal state point in the next decision cycle; Each feasible next state point obtained by enumeration is appended to the end of its corresponding original path to form a new candidate path; For each new candidate path, the comprehensive production risk index corresponding to the end state point of the path is used as the dividend, and the comprehensive production benefit index is used as the divisor. The resulting quotient is determined as the evaluation cost of the path. All new candidate paths are sorted in ascending order based on their evaluated cost value, and the top K candidate paths are selected and retained. The K retained paths are used as a new set of candidate paths, and the current iteration step number is recorded; Determine whether the current iteration step has reached L, and whether there is an end state point in the new candidate path set that has entered the target region; If not, continue enumerating; if yes, output the current set of candidate paths as the final K candidate decision paths to be retained.

[0091] Specifically, the current state point is used as the search starting point. This current state point is a coordinate point in the two-dimensional decision space that represents the production situation before adjustment, calculated based on real-time production risk and benefit indicators. The initialization operation involves creating a path that contains only the coordinates of this starting point and setting this single path as the initial set of candidate paths.

[0092] Simultaneously, two key control parameters are set: the maximum iteration step size L and the number of paths retained per generation K. The maximum iteration step size L defines the maximum number of steps or decision cycles the algorithm is allowed to explore forward, preventing the search process from going indefinitely; the number of paths retained per generation K defines the number of optimal paths allowed to be retained after each round of path expansion, controlling the search width and computational complexity. More specifically, this step completes the preparatory work before the algorithm runs, clarifying the initial state of the search and the resource boundaries.

[0093] Specifically, for each path in the candidate path set, all feasible next state points for its terminal state point within the next decision cycle are enumerated. In detail, the terminal state point refers to the coordinate point corresponding to the last position of a path, representing the production state achieved by adjusting along that path to that point. Feasible next state points refer to all new coordinate points in the two-dimensional decision space that, starting from the current terminal state point, can be reached within the next decision cycle through a single adjustment action, based on production scheduling rules and physical constraints, such as the shortest interval time between processes and the maximum capacity of equipment.

[0094] Specifically, performing the enumeration operation means, based on a predefined set of scheduling actions, such as delaying a process by 5 minutes or adding a backup device to a process, calculating the changes in production risk and benefit indicators that will result from the execution of each action, thereby mapping the corresponding new coordinate points in the decision space. More specifically, this step explores all possible immediate subsequent development directions for each existing path, generating a set of alternative expansion nodes.

[0095] Specifically, each feasible next state point obtained through enumeration is appended to the end of its corresponding original path, forming a new candidate path. In other words, the append operation means that for each new state point generated by the enumeration, it is treated as a new path node and connected to the end of the original path that generated it. In this way, an original path can generate multiple new paths, each one node longer, based on the multiple feasible next state points it has enumerated. More specifically, this operation realizes the expansion and growth of paths, allowing the search to advance from the current state to possible future state spaces.

[0096] Specifically, for each new candidate path, a division operation is performed using the comprehensive production risk index corresponding to the path's terminal state point as the dividend and the comprehensive production efficiency index as the divisor. The resulting quotient is determined as the evaluation cost of the path. More specifically, the production risk index is the value of the path's terminal state point on the first coordinate axis in the two-dimensional decision grid space, i.e., the weighted sum of process delay risk and quality deviation risk.

[0097] The production efficiency index is the value on the second coordinate axis of the terminal state point, which is the weighted sum of capacity utilization and expected revenue. Specifically, performing the division and determination operation means that for each newly generated path, the risk comprehensive index of its terminal state point is calculated and divided by its benefit comprehensive index to obtain a new value. More specifically, this quotient is defined as the evaluation cost of the path, which quantifies the risk ratio required to obtain a unit of benefit; the lower the cost, the better the balance between risk and benefit that the path leads to, and the higher the evaluation quality of the path.

[0098] Specifically, all new candidate paths are sorted in ascending order based on their evaluated cost values, and the top K paths are selected and retained. This sorting and selection process involves arranging all generated paths in ascending order of their calculated evaluated cost values. Then, only the top K paths are retained, while the rest are discarded. More specifically, this step is the core of bundle search, or bundle or pruning operation, which ensures that only the K most promising and best-evaluated paths are retained for further exploration in each generation of expansion, thus significantly reducing the search space and improving search efficiency.

[0099] Specifically, the K retained paths are used as the new candidate path set, and the current iteration number is recorded. More specifically, the K retained paths after pruning represent the currently explored optimal frontier. Performing the update operation means completely replacing the previous round's candidate path set with these K paths, serving as the starting point for the next iteration expansion. Simultaneously, the counter recording the iteration count is incremented. More specifically, this step completes an update of the search state, preparing for the next iteration.

[0100] Specifically, the algorithm's termination condition check involves determining whether the current iteration step count has reached L and whether there exists an end state point in the new candidate path set that has entered the target region. This check includes two parallel conditions: first, whether the recorded current iteration step count has reached or exceeded the preset maximum iteration step size L; and second, whether there is at least one path in the new candidate path set whose final state point's coordinates fall within the previously defined target region. More specifically, these two conditions correspond to the termination requirements of the search process in terms of depth (step count) and objective (achieving the goal), respectively.

[0101] Specifically, the subsequent action is determined based on the judgment result: if not, the enumeration continues; if yes, the current set of candidate paths is output as the final K candidate decision paths. Specifically, executing the decision and output operation means that if neither the maximum number of steps has been reached nor a path enters the target region, the algorithm will start a new round of enumeration, evaluation, sorting, and pruning based on the current new set of candidate paths. If any termination condition is met, the algorithm immediately stops iterating and outputs the currently held set of candidate paths, i.e., the K paths, as the final search results. More specifically, these K candidate decision paths, finally output, are a sequence of scheduling strategies that, starting from the starting point, lead to the satisfactory region (or the farthest position explored) within a finite number of steps, and have undergone multiple rounds of optimization. They constitute the original set for subsequently selecting the optimal decision paths.

[0102] S6. For each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. Specifically, the fluctuation value sequence is a sequence obtained by first-order difference calculation of the estimated execution time sequence of the process, reflecting the continuous change in the execution time of the process. Positive fluctuations indicate time extension, and negative fluctuations indicate time shortening. The symbolic transfer entropy calculation is based on the symbolic transfer entropy model, quantifying the influence intensity of fluctuations in preceding processes on fluctuations in subsequent processes. The core is to calculate the amount of information transferred through symbolic transfer probability. The total entropy value of disturbance propagation is the sum of the symbolic transfer entropies of all directly dependent process pairs in the process influence chain, quantifying the disturbance risk of the flexible scheduling plan on the stability of the production sequence. The smaller the entropy value, the higher the stability. The adjusted voltage amplitude sequence and active power sequence are time-series data of the voltage of key nodes and the active power of key lines of the power grid obtained by the power flow calculation model after the simulated execution of the scheduling plan, reflecting the adjusted power grid operating status. The real-time security entropy value is a comprehensive index calculated based on the power grid security situation assessment function. It is obtained by weighted summation of three entropy components: voltage stability, power flow load, and power flow balance, quantifying the power grid operation security risk. The smaller the entropy value, the higher the security.

[0103] Schematic, for each flexible scheduling plan, an estimated execution time series is generated based on the process adjustment instructions, and a fluctuation value series is obtained through first-order differencing. The fluctuation values ​​are symbolized into a sequence of "+1", "-1", and "0" according to thresholds. The symbol transition probability of directly dependent process pairs is calculated, and the total entropy value of disturbance propagation is obtained by substituting it into the transfer entropy formula. The plan is executed in a simulation, and the voltage of key nodes and the active power series of key lines are obtained through a power grid flow calculation model. The voltage stability entropy component (sum of squares of deviation from rated voltage), the power flow load entropy component (variance of load rate), and the power flow equilibrium entropy component (sequence variance of active power standard deviation) are calculated, and the weighted summation is used to obtain the real-time safety entropy value. By quantifying the plan risk from the dual dimensions of production stability and power grid security, a scientific basis for optimal decision-making is provided.

[0104] Preferably, for each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated, including: For each flexible scheduling plan in the flexible scheduling plan set, the estimated execution time sequence of each process is generated based on the adjustment instructions of the corresponding process in the process influence chain. For each process, the estimated execution time series is calculated using first-order difference to obtain the fluctuation value series of process execution time. Based on the pre-defined symbolic transfer entropy model, each fluctuation value in the fluctuation value sequence is symbolized in time series to obtain a symbolized fluctuation sequence. For each process pair in the process influence chain that has a direct dependency relationship, based on the symbolic fluctuation sequence of the preceding process and the symbolic fluctuation sequence of the following process, the symbolic transfer probability is calculated and substituted into the preset transfer entropy formula to calculate the symbolic transfer entropy from the preceding process to the following process. The total entropy of disturbance propagation is obtained by summing the symbolic propagation entropy of all process pairs with direct dependencies in the process influence chain. After each flexible scheduling plan is executed in the simulation, the voltage amplitude sequence of the grid nodes and the active power sequence of the grid lines are obtained after adjustment. The sum of the squares of the deviations of each data point from the rated voltage value in the voltage amplitude sequence is taken as the voltage stability entropy component; The variance of the ratio of each data point in the active power sequence to the transmission capacity limit of the line is used as the power flow load entropy component, and the variance of the sequence of the standard deviation of the active power values ​​of each critical line at the same time is used as the power flow equilibrium entropy component. The real-time safety entropy value is obtained by weighted summation of the voltage stability entropy component, the power flow load entropy component, and the power flow balance entropy component.

[0105] Specifically, the symbol transfer entropy model is used to analyze the direction and intensity of information transfer between two discrete symbol sequences. It quantifies the impact of fluctuations in preceding processes on subsequent processes by calculating symbol transition probabilities. The symbolized fluctuation sequence is a sequence of discrete symbols converted from a continuous fluctuation value sequence according to a preset threshold, preserving the fluctuation direction and pattern. The symbol transition probability is the conditional probability that a subsequent process will exhibit a certain symbol state when the preceding process is in a certain symbol state. The voltage stability entropy component characterizes the degree to which the grid voltage deviates from its rated value, calculated as the sum of the squares of the deviations of each data point in the voltage amplitude sequence from the rated voltage value. The power flow load entropy component characterizes the degree of fluctuation in the grid line load rate, calculated as the variance of the ratio of each data point in the active power sequence to the line transmission capacity limit. The power flow equilibrium entropy component characterizes the degree of balance in the load distribution of the grid lines, calculated as the series variance of the standard deviation of the active power values ​​of each critical line at the same time.

[0106] Specifically, for each flexible scheduling plan, based on the plan's adjustment instructions for processes in the process influence chain, an estimated execution time sequence for each process is generated. Specifically, the flexible scheduling plan includes specific adjustment instructions for the start and stop times, production order, and resource allocation of related processes in the process influence chain. The execution generation operation refers to parsing these instructions, simulating the adjusted execution logic and timing of each process, predicting its planned start and end times, and arranging them chronologically to form an independent estimated execution time sequence for each process. More specifically, the estimated execution time sequence, in the form of discrete time points, characterizes the expected time interval occupied by each process under the plan and its temporal position in the overall production process, providing a precise temporal basis for subsequent quantification of time fluctuations.

[0107] Specifically, a first-order difference calculation is performed on the estimated execution time series of each process to obtain a sequence of fluctuation values ​​for the process execution time. In detail, the first-order difference calculation is a mathematical operation used to reveal the changing patterns of adjacent terms in a sequence. Specifically, performing the first-order difference calculation means, for a given process's estimated execution time series, sequentially calculating the difference between the value at each subsequent time point and the value at the previous time point, resulting in a new sequence. More specifically, this new sequence is the sequence of fluctuation values ​​for the process execution time, where each value represents the change in the execution time interval between two consecutive planned time points; positive fluctuation values ​​indicate an extended time interval, negative fluctuation values ​​indicate a shortened time interval, and zero values ​​indicate no change in the interval, thus effectively extracting the temporal uncertainty characteristics introduced by the scheduling plan.

[0108] Specifically, based on the symbolic transfer entropy model, each fluctuation value in the fluctuation value sequence is symbolized over time to obtain the symbolized fluctuation sequence for this process. In detail, the symbolic transfer entropy model is a model used to analyze the direction and intensity of information transfer between two discrete symbol sequences, and its core calculation formula is transfer entropy. Symbolization is the process of mapping continuous numerical values ​​to a finite number of discrete symbols. Performing the symbolization operation means setting one or more numerical thresholds and classifying each numerical fluctuation value in the fluctuation value sequence into a predefined symbol based on its relationship with the threshold. For example, positive fluctuations are mapped to the symbol "+1", negative fluctuations to the symbol "-1", and fluctuations close to zero to the symbol "0". More specifically, after this operation, the original numerical fluctuation sequence is transformed into a symbolized fluctuation sequence composed of discrete symbols. This sequence retains the direction and pattern information of the fluctuations but filters out specific numerical details, preparing for the calculation of information transfer between symbols.

[0109] Specifically, for each pair of processes with direct dependencies in the process influence chain topology, based on the symbolic fluctuation sequences of the preceding and subsequent processes, the symbolic transfer entropy from the preceding process to the subsequent process is calculated by calculating the symbolic transition probability and substituting it into the transfer entropy formula. In detail, the process influence chain topology defines the network of dependencies between processes, where a direct dependency means that one process is the immediate predecessor of another. The symbolic transition probability describes the conditional probability that the subsequent process will exhibit a certain symbol when the preceding process is in a certain symbolic state. The standard formula for calculating transfer entropy is used to quantify the degree of uncertainty reduction in the prediction of the future state of one time series for another. Specifically, performing the calculation operation means that for each pair of processes with direct dependencies, the symbolic fluctuation sequence X of its preceding process and the symbolic fluctuation sequence Y of its subsequent process are collected. Based on the joint symbolic state occurrence frequency of sequences X and Y, the required probability distribution is estimated.

[0110] Then, these probabilities are substituted into the standard formula for calculating transfer entropy for calculation: in, The symbol represents the entropy value. , Let Y and X represent the symbol states of the sequences Y and X at time t, respectively. This represents the symbol state of sequence Y at time t+1. This represents probability. More specifically, it is the calculated symbolic propagation entropy value. It numerically measures the amount of information contained in the fluctuation pattern of the preceding process that can reduce the uncertainty of future fluctuation prediction of the subsequent process. The larger the entropy value, the stronger the influence of the preceding process on the fluctuation of the subsequent process, and the more significant the information transmission.

[0111] Specifically, the calculated symbolic transfer entropy is summed for all processes with direct dependencies in the process influence chain to obtain the total entropy value of disturbance propagation. Performing the summation operation involves traversing each directly dependent edge in the process influence chain, extracting its corresponding symbolic transfer entropy value, and arithmetically summing them. More specifically, the final total entropy value of disturbance propagation is a single scalar that aggregates the total intensity of all fluctuation information transmission caused by the scheduling plan across the entire production process dependency network. The higher this total entropy value, the wider and deeper the potential production timing disruptions caused by the plan propagate between processes, meaning a greater potential disturbance risk to the internal stability of the production process.

[0112] Specifically, for each flexible dispatch plan, the adjustment of load demand according to the plan is executed, and the voltage amplitude sequence of key nodes and the active power sequence of key lines are obtained after adjustment based on the power grid flow calculation model. In detail, executing the load demand adjustment of the plan means converting the changes in the electricity consumption time and power level of each production process in the plan into corresponding load change inputs on the grid side. The power grid flow calculation model is a steady-state mathematical model used in power system analysis to solve for the voltage distribution of each node and the power distribution of each branch under given network topology, parameters, and load conditions. Simulation operation based on the power grid flow calculation model means using the adjusted load distribution as model input, performing one or more flow calculations, and outputting the time-varying sequence of voltage amplitude of pre-selected key nodes and the time-varying sequence of active power transmitted by key lines within the simulation period. More specifically, these electrical quantity time series objectively reflect the dynamic evolution of the power grid operating state after the implementation of the dispatch plan.

[0113] Specifically, based on the power grid security assessment function, the sum of the squares of the deviations of each data point in the voltage amplitude sequence from its rated voltage value is used as the voltage stability entropy component; the variance of the ratio of each data point value in the active power sequence to its corresponding line transmission capacity limit is used as the power flow load entropy component; and the series variance of the standard deviations of the active power values ​​of each critical line at the same time is used as the power flow equilibrium entropy component. In detail, the power grid security assessment function is a function designed to comprehensively quantify the security risks of power grid operation. It achieves the assessment by integrating multiple dimensions of insecurity factors. The voltage stability entropy component aims to penalize the degree of voltage deviation from the rated value. The power flow load entropy component aims to measure the severity of fluctuations in line load rate around its average level. The power flow equilibrium entropy component aims to assess the imbalance in load distribution among different lines and its changes over time.

[0114] Simultaneously, the calculation of each entropy component involves: for the voltage amplitude sequence, calculating the square of the difference between the voltage value and the rated voltage at each moment, and summing the squared values ​​at all moments to obtain the voltage stability entropy component S_v; for the active power sequence of each critical line, calculating the ratio of the power value at each moment to the transmission capacity limit of that line to obtain the load rate sequence of that line, and then calculating the variance of the load rate sequences of all lines to obtain the power flow load entropy component S_l; for each sampling moment, calculating the standard deviation of the active power values ​​of all critical lines at that moment to obtain a sequence describing the balance at that moment, and then calculating the variance of this sequence to obtain the power flow balance entropy component S_b. More specifically, these three entropy components extract the potential risk characteristics of the power grid operating state from three key aspects: voltage level, line load fluctuation, and network load distribution.

[0115] Specifically, the voltage stability entropy component, power flow load entropy component, and power flow equilibrium entropy component are substituted into the power grid security situation assessment function, weighted, and summed. The output value of the function is then used as the real-time security entropy value. In detail, the standard form of the power grid security situation assessment function is a linear weighted sum of the security components, i.e.: in, , , These are preset weighting coefficients for the voltage stability entropy component, the power flow load entropy component, and the power flow equilibrium entropy component, respectively, with the sum of these weighting coefficients being 1. The weight values ​​are preset based on the importance of each component to power grid security. Performing a weighted summation operation means applying the calculated voltage stability entropy component... , power flow load entropy component and the entropy component of the current equilibrium Multiply by their respective weighting coefficients , , Then, the three products are added together. The sum S is the real-time security entropy value, which is a comprehensive risk assessment indicator. The higher the entropy value, the greater the overall risk that the power grid's operating state will deviate from the ideal state of safety, stability, and equilibrium after the implementation of the contingency plan. In other words, the more significant the negative impact of the contingency plan on the external security of the power system.

[0116] S7. Perform Pareto mapping based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

[0117] Specifically, Pareto mapping refers to the process of mapping the total entropy value of disturbance propagation and the real-time safety entropy value of a flexible scheduling plan to a two-dimensional target space, and selecting the optimal plan by identifying Pareto non-dominated solutions, i.e., Pareto front analysis.

[0118] Schematic, the dual entropy values ​​of each flexible scheduling plan are combined into a two-dimensional evaluation vector and mapped onto a two-dimensional target space with dual entropy values ​​as axes; Pareto non-dominated coordinate points are identified to form a Pareto front solution set; a preset reference point (the theoretical minimum point of dual entropy value) is set, and the weighted Chebyshev distance from each point in the solution set to the ideal point is calculated; the plan corresponding to the point with the smallest distance is selected as the optimal flexible scheduling plan. By screening the optimal solution that takes into account both production and power grid objectives, the subjective nature of decision-making in existing technologies is addressed.

[0119] Preferably, a Pareto mapping is performed based on the total entropy value of the disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan, including: The total entropy value of disturbance propagation and the real-time safety entropy value corresponding to each flexible scheduling plan are used as a two-dimensional evaluation vector; A two-dimensional target space is constructed with the total entropy of disturbance propagation as the first dimension and the real-time security entropy as the second dimension, and all two-dimensional evaluation vectors are mapped to the two-dimensional target space respectively; wherein, each two-dimensional evaluation vector corresponds to a coordinate point in the two-dimensional target space; In the two-dimensional target space, identify all Pareto non-dominated coordinate points, construct the Pareto front solution set, and calculate the weighted Chebyshev distance from each coordinate point in the Pareto front solution set to the preset reference point. The flexible scheduling plan corresponding to the coordinate point with the smallest weighted Chebyshev distance is taken as the final flexible scheduling plan.

[0120] Specifically, the two-dimensional evaluation vector is an ordered combination of the total perturbation propagation entropy value and the real-time safety entropy value corresponding to each flexible dispatch plan. It is a quantitative representation of the flexible dispatch plan in the dual-objective space of production stability and power grid security. The two-dimensional objective space is a planar space constructed with the total perturbation propagation entropy value as the first dimension (horizontal axis) and the real-time safety entropy value as the second dimension (vertical axis), used to map the dual-objective evaluation results of all plans. Pareto non-dominated coordinate points are points in the two-dimensional objective space where no other coordinate point can strictly optimize the other objective value without worsening one objective value. The Pareto front solution set is the set composed of all Pareto non-dominated coordinate points. The weighted Chebyshev distance is a quantitative indicator measuring the distance between the coordinate points in the Pareto front solution set and the ideal point.

[0121] Specifically, the total entropy value of disturbance propagation and the real-time safety entropy value corresponding to each plan in the flexible dispatch plan set are combined into a two-dimensional evaluation vector. The total entropy value of disturbance propagation is a scalar that quantifies the potential disturbance risk of the dispatch plan to the temporal stability of the production process; the real-time safety entropy value is also a scalar that quantifies the comprehensive impact risk of the plan on the external safety of the power grid operation after its implementation. Performing the combination operation means that for each plan in the plan set, the two scalar entropy values ​​calculated are arranged in a fixed order, for example, the total entropy value of disturbance propagation is placed in the first position and the real-time safety entropy value in the second position, forming a mathematical vector with two components. More specifically, this operation abstracts each complex dispatch plan into a concise mathematical object that can be measured and compared across the two optimization objective dimensions of internal production disturbance and external power grid safety—that is, a two-dimensional evaluation vector—laying a quantitative foundation for subsequent multi-objective comprehensive evaluation.

[0122] Specifically, all two-dimensional evaluation vectors are mapped to a two-dimensional target space with the total entropy of disturbance propagation as the first dimension and the real-time safety entropy as the second dimension. Each vector corresponds to a coordinate point in this space. More specifically, the two-dimensional target space is a virtual mathematical plane. Its horizontal axis is defined as the total entropy of disturbance propagation, with the direction typically set so that an increase in this value represents an increase in the risk of internal disturbances in production. Its vertical axis is defined as the real-time safety entropy, with the direction typically set so that an increase in this value represents an increase in the safety risk external to the power grid. Performing the mapping operation means that for each evaluation vector, based on the values ​​of its two components, a unique position is determined in this two-dimensional plane. The horizontal coordinate of this position is the value of the first component of the vector, and the vertical coordinate is the value of the second component. More specifically, through this mapping, all schemes in the entire flexible dispatch plan set are transformed into a series of discrete coordinate points in this two-dimensional target space, allowing for a direct observation of the distribution and relative advantages and disadvantages of all schemes across the two major objectives from a dual-entropy perspective.

[0123] Specifically, in the two-dimensional target space, all Pareto non-dominated coordinate points are identified, forming the Pareto front solution set. Pareto non-domination is a core concept in multi-objective optimization, used to describe the superiority / inferiority relationship between solutions. A coordinate point (corresponding to a flexible scheduling plan) is called a non-dominated point because, among all plan points, no other point can be strictly superior to it in any other objective without worsening at least one objective. Specifically, the identification operation typically employs a pairwise comparison traversal method: each coordinate point in the target space is examined sequentially, comparing it with all other points; if a point A has a total perturbation propagation entropy value no greater than that of point B, and a real-time safety entropy value strictly less than that of point B, or vice versa, then point A is said to dominate point B; if a point is not dominated by any other point, then that point is a Pareto non-dominated point. Collecting all such non-dominated points constitutes the Pareto front solution set. More specifically, this solution set represents the set of elite solutions in the existing set of contingency plans that achieve the best trade-off between the two conflicting objectives of internal production stability and external power grid security, and cannot be further improved simultaneously. These solutions are distributed on the lower left boundary of the objective space (assuming the objective is to minimize the entropy value), forming a frontier curve or broken line.

[0124] Specifically, the weighted Chebyshev distance from each coordinate point in the Pareto front solution set to a preset reference point is calculated. The preset reference point is a virtual reference point whose coordinates are defined by the decision-maker. It is typically composed of the theoretical optimal values ​​of the two objectives on the Pareto front solution set, i.e., the point determined by the minimum possible value of the total entropy of the perturbation propagation and the minimum possible value of the real-time safety entropy. It represents the ideal state in which both objectives theoretically reach their optimal state simultaneously. The weighted Chebyshev distance is a metric for measuring the distance of a solution to this ideal point. Its characteristics include the ability to balance the differences across the various objective dimensions and its sensitivity to uneven distribution of objective values. Specifically, the calculation involves first assigning weights to the two entropy objectives according to decision preferences; then, for each coordinate point in the front solution set, calculating the absolute difference between its two coordinate values ​​and the corresponding coordinate values ​​of the preset reference point after weight adjustment; finally, taking the maximum of these two differences as the weighted Chebyshev distance from that point to the preset reference point. More specifically, this distance value quantifies the overall gap between each elite solution and the ideal state of dual excellence. The smaller the distance, the closer the solution's overall performance on both objectives is to the ideal state.

[0125] Specifically, the flexible scheduling plan corresponding to the coordinate point with the smallest weighted Chebyshev distance is selected as the optimal flexible scheduling plan. Specifically, the selection operation involves, after completing the calculation in step four, traversing all points in the Pareto front solution set corresponding to the weighted Chebyshev distance values, finding the smallest distance value, and locating its corresponding coordinate point. Then, through reverse mapping of this coordinate point, the original two-dimensional evaluation vector that initially generated this coordinate point is found, thereby determining the specific flexible scheduling plan represented by this vector. More specifically, this selected plan, that is, the plan that most closely approximates the theoretically optimal reference point after comprehensive consideration in the elite plan set, is ultimately determined as the optimal flexible scheduling plan. This method ensures that the final decision is not subjectively arbitrary, but objectively generated under a clear multi-objective optimization framework (Pareto front) and mathematical decision rules (minimizing the weighted Chebyshev distance), taking into account both the inherent stability of the production process and the safety of power system operation.

[0126] By implementing this embodiment, real-time telemetry data, production plan data, status time-series data from equipment IoT sensors, and real-time work order data are acquired simultaneously. This breaks the isolated processing mode of load data and production process data in existing technologies, laying a data foundation for establishing the correlation between load fluctuations and the production process. Then, by separating the high-frequency transient components and low-frequency steady-state baseband components of the load time-series data through wavelet transform, the rapid load fluctuation characteristics caused by sudden events such as equipment start-up and shutdown and process switching can be accurately captured, solving the problem of existing technologies lacking dynamic perception capabilities for transient load fluctuations. Based on preset process constraints, this invention anchors and matches the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data, directly establishing a causal relationship between load fluctuation characteristics and production links, overcoming the bottleneck of existing technologies that struggle to build effective correlations between load fluctuations and production links. By identifying target process links, calculating the intensity value of process interruption impact, and generating process impact chains, dynamic analysis of the production source and impact range of load fluctuations is achieved, changing the status quo of existing technologies that can only passively respond to load fluctuations. Based on the process influence chain, production risk and efficiency indicators are extracted, generating a set of flexible scheduling plans in the decision grid space. By calculating symbolic transfer entropy and assessing grid security entropy, combined with Pareto mapping, the optimal flexible scheduling plan is selected. This overcomes the limitations of existing technologies that rely on fixed thresholds or static scheduling, achieving dynamic optimization scheduling that balances production efficiency and grid security. It ensures targeted scheduling decisions through precise correlation and impact analysis, preventing a decline in production efficiency, while dynamically assessing grid voltage and active power sequences to ensure grid load balance, effectively reducing grid security risks and ultimately achieving synergistic protection of industrial production continuity and grid security and stability.

[0127] See Figure 2 This is a schematic diagram of a load scheduling device based on process dynamic analysis according to an embodiment of the present invention, comprising: The scheduling data acquisition module is used to acquire real-time telemetry data, production plan data, status time-series data of equipment IoT sensors, and real-time work order data. The load time series data transformation module is used to extract load time series data based on real-time telemetry data, production plan data and status time series data, and perform wavelet transform on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. The process constraint matching module is used to perform anchoring matching based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to the preset process constraint relationship, and obtain the matching result; The process influence chain generation module is used to identify the target process step from real-time work order data based on the matching results, calculate the influence intensity value to characterize the process interruption based on the process constraint relationship, the target process step and the matching results, and perform topology analysis based on the influence intensity value and the target process step to generate the process influence chain. The rolling search module for scheduling plans is used to calculate production risk indicators and production benefit indicators based on the influence intensity value in the process influence chain and the process constraint relationship. It then performs a rolling path search in the decision grid space composed of the production risk indicators and production benefit indicators to generate a set of flexible scheduling plans. The scheduling plan entropy calculation module is used to calculate the symbol transfer entropy for each flexible scheduling plan in the flexible scheduling plan set based on the fluctuation value sequence in the corresponding process influence chain, so as to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. The scheduling plan determination module is used to perform Pareto mapping based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

[0128] This invention provides a load scheduling device based on process dynamic analysis. The device acquires real-time telemetry data, production plan data, status time-series data from equipment IoT sensors, and real-time work order data via a scheduling data acquisition module. In a load time-series data transformation module, load time-series data is extracted from the real-time telemetry data, production plan data, and status time-series data. Wavelet transform is performed on the load time-series data to separate high-frequency transient components and low-frequency steady-state baseband components. In a process constraint matching module, anchoring matching is performed based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to preset process constraint relationships to obtain matching results. In a process influence chain generation module, target process steps are identified from the real-time work order data based on the matching results. The impact intensity value used to characterize process interruption is calculated based on the process constraint relationships, the target process steps, and the matching results. Based on the influence intensity value and the target process link, topology analysis is performed to generate a process influence chain. In the scheduling plan rolling search module, production risk indicators and production benefit indicators are calculated based on the influence intensity values ​​and process constraints in the process influence chain. A rolling path search is performed in the decision grid space composed of the production risk indicators and production benefit indicators to generate a flexible scheduling plan set. In the scheduling plan entropy calculation module, for each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. Finally, in the scheduling plan determination module, Pareto mapping is performed based on the total entropy value of disturbance propagation and the real-time safety entropy value to determine the final flexible scheduling plan.

[0129] By synchronously acquiring real-time telemetry data, production plan data, status time-series data from equipment IoT sensors, and real-time work order data, this invention breaks the isolated processing mode of load data and production process data in existing technologies, laying a data foundation for establishing the correlation between load fluctuations and the production process. Then, by using wavelet transform to separate the high-frequency transient components and low-frequency steady-state baseband components of the load time-series data, it can accurately capture the rapid load fluctuation characteristics caused by sudden events such as equipment start-up and shutdown, and process switching, solving the problem of existing technologies lacking dynamic perception capabilities for transient load fluctuations. Based on preset process constraints, this invention anchors and matches the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data, directly establishing a causal relationship between load fluctuation characteristics and production processes, overcoming the bottleneck of existing technologies that struggle to build effective correlations between load fluctuations and production processes. By identifying target process processes, calculating the intensity of process interruption impacts, and generating process impact chains, it achieves dynamic analysis of the production source and impact range of load fluctuations, changing the current situation where existing technologies can only passively respond to load fluctuations. Based on the process influence chain, production risk and efficiency indicators are extracted, generating a set of flexible scheduling plans in the decision grid space. By calculating symbolic transfer entropy and assessing grid security entropy, combined with Pareto mapping, the optimal flexible scheduling plan is selected. This overcomes the limitations of existing technologies that rely on fixed thresholds or static scheduling, achieving dynamic optimization scheduling that balances production efficiency and grid security. It ensures targeted scheduling decisions through precise correlation and impact analysis, preventing a decline in production efficiency, while dynamically assessing grid voltage and active power sequences to ensure grid load balance, effectively reducing grid security risks and ultimately achieving synergistic protection of industrial production continuity and grid security and stability.

[0130] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0131] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0132] Another embodiment of the present invention provides a terminal device, 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 a load scheduling method based on process dynamic analysis as described in the above embodiments. The terminal device may be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The terminal device may include, but is not limited to, a processor and a memory.

[0133] 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. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0134] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. 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 use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0135] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a load scheduling method based on process dynamic analysis as described in the above embodiment.

[0136] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is 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 file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0137] 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 load scheduling method based on process dynamic analysis, characterized in that, include: Acquire real-time telemetry data, production plan data, status time-series data of equipment IoT sensors, and real-time work order data; Load time series data is extracted from real-time telemetry data, production plan data and status time series data. Wavelet transform is performed on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. Anchoring and matching are performed based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to the preset process constraint relationship to obtain the matching result; Based on the matching results, the target process step is identified from the real-time work order data. The impact intensity value used to characterize the process interruption is calculated according to the process constraint relationship, the target process step and the matching results. Based on the impact intensity value and the target process step, topology analysis is performed to generate the process impact chain. Based on the influence intensity value in the process influence chain and the process constraint relationship, production risk indicators and production benefit indicators are calculated. A rolling path search is performed in the decision grid space composed of production risk indicators and production benefit indicators to generate a set of flexible scheduling plans. For each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. Pareto mapping is performed based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

2. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, Load time-series data is extracted based on real-time telemetry data, production plan data, and status time-series data, including: The first data stream is obtained by aligning the timestamps of each data item in the real-time telemetry data, production plan data, and status time sequence data. The first data stream is converted into a data format according to a preset communication protocol to obtain a protocol data stream. The protocol data stream is then aggregated and encapsulated based on a preset data topic to generate a real-time status data stream. Active power time-series data is parsed and extracted from the real-time status data stream and used as load time-series data.

3. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, Wavelet transform is performed on the load time series data to separate the high-frequency transient components and the low-frequency steady-state baseband components, including: Based on the load time series data, a discrete wavelet transform with a preset decomposition scale is performed to obtain the approximate coefficients and detail coefficients of the load time series data at each decomposition scale. Wavelet reconstruction is performed on the approximation coefficients at the maximum decomposition scale to obtain the low-frequency steady-state baseband components of the load time series data. Wavelet reconstruction is performed on the detail coefficients at all decomposition scales except the maximum decomposition scale. All detail components obtained from the wavelet reconstruction are superimposed to obtain the high-frequency transient components of the load time series data.

4. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, Anchoring and matching are performed based on high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to preset process constraints to obtain matching results, including: The local extreme points of the low-frequency steady-state baseband components are used as power feature points, and the starting point of the high-frequency transient components exceeding the preset component threshold is used as transient event feature points. Based on real-time work order data, extract equipment status change signals and process code switching signals, and use these signals as work order event feature points. The initial association is constructed by matching timestamps based on power feature points, transient event feature points, and work order event feature points; wherein, if the absolute time difference between any pair of feature points is less than the allowable time tolerance of the corresponding process in the preset process constraint relationship, the initial association is constructed. Based on the equipment start-stop power logic and process flow sequence in the process constraint relationship, the consistency of the feature point pairs that have been initially associated is checked, and the feature point pairs that contradict the equipment start-stop power logic or process flow sequence are deleted to obtain the verified associated point pairs. Each verified associated point pair is encapsulated as a mapping entry, all mapping entries are aggregated into a mapping relationship set, and the mapping relationship set is used as the matching result.

5. The load scheduling method based on process dynamic analysis as described in claim 4, characterized in that, Based on process constraints, target process steps, and matching results, the impact intensity value used to characterize process interruption is calculated. Then, based on the impact intensity value and the target process step, topology analysis is performed to generate a process influence chain, including: Find all subsequent processes that have the target process as a direct predecessor from the process constraints, and construct a set of direct subsequent processes; Based on the dependencies between processes in the process constraints, starting from the set of directly dependent processes, iteratively query all indirectly dependent processes to construct the complete set of affected processes; Based on the matching results, determine the amplitude corresponding to the transient event feature point associated with the work order event feature point, and take the transient event feature point in the mapping relationship set corresponding to the amplitude as the load transient feature point; The baseline influence coefficient of the target process step in the process constraint relationship is multiplied with the amplitude of the load transient characteristic point to obtain the influence intensity value used to characterize the process interruption. Starting with the impact intensity value, based on the dependencies between processes, the impact intensity propagation calculation is performed on each process in the set of all affected processes. The dependency path is multiplied by the corresponding propagation attenuation factor. For each process in the set of all affected processes, the impact intensity value of all preceding processes of the current process is attenuated according to the impact propagation weight factor and then summed to obtain the cumulative impact intensity value of the current process. The impact propagation weight factor is preset according to the urgency level of the process, and the propagation attenuation factor is preset according to the dependency path between processes. The target process step and subsequent processes whose cumulative influence intensity value exceeds the preset intensity threshold are organized into a topological structure according to their dependencies to generate a process influence chain.

6. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, Production risk indicators include process delay risk indicators and quality deviation risk indicators; production efficiency indicators include capacity utilization rate indicators and expected return indicators. Based on the influence intensity values ​​in the process influence chain and the process constraint relationships, production risk indicators and production benefit indicators are calculated, including: The cumulative influence intensity value of each process is determined based on the influence intensity value in the process influence chain; Multiply the cumulative impact strength value and the time conversion coefficient in the process constraint relationship to obtain the delay time of each process. The ratio of the delay time of each process to the preset standard working hours is used as the process delay risk indicator, and the product of the process quality sensitivity coefficient and the cumulative impact intensity value in the process constraint relationship is used as the quality deviation risk indicator. Add the standard working hours of each process to the delay time to obtain the estimated working hours of each process; The ratio of estimated working hours to the preset total planned working hours is used as the capacity utilization rate indicator. The expected revenue indicator is obtained by summing the product of the output unit price of each process and the preset output quantity.

7. The load scheduling method based on process dynamic analysis as described in claim 6, characterized in that, A rolling path search is performed within the decision grid space comprised of production risk indicators and production efficiency indicators to generate a set of flexible scheduling plans, including: A two-dimensional decision grid space is constructed using the weighted sum of process delay risk indicators and quality deviation risk indicators as the first coordinate axis, and the weighted sum of capacity utilization rate indicators and expected return indicators as the second coordinate axis. Based on the process delay risk index, quality deviation risk index, capacity utilization rate index, and expected return index under the current production status, calculate the coordinates of the current status in the two-dimensional decision grid space and use them as the initial point. Based on the production time and output constraints in the production plan data, the target area is determined in the two-dimensional decision grid space. In the two-dimensional decision grid space, starting from the initial point and ending at the target area, a path search is performed according to the preset cluster search method to obtain several candidate decision paths and their corresponding costs. Candidate decision paths with a cost value less than a preset cost threshold are selected as the preferred decision paths. Based on the coordinates of the decision state point of each preferred decision path, the start and stop times, production order, and resource allocation parameters of the corresponding processes in the process influence chain are adjusted to generate a flexible scheduling plan. The flexible scheduling plans of all preferred decision paths are then integrated to obtain a flexible scheduling plan set.

8. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, For each flexible scheduling plan in the flexible scheduling plan set, the symbol transfer entropy is calculated based on the fluctuation value sequence in the corresponding process influence chain to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated, including: For each flexible scheduling plan in the flexible scheduling plan set, the estimated execution time sequence of each process is generated based on the adjustment instructions of the corresponding process in the process influence chain. For each process, the estimated execution time series is calculated using first-order difference to obtain the fluctuation value series of process execution time. Based on the pre-defined symbolic transfer entropy model, each fluctuation value in the fluctuation value sequence is symbolized in time series to obtain a symbolized fluctuation sequence. For each process pair in the process influence chain that has a direct dependency relationship, based on the symbolic fluctuation sequence of the preceding process and the symbolic fluctuation sequence of the following process, the symbolic transfer probability is calculated and substituted into the preset transfer entropy formula to calculate the symbolic transfer entropy from the preceding process to the following process. The total entropy of disturbance propagation is obtained by summing the symbolic propagation entropy of all process pairs with direct dependencies in the process influence chain. After each flexible scheduling plan is executed in the simulation, the voltage amplitude sequence of the grid nodes and the active power sequence of the grid lines are obtained after adjustment. The sum of the squares of the deviations of each data point from the rated voltage value in the voltage amplitude sequence is taken as the voltage stability entropy component; The variance of the ratio of each data point in the active power sequence to the transmission capacity limit of the line is used as the power flow load entropy component, and the variance of the sequence of the standard deviation of the active power values ​​of each critical line at the same time is used as the power flow equilibrium entropy component. The real-time safety entropy value is obtained by weighted summation of the voltage stability entropy component, the power flow load entropy component, and the power flow equilibrium entropy component.

9. The load scheduling method based on process dynamic analysis as described in claim 1, characterized in that, Based on the total entropy value of the disturbance propagation and the real-time safety entropy value, a Pareto mapping is performed to determine the final flexible scheduling plan, including: The total entropy value of disturbance propagation and the real-time safety entropy value corresponding to each flexible scheduling plan are used as a two-dimensional evaluation vector; A two-dimensional target space is constructed with the total entropy of disturbance propagation as the first dimension and the real-time security entropy as the second dimension, and all two-dimensional evaluation vectors are mapped to the two-dimensional target space respectively; wherein, each two-dimensional evaluation vector corresponds to a coordinate point in the two-dimensional target space; In the two-dimensional target space, identify all Pareto non-dominated coordinate points, construct the Pareto front solution set, and calculate the weighted Chebyshev distance from each coordinate point in the Pareto front solution set to the preset reference point. The flexible scheduling plan corresponding to the coordinate point with the smallest weighted Chebyshev distance is taken as the final flexible scheduling plan.

10. A load scheduling device based on process dynamic analysis, characterized in that, include: The scheduling data acquisition module is used to acquire real-time telemetry data, production plan data, status time-series data of equipment IoT sensors, and real-time work order data. The load time series data transformation module is used to extract load time series data based on real-time telemetry data, production plan data and status time series data, and perform wavelet transform on the load time series data to separate high-frequency transient components and low-frequency steady-state baseband components. The process constraint matching module is used to perform anchoring matching based on the high-frequency transient components, low-frequency steady-state baseband components, and real-time work order data according to the preset process constraint relationship, and obtain the matching result; The process influence chain generation module is used to identify the target process step from real-time work order data based on the matching results, calculate the influence intensity value to characterize the process interruption based on the process constraint relationship, the target process step and the matching results, and perform topology analysis based on the influence intensity value and the target process step to generate the process influence chain. The rolling search module for scheduling plans is used to calculate production risk indicators and production benefit indicators based on the influence intensity value in the process influence chain and the process constraint relationship. It then performs a rolling path search in the decision grid space composed of the production risk indicators and production benefit indicators to generate a set of flexible scheduling plans. The scheduling plan entropy calculation module is used to calculate the symbol transfer entropy for each flexible scheduling plan in the flexible scheduling plan set based on the fluctuation value sequence in the corresponding process influence chain, so as to obtain the total entropy value of disturbance propagation. Based on the adjusted voltage amplitude sequence and active power sequence corresponding to the flexible scheduling plan, the real-time safety entropy value is calculated. The scheduling plan determination module is used to perform Pareto mapping based on the total entropy value of disturbance propagation and the real-time security entropy value to determine the final flexible scheduling plan.

Images