An intelligent online learning system and method for optimizing energy consumption of multiple devices

Abstract

The present invention discloses an intelligent online learning system and method for optimizing the energy consumption of multiple devices, which specifically relates to the field of multi-device energy consumption optimization and is used to solve the problems of energy consumption fluctuation and low operating efficiency caused by unoptimized feedback paths. The system captures key characteristics of device operation, uses causal reasoning to construct a feedback interaction matrix, extracts feedback path weights and action chain characteristics, and analyzes the dynamic interaction relationship between devices; combines cluster analysis and reinforcement learning to optimize classification boundaries, marks high-risk feedback modes and generates optimization templates to provide a basis for path adjustment and strategy optimization; dynamically adjusts feedback path weights, generates device adjustment instructions and combines feedback optimization templates to achieve efficient control of device collaborative operation; thereby effectively suppressing the negative impact of high-risk feedback paths on operation, reducing the cumulative effect of energy consumption caused by feedback enhancement between devices, optimizing device collaborative relationships, and improving the energy consumption allocation efficiency and operational stability of the system in a dynamic load environment.

Description

Technical Field

[0001] The present invention relates to the field of multi-device energy consumption optimization, and more specifically, to an intelligent online learning system and method for multi-device energy consumption optimization. Background Art

[0002] In industrial scenarios where multiple devices operate in coordination, in order to achieve energy consumption optimization, devices usually deploy intelligent online learning algorithms to collect historical operating data in real time, predict future load demand and adjust their own operating strategies. This prediction-based optimization method can reduce the instantaneous energy consumption of equipment to a certain extent and improve resource utilization efficiency, so it is widely used in modern intelligent manufacturing systems. However, when multiple devices in the system operate simultaneously and adopt independent predictive optimization strategies, the predictive behaviors between devices may produce complex dynamic interactions, making the overall operating mode far beyond the simple superposition of individual optimization logic.

[0003] In the prior art, energy consumption optimization algorithms for multiple devices are usually centered on a single device, lacking in-depth modeling and analysis of the interaction of predictive behaviors between devices. When predicting future demand, a device may be indirectly affected by the adjustment behavior of other devices, causing its prediction results to shift. For example, the predictive adjustment of a device may be misjudged by other devices as a systemic demand change, thereby triggering unnecessary energy consumption adjustments. This dynamic feedback effect between devices caused by the predictive optimization strategy may form a self-reinforcing feedback loop, ultimately leading to an increase in overall energy consumption or even systemic fluctuations. Traditional energy consumption optimization methods fail to effectively identify and circumvent this potential problem, making it difficult for the optimization strategy to be stably executed in highly complex multi-device scenarios, thereby limiting the further improvement of the overall optimization effect.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above defects of the prior art, the embodiment of the present invention provides an intelligent online learning system and method for optimizing the energy consumption of multiple devices, which can capture the operating status of the device, predict the behavior data and load synchronization signal in real time, extract key features, ensure a high-precision data foundation, and provide accurate input for subsequent feedback path identification and adjustment; construct a feedback interaction matrix based on causal reasoning and extract the feedback path weight and action chain characteristics to achieve in-depth analysis of the dynamic interaction relationship between devices, and lay a technical foundation for identifying high-risk feedback patterns; optimize the classification boundaries through cluster analysis combined with reinforcement learning, accurately mark high-risk feedback patterns and generate optimization templates, and provide a clear basis for feedback path adjustment and device strategy optimization; adopt dynamic adjustment of feedback path weights, generate device adjustment instructions and combine closed-loop feedback optimization templates to achieve efficient control of collaborative operation between devices. In this way, it not only effectively reduces the impact of high-risk feedback paths on device energy consumption, but also significantly improves the stability and dynamic adaptability of the system, while optimizing the energy consumption distribution and equipment operation efficiency, and solves the energy consumption accumulation effect and response inconsistency caused by the unoptimized feedback path in the collaborative operation of multiple devices, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] S1, for a multi-device system with collaborative operation tasks and sharing and predicting behaviors through physical or logical connections, collect various data and signals of the devices respectively, extract the corresponding features, and generate a device behavior feature data packet in a unified format as the input of causal analysis.

[0008] S2, based on the device behavior feature data packet, defines the causal influence intensity between devices and constructs the feedback interaction matrix between devices; extracts feedback path features and action chain features, quantifies the impact of each feedback path on the overall performance of the system, and generates feature inputs for high-risk feedback pattern identification.

[0009] S3, based on the feedback path characteristics, adopts clustering algorithm combined with reinforcement learning to optimize the classification boundary, classifies the feedback pattern and evaluates the abnormality, and calculates the abnormal intensity of the feedback pattern; when the abnormal intensity is greater than the preset threshold and there is abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing abnormal intensity, feature boundary and optimization suggestions is generated.

[0010] S4, based on the risk feedback annotation and optimization template, dynamically adjust the feedback path weight to weaken the impact of high-risk paths; combine the adjusted feedback path weight and action chain characteristics to generate a device adjustment instruction set, and send the adjustment instruction set to the device for execution.

[0011] In a preferred embodiment, step S1 includes the following contents:

[0012] S1.1. For each device in a multi-device system that has collaborative operation tasks and shares operation loads, energy consumption data, and predicted behaviors through physical or logical connections, three types of data are collected, including operation status data, predicted behavior data, and load synchronization signals.

[0013] S1.2, preprocess the collected data.

[0014] S1.3, extract data features, including load change rate, response time and energy consumption fluctuation rate, to form a feature data packet.

[0015] S1.4, combining the extracted feature quantities to generate a device behavior feature data packet in a unified format.

[0016] In a preferred embodiment, step S2 includes the following contents:

[0017] S2.1, based on the multi-dimensional characteristics of the device behavior feature data packets, define the causal influence intensity between devices.

[0018] S2.2, based on the causal relationship modeling, construct a feedback path matrix between devices to describe the direct action chain between devices in the entire multi-device collaborative system.

[0019] S2.3, calculate the feedback strength by calculating the weighted sum of the feedback path weights between all devices.

[0020] S2.4, finally, based on the feedback path matrix and feedback strength, the action chain characteristics between devices are extracted to quantify the impact of each feedback path on the overall system performance.

[0021] In a preferred embodiment, step S3 includes the following contents:

[0022] S3.1, first, based on the feedback interaction matrix and action chain characteristics, construct the characteristic vector of the feedback pattern , used to describe the main feedback characteristics of the equipment in collaborative operation; the eigenvector is defined as follows: ;in, :equipment Total output impact on other devices; : Other device to device Total input impact of :equipment The feedback modes are grouped by using density clustering algorithm, and the equipment combinations with similar feedback characteristics in operation are mined to preliminarily divide several feedback modes.

[0023] S3.2, after cluster analysis is completed, each type of feedback pattern is evaluated for abnormality; by calculating the abnormal intensity index of the feedback pattern , comprehensively evaluate the feedback path strength, action chain characteristics and comprehensive abnormality of energy consumption fluctuation rate: ;in, : is the total number of devices; Equipment in this category About equipment The feedback path weight reflects the direct connection between devices in the mode; : No. Average action chain characteristics of equipment within the class mode; : No. Average energy consumption fluctuation rate of equipment within the same mode; : The number of devices in the cluster.

[0024] S3.3, Optimizing cluster classification boundaries using reinforcement learning.

[0025] Step S3 also includes the following:

[0026] S3.4. First, based on the classification results and abnormality assessment, the process of judging and marking high-risk feedback patterns is as follows: by calculating the abnormal intensity index of each type of feedback pattern, it is compared with the preset abnormal intensity threshold. When the abnormal intensity index is greater than or equal to the abnormal intensity threshold, the corresponding feedback pattern is marked as a high-risk feedback pattern; at the same time, the characteristic boundary of the pattern is combined to analyze whether the characteristic value distribution exceeds the normal range. If the characteristic boundary satisfies the existence of abnormal distribution, the corresponding feedback pattern is confirmed to be a high-risk pattern; finally, a corresponding optimization template is generated for each type of high-risk feedback pattern.

[0027] In a preferred embodiment, step S4 includes the following contents:

[0028] S4.1, the weight adjustment of the feedback path is based on the original feedback weight, and dynamically weakens the impact of the high-risk path according to the abnormal intensity of the high-risk feedback pattern and the adjustment suggestions in the optimization template; the adjusted feedback path weight proportionally reduces the contribution of the high-risk path while ensuring that the low-risk path maintains its function.

[0029] S4.2, generating a device adjustment instruction set according to the adjusted feedback path weight and action chain characteristics, including power adjustment amplitude, load distribution priority and response correction value.

[0030] The power adjustment amplitude is based on the rate of change of the feedback intensity. By analyzing the sensitivity of the feedback intensity to the power output of the device, the power output direction and amplitude that need to be adjusted for each device are determined.

[0031] The load distribution priority is based on the adjusted feedback path weight and the action chain characteristics of the device. The load priority is allocated by calculating the ratio of the feedback path weight and the action chain characteristics of each device in the feedback path network.

[0032] The response correction value is based on the current response delay of the device and optimizes its adaptability to load changes by dynamically adjusting the weight ratio of its self-feedback path and the paths associated with other devices.

[0033] S4.3, sends the adjustment instruction set to the device and monitors the adjustment effect in real time.

[0034] An intelligent online learning system for optimizing energy consumption of multiple devices includes a behavior capture module, an interaction analysis module, a pattern recognition module and a feedback control module.

[0035] Behavior capture module: For multi-device systems that have collaborative tasks and share and predict behaviors through physical or logical connections, various data and signals of the devices are collected separately, and the corresponding features are extracted to generate device behavior feature data packets in a unified format as input for causal analysis, and the device behavior feature data packets are passed to the interactive analysis module.

[0036] Interaction analysis module: Based on the device behavior feature data packet, define the causal influence intensity between devices and build the feedback interaction matrix between devices; extract feedback path features and action chain features, quantify the impact of each feedback path on the overall system performance, generate feature inputs for high-risk feedback pattern recognition, and pass the feedback path features to the pattern recognition module.

[0037] Pattern recognition module: Based on the characteristics of the feedback path, a clustering algorithm combined with reinforcement learning is used to optimize the classification boundaries, classify the feedback patterns and evaluate their abnormality, and calculate the abnormal intensity of the feedback patterns; when the abnormal intensity is greater than the preset threshold and there is an abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing the abnormal intensity, feature boundary and optimization suggestions is generated, and the optimization template and recognition results are sent to the feedback control module.

[0038] Feedback control module: According to the risk feedback labeling and optimization template, the feedback path weight is dynamically adjusted to weaken the impact of high-risk paths; based on the adjusted feedback path weight and action chain characteristics, a device adjustment instruction set is generated and sent to the device for execution.

[0039] The technical effects and advantages of the intelligent online learning system and method for optimizing energy consumption of multiple devices of the present invention are as follows:

[0040] The present invention captures the equipment operation status, predicts the behavior data and the load synchronization signal in real time, extracts the key features, ensures the high-precision data foundation, and provides accurate input for the subsequent feedback path identification and adjustment; constructs the feedback interaction matrix based on causal reasoning and extracts the feedback path weight and action chain characteristics, realizes the in-depth analysis of the dynamic interaction relationship between devices, and lays a technical foundation for identifying high-risk feedback patterns; optimizes the classification boundary through cluster analysis combined with reinforcement learning, accurately marks the high-risk feedback pattern and generates the optimization template, and provides a clear basis for feedback path adjustment and equipment strategy optimization; adopts dynamic adjustment of feedback path weight, generates equipment adjustment instructions and combines closed-loop feedback optimization template to realize efficient control of collaborative operation between devices. In this way, it not only effectively reduces the impact of high-risk feedback paths on equipment energy consumption, but also significantly improves the stability and dynamic adaptability of the system, optimizes the energy consumption distribution and equipment operation efficiency, and solves the energy consumption accumulation effect and response inconsistency caused by the unoptimized feedback path in the collaborative operation of multiple devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 The present invention is a flowchart of an intelligent online learning method for optimizing energy consumption of multiple devices.

[0042] Figure 2 This is a structural schematic diagram of an intelligent online learning system for optimizing energy consumption of multiple devices according to the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Embodiment 1: Figure 1 The present invention provides an intelligent online learning method for optimizing energy consumption of multiple devices, comprising:

[0045] S1, for a multi-device system with collaborative operation tasks and sharing and predicting behaviors through physical or logical connections, collect various data and signals of the devices respectively, extract the corresponding features, and generate a device behavior feature data packet in a unified format as the input of causal analysis.

[0046] S2, based on the device behavior feature data packet, defines the causal influence intensity between devices and constructs the feedback interaction matrix between devices; extracts feedback path features and action chain features, quantifies the impact of each feedback path on the overall performance of the system, and generates feature inputs for high-risk feedback pattern identification.

[0047] S3, based on the feedback path characteristics, adopts clustering algorithm combined with reinforcement learning to optimize the classification boundary, classifies the feedback pattern and evaluates the abnormality, and calculates the abnormal intensity of the feedback pattern; when the abnormal intensity is greater than the preset threshold and there is abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing abnormal intensity, feature boundary and optimization suggestions is generated.

[0048] S4, based on the risk feedback annotation and optimization template, dynamically adjust the feedback path weight to weaken the impact of high-risk paths; combine the adjusted feedback path weight and action chain characteristics to generate a device adjustment instruction set, and send the adjustment instruction set to the device for execution.

[0049] In the collaborative operation of multiple devices, the operating status, load demand and energy consumption performance of each device are the basic data for energy consumption optimization. However, due to the dynamic changes in the operating characteristics of the equipment and the external load, it is difficult for traditional data collection methods to accurately capture the correlation characteristics between state changes and energy consumption adjustments. Therefore, a behavior capture method based on multi-dimensional data fusion is proposed. By collecting the core operating data of the equipment in real time and performing complex calculations and feature extraction on it, a unified device behavior feature data packet is generated as the basis for subsequent causal analysis.

[0050] Step S1 includes the following contents:

[0051] S1.1. For each device in a multi-device system that has collaborative operation tasks and shares operation loads, energy consumption data, and predicted behaviors through physical or logical connections, three types of data are collected, including operation status data, predicted behavior data, and load synchronization signals.

[0052] Operation status data : Includes the power output of the device (Instantaneous sampling), current fluctuation (short-term changes);

[0053] The operating status data directly reflects the current working condition of the equipment and is the basic data source for analyzing the equipment's load capacity, operating efficiency, and power stability. By collecting power output and current fluctuations, the equipment's workload characteristics and power change patterns can be dynamically monitored, providing a basis for evaluating the equipment's response behavior and operating stability. In addition, the operating status data can reveal the interaction effects between the equipment and the external environment or task changes, and is an important prerequisite for calculating features in other analysis modules.

[0054] Predictive behavior data : Includes the load prediction value of the device itself and predictive adjustments ;

[0055] Predictive behavior data reflects the device's estimation of future load demand and the corresponding adjustment strategy. This data is the core information for evaluating the consistency between the device's prediction and actual operating status. By capturing the device's predicted output, the role of predictive adjustment in energy consumption optimization can be directly analyzed. Especially in collaborative operation scenarios, predictive behavior data can reveal the initiative and adaptability of the device in collaborative adjustment, and is key data for further research on predictive feedback loops.

[0056] Load synchronization signal : Records the time points of load triggering and response between devices, indicating the load coordination characteristics between devices.

[0057] The load synchronization signal records the time points of load triggering and response between devices, directly reflecting the collaborative relationship and synchronization performance between devices. In the multi-device collaborative scenario, the load synchronization signal is the basic data source for analyzing the causal relationship and feedback path between devices. By capturing the time characteristics of triggering and response, the collaborative hysteresis effect and dynamic coupling characteristics between devices can be effectively identified, providing a high-quality time benchmark for building a feedback interaction matrix.

[0058] S1.2, preprocess the collected data, including noise filtering, smoothing and data standardization.

[0059] Noise filtering: Using nonlinear dynamic threshold filtering method, the noise limit is calculated based on the historical data of equipment operation, and the excessively high or low instantaneous abnormal values ​​are filtered out;

[0060] Smoothing: Based on multi-order Gaussian convolution, the unstable short-period fluctuations in the device data are smoothed while retaining the key change points;

[0061] Data standardization: Normalize the collected multidimensional data to ensure that parameters of different dimensions have consistent calculation weights.

[0062] S1.3, extract data features, including load change rate, response time and energy consumption fluctuation rate, to form a feature data packet.

[0063] Calculate the load change rate using the relationship between the equipment power and the load forecast value , the load deviation mean of the equipment in the time window is calculated by integration, which is used to measure the difference between the actual power output of the equipment and the predicted load demand. By integrating and normalizing the absolute value of the power deviation in a short period of time, the stability and adaptability of the equipment's load regulation capability are captured, the interference of short-term fluctuations is avoided, and the degree of deviation between the equipment's operating status and the predicted behavior is reflected.

[0064] Calculate the response time of the device when the load changes according to the load synchronization signal By counting and averaging the time difference between each trigger and response event, the response speed and load synchronization performance of the device are evaluated. By accumulating multiple response time deviations, the hysteresis characteristics of device load regulation are revealed, which helps to identify the collaboration efficiency and potential bottlenecks between devices.

[0065] Energy consumption fluctuation rate The calculation of is based on the variance of power output to define the stability index of power fluctuation. It reflects the stability and smoothness of the power output of the equipment by calculating the root mean square value of the power deviation from its mean value within the time window. This index quantifies the dynamic change amplitude of the power output of the equipment, and is used to evaluate the stability of the equipment's operating status and whether the load change causes drastic fluctuations in the power output.

[0066] S1.4, combine the extracted features to generate a device behavior feature data packet in a unified format : ; Each feature value has a unified timestamp, forming standardized input data and providing a high-quality analysis basis for subsequent steps.

[0067] Step S1 comprehensively captures the current operating characteristics, predictive regulation capabilities, and collaborative response characteristics of the equipment by collecting equipment operating status data, predictive behavior data, and load synchronization signals. Preprocessing methods such as noise filtering, Gaussian smoothing, and feature normalization are used to ensure data accuracy and consistency. Complex calculation formulas for load change rate, power response delay, and energy consumption fluctuation rate are used to extract core features and generate a unified device behavior feature data package, providing a high-quality, standardized input basis for subsequent causal analysis.

[0068] In a complex environment where multiple devices operate in coordination, the mutual influence between devices is dynamic, nonlinear, and time-dependent. Changes in the operating state of a device may affect the energy consumption and operating efficiency of other devices through feedback paths. This feedback relationship may be direct (significant energy consumption regulation linkage between devices) or indirect (indirect effect through other devices). Traditional methods usually rely on static causal analysis, which can only capture significant direct correlations between devices, while ignoring subtle dynamic interaction characteristics and complex feedback paths, making it difficult to reveal the hidden mechanism of action between devices. To solve this problem, this step extracts the interaction characteristics and feedback strength between devices through accurate modeling of causal relationships between devices and dynamic identification of feedback paths, providing reliable input for subsequent risk pattern identification.

[0069] Step S2 includes the following contents:

[0070] S2.1, the goal of causal relationship modeling is to reveal the dynamic impact strength between devices, that is, how the change of the operating state of one device directly or indirectly causes the adjustment of the operating state of other devices. To achieve this goal, based on the multi-dimensional characteristics of the device behavior feature data packet (such as load change rate, power response delay and energy consumption fluctuation rate), the device About equipment The strength of causal influence : ;in: Indicates the device The load change rate reflects its load regulation characteristics; is the suppression factor of load response time delay on causality, reflecting that the greater the delay, the weaker the impact on other devices; For equipment The response time constant quantifies the time sensitivity of the device to load changes.

[0071] The formula captures the device in derivative form The load change of the equipment The sensitivity of state adjustment, combined with the exponential form of the time delay factor, comprehensively considers the timeliness and intensity of dynamic impacts.

[0072] S2.2, based on the causal relationship modeling, further construct the feedback path matrix between devices The elements of this matrix are Indicates the device About equipment The feedback path weight is calculated as follows: ;in: is the total number of devices, indicating the range of all devices in the feedback network. Indicates the device number, ranging from 1 to , indicating all devices in the system. For equipment About equipment The strength of the causal influence.

[0073] This formula ensures the standardization of feedback path weights through normalization processing, revealing the relative importance of direct feedback paths between devices. The feedback path matrix is ​​used to describe the direct action chain between devices in the entire multi-device collaborative system.

[0074] S2.3, the purpose of feedback strength quantification is to evaluate the feedback complexity and interaction strength of the entire system. The feedback strength is calculated by calculating the weighted sum of the feedback path weights between all devices. : ;in: For equipment About equipment The feedback path weight of For equipment The energy consumption fluctuation rate reflects its operation stability; It is a stability suppression factor. The greater the energy consumption fluctuation, the smaller the contribution of the feedback path to the overall system.

[0075] The formula dynamically links the system feedback intensity with the equipment operation stability by adding the exponential form of energy consumption fluctuation rate, laying a quantitative foundation for identifying high-risk feedback loops.

[0076] S2.4, finally, based on the feedback path matrix and feedback strength, extract the action chain characteristics between devices , quantify the impact of each feedback path on the overall system performance. The formula is as follows: ;in: is the Kroneckerdelta function, which is used to filter the direct feedback path; is the feedback strength, which is used to measure the comprehensive effect of the feedback path on system performance.

[0077] The Kroneckerdelta function is a discrete function usually used in matrix or index operations. It is defined as follows: when two indices are equal, the function value is 1, otherwise it is 0, recorded as (when )or (when ). This function is often used to filter specific elements or eliminate irrelevant items. In feedback path analysis, it is used to extract direct feedback relationships between devices, ignoring indirect or irrelevant interactions, thereby focusing on the key features of the target path or action chain.

[0078] The action chain feature extracts the key influencing characteristics of the feedback path, providing multi-dimensional high-value feature input for subsequent risk pattern identification.

[0079] Step S2 reveals the dynamic interaction relationship between devices through causal modeling and feedback path identification, extracts the weight characteristics of direct and indirect feedback paths, and systematically evaluates the degree of mutual influence in the collaborative operation of devices through feedback intensity quantification and action chain feature extraction, providing a high-value input basis for identifying high-risk feedback patterns and formulating optimization strategies. Its role is to construct a feedback interaction matrix and key features between devices to ensure that subsequent analysis can accurately locate potential problems and effectively guide optimization.

[0080] In the collaborative operation of multiple devices, complex feedback interactions between devices may cause the system to enter a high-risk operation mode, such as the cyclic reinforcement of the feedback path causing a surge in energy consumption or a delayed response. These high-risk modes are usually hidden and only appear under specific operating conditions. Traditional methods mostly use static analysis, which can only identify significant anomalies and have difficulty capturing potential problems behind complex interactions. Based on the feedback interaction matrix and action chain features generated in step S2, step S3 extracts and annotates high-risk feedback modes through dynamic pattern recognition methods, and provides accurate strategic basis for subsequent optimization control.

[0081] Step S3 includes the following contents:

[0082] S3.1, first, based on the feedback interaction matrix and action chain features generated in step S2, construct the feature vector of the feedback pattern , which is used to describe the main feedback characteristics of the equipment in collaborative operation. The feature vector is defined as follows: ;in, :equipment The total output impact on other devices reflects its active regulation capability; : Other device to device The total input impact of a, a measure of its state of control; :equipment The action chain characteristics of the factor reveal its comprehensive force in the feedback path.

[0083] Based on these feature vectors, a density clustering algorithm (such as DBSCAN) is used to group feedback patterns, explore device combinations with similar feedback characteristics during system operation, and preliminarily divide several feedback patterns. In this way, the complexity of feedback paths can be converted into discrete pattern categories, providing structured input for subsequent analysis.

[0084] S3.2, after the cluster analysis is completed, each type of feedback pattern is further evaluated for abnormality. Abnormality reflects whether the pattern deviates from the normal collaborative characteristics and has high risk attributes. By calculating the abnormal intensity index of the feedback pattern , comprehensively evaluate the feedback path strength, action chain characteristics and comprehensive abnormality of energy consumption fluctuation rate: ;in, : is the total number of devices; Equipment in this category About equipment The feedback path weight reflects the direct connection between devices in the mode; : No. Average action chain characteristics of equipment within the class mode; : No. The average energy consumption fluctuation rate of the equipment within the same mode measures the operational stability of the equipment; : The number of devices in the cluster, ensuring that the calculation results are related to the size of the device distribution.

[0085] This formula captures the complex coupling relationship between feedback paths, energy consumption fluctuations and action chain characteristics through weight normalization and nonlinear scaling, quantifies the abnormality of each type of pattern, and lays a basis for the screening of high-risk patterns.

[0086] S3.3, with the dynamic changes of equipment operating conditions and load environment, the classification boundary of feedback patterns may drift. In order to improve the adaptability and accuracy of recognition, reinforcement learning is used to optimize the clustering classification boundary. The goal of reinforcement learning is to adjust the key parameters of the clustering algorithm (such as neighborhood radius and density threshold) to maximize the recognition effect of high-risk patterns.

[0087] Defining the reward function for: ;in, : A collection of feedback patterns marked as high risk; : A set of feedback patterns marked as non-high risk; , : Weigh the benefits of labeling high-risk patterns and the penalties of excluding non-high-risk patterns respectively.

[0088] Reinforcement learning dynamically adjusts clustering classification parameters through multiple rounds of training and strategy updates, making the classification boundaries more adaptable to the real-time operating status of the system and ensuring accurate labeling of risk patterns.

[0089] S3.4. First, based on the classification results and abnormality assessment, the process of judging and marking high-risk feedback patterns is as follows: by calculating the abnormal intensity index of each type of feedback pattern, it is compared with the preset abnormal intensity threshold. When the abnormal intensity index is greater than or equal to the abnormal intensity threshold, the corresponding feedback pattern is marked as a high-risk feedback pattern; at the same time, the characteristic boundary of the pattern is combined to analyze whether the characteristic value distribution exceeds the normal range. If there is a significant abnormal distribution within the characteristic boundary (such as the feedback path weight concentration is higher than the corresponding standard or the action chain characteristic abnormality exceeds the standard range), the corresponding feedback pattern is further confirmed as a high-risk pattern to ensure that the identification result comprehensively considers the degree of deviation of the abnormal intensity and the characteristic distribution.

[0090] Finally, generate corresponding optimization templates for each type of high-risk feedback mode The template structure is as follows: ;in, : The abnormal intensity of the corresponding feedback pattern, used to quantify the degree of risk; : Corresponding to the characteristic boundary of the feedback mode, defining the upper and lower limits of the feedback path strength and action chain characteristics; : Optimization suggestions, including specific measures such as feedback path adjustment, priority allocation and device response strategy.

[0091] The generated optimization template will serve as a direct basis for subsequent optimization control and guide the system's coordinated adjustment strategy under high-risk operation mode.

[0092] Step S3 identifies high-risk feedback patterns in the collaborative operation of multiple devices through cluster analysis and abnormality evaluation of feedback patterns, combined with reinforcement learning to optimize classification boundaries, extracts key features of these patterns and generates optimization templates, providing a reliable decision-making basis for multi-device energy consumption optimization. This process not only systematically analyzes the dynamic characteristics of feedback interaction, but also enhances adaptability to complex operating environments through dynamic adjustments, laying a solid foundation for the subsequent collaborative optimization strategy formulation.

[0093] Step S3 has marked high-risk feedback patterns and generated optimization templates, clarifying the characteristic boundaries of feedback paths and potential abnormal paths. However, this information only provides risk labeling and adjustment basis. How to use these results to dynamically generate and execute equipment adjustment strategies is the key to ensuring that high-risk feedback patterns are effectively suppressed and system energy consumption optimization goals are achieved. Step S4 combines the feedback interaction matrix, action chain characteristics and optimization templates to generate specific equipment operation instructions and closed-loop optimization to ensure dynamic stability and global energy consumption optimization effects in the coordinated operation of multiple devices.

[0094] Step S4 includes the following contents:

[0095] S4.1, the weight adjustment of the feedback path is based on the original feedback weight, and the impact of the high-risk path is dynamically weakened according to the abnormal intensity of the high-risk feedback pattern and the adjustment suggestions in the optimization template. The adjusted feedback path weight proportionally reduces the contribution of the high-risk path while ensuring that the low-risk path maintains its function. The higher the abnormal intensity, the greater the weakening of the path weight; the adjustment suggestion value in the optimization template is used to further guide the dynamic allocation of path priorities, thereby achieving the suppression of high-risk paths and the overall balance of the system feedback network. For example, the adjusted feedback path weight The expression can be: ;in, : The original feedback path weight, provided by the feedback interaction matrix generated in step S2; : The abnormal intensity of the high-risk pattern, which is used to measure the risk level of the feedback path; : The recommended path adjustment value given in the optimization template reflects the adjustment priority of the path weight.

[0096] By combining anomaly intensity and optimization suggestions, the weight of high-risk paths can be dynamically weakened to reduce their negative impact on system operation, while retaining the normal functions of low-risk paths to ensure the overall balance and stability of the feedback network.

[0097] S4.2, based on the adjusted feedback path weight and action chain characteristics, generate a device adjustment instruction set, including the power adjustment amplitude , Load distribution priority and response correction value Each instruction is calculated using the following formula:

[0098] The power adjustment amplitude is based on the rate of change of feedback intensity. By analyzing the sensitivity of feedback intensity to device power output, the power output direction and amplitude that needs to be adjusted for each device are determined. Based on the objective function gradient, the device power output is adjusted to reduce the overall feedback intensity, and the devices with a large impact on the feedback path are adjusted first. The adjustment direction gives priority to reducing the power output of devices that contribute more to the system feedback intensity. At the same time, combined with the impact of dynamic changes in the feedback path, ensure that the adjusted power output can effectively reduce the overall feedback intensity and achieve the transition from local optimization to global goals. For example, the power adjustment amplitude The expression can be: ;in: : The total feedback strength of the system, calculated by the feedback matrix in step S2; : The influence of power output on the system feedback strength; : Proportional factor, controls the adjustment range.

[0099] The load distribution priority is based on the adjusted feedback path weight and the action chain characteristics of the device. The load priority is assigned by calculating the ratio of the feedback path weight and the action chain characteristics of each device in the feedback path network. Devices with higher path weights have a greater impact on other devices and are given priority in obtaining load distribution resources. At the same time, combined with the action chain characteristics of the device, its comprehensive role in the feedback network is ensured to be brought into play, so that the load distribution strategy can take into account both the optimization of local paths and the coordination of the overall system. For example, the load distribution priority The expression can be: ;in: :equipment The action chain characteristics.

[0100] The response correction value is based on the current response delay of the device, and optimizes its adaptability to load changes by dynamically adjusting the weight ratio of its self-feedback path and the path associated with other devices. If the weight of the device's self-feedback path is large, the excessive dependence of the device on its own historical state during operation is weakened by reducing the impact of the self-feedback path; if the weight of the path associated with other devices is high, its response ability to external load signals is strengthened first to improve the overall coordination and dynamic adaptability of the system. For example, the response correction value The expression can be: ;in: :equipment The current response delay of :equipment The self-feedback path weight of : Prevent small amounts from having a denominator of zero.

[0101] S4.3, send the adjustment instruction set to the device, and monitor the adjustment effect in real time, including dynamic changes in the feedback path, power output adjustment, and system energy consumption fluctuations. For example, the adaptive update formula of the optimization template can be: ; : Original optimized template; : Learning rate, used to balance the update rate; : The impact of the optimization template on the system energy consumption fluctuation.

[0102] By updating and optimizing the template through real-time feedback, the adaptability of the template to the actual operating environment is gradually improved, ensuring the continued effectiveness of the adjustment strategy under dynamically changing conditions.

[0103] Step S4 dynamically adjusts the feedback path weight by combining the feedback interaction matrix, high-risk feedback annotation and optimization template, generates equipment adjustment instructions, and updates the optimization template through a real-time feedback mechanism, aiming to eliminate the negative impact of high-risk feedback paths on system operation, optimize the collaborative relationship between devices, and achieve global energy consumption optimization and dynamic stability improvement of multi-device systems. Its role is to ensure the pertinence and adaptability of the feedback control strategy, effectively reduce the overall energy consumption fluctuations of the system and improve operating efficiency.

[0104] Embodiment 2: Figure 2 The invention provides an intelligent online learning system for optimizing energy consumption of multiple devices, comprising a behavior capture module, an interaction analysis module, a pattern recognition module and a feedback control module.

[0105] Behavior capture module: For multi-device systems that have collaborative tasks and share and predict behaviors through physical or logical connections, various data and signals of the devices are collected separately, and the corresponding features are extracted to generate device behavior feature data packets in a unified format as input for causal analysis, and the device behavior feature data packets are passed to the interactive analysis module.

[0106] Interaction analysis module: Based on the device behavior feature data packet, define the causal influence intensity between devices and build the feedback interaction matrix between devices; extract feedback path features and action chain features, quantify the impact of each feedback path on the overall system performance, generate feature inputs for high-risk feedback pattern recognition, and pass the feedback path features to the pattern recognition module.

[0107] Pattern recognition module: Based on the characteristics of the feedback path, a clustering algorithm combined with reinforcement learning is used to optimize the classification boundaries, classify the feedback patterns and evaluate their abnormality, and calculate the abnormal intensity of the feedback patterns; when the abnormal intensity is greater than the preset threshold and there is an abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing the abnormal intensity, feature boundary and optimization suggestions is generated, and the optimization template and recognition results are sent to the feedback control module.

[0108] Feedback control module: According to the risk feedback labeling and optimization template, the feedback path weight is dynamically adjusted to weaken the impact of high-risk paths; based on the adjusted feedback path weight and action chain characteristics, a device adjustment instruction set is generated and sent to the device for execution.

[0109] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0110] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0111] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0112] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An intelligent online learning method for optimizing multi-device energy consumption, characterized in that: Includes steps: S1, for a multi-device system with collaborative operation tasks and sharing and predicting behaviors through physical or logical connections, collect various data and signals of the devices respectively, extract corresponding features, and generate a unified format of device behavior feature data packets as input for causal relationship analysis; S2, based on the device behavior feature data package, defines the causal influence strength between devices and constructs the feedback interaction matrix between devices; extracts feedback path features and action chain features, quantifies the impact of each feedback path on the overall system performance, and generates feature inputs for high-risk feedback pattern recognition; S3, based on the feedback path characteristics, adopts clustering algorithm combined with reinforcement learning to optimize the classification boundary, classifies the feedback pattern and evaluates the abnormality, and calculates the abnormal intensity of the feedback pattern; when the abnormal intensity is greater than the preset threshold and there is an abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing abnormal intensity, feature boundary and optimization suggestions is generated; Step S3 includes the following contents: S3.1, first, based on the feedback interaction matrix and action chain characteristics, construct the characteristic vector of the feedback pattern , used to describe the main feedback characteristics of the equipment in collaborative operation; the eigenvector is defined as follows: ;in, :equipment Total output impact on other devices; : Other device to device Total input impact of :equipment The feedback modes are grouped by using density clustering algorithm, and the equipment combinations with similar feedback characteristics in operation are mined to preliminarily divide several feedback modes. S3.2, after cluster analysis is completed, each type of feedback pattern is evaluated for abnormality; by calculating the abnormal intensity index of the feedback pattern , comprehensively evaluate the feedback path strength, action chain characteristics and comprehensive abnormality of energy consumption fluctuation rate: ;in, : is the total number of devices; Equipment in this category About equipment The feedback path weight reflects the direct connection between devices in the mode; : No. Average action chain characteristics of equipment within the class mode; : No. Average energy consumption fluctuation rate of equipment within the same mode; : The number of devices in the cluster; S3.3, Optimizing cluster classification boundaries using reinforcement learning; S4, based on the risk feedback annotation and optimization template, dynamically adjust the feedback path weight to weaken the impact of high-risk paths; combine the adjusted feedback path weight and action chain characteristics to generate a device adjustment instruction set, and send the adjustment instruction set to the device for execution.

2. The intelligent online learning method for optimizing multi-device energy consumption according to claim 1, characterized in that: Step S1 includes the following contents: S1.1, for each device in a multi-device system with collaborative operation tasks and sharing operation load, energy consumption data and predicted behavior through physical or logical connections, three types of data are collected, including operation status data, predicted behavior data and load synchronization signals; S1.2, preprocessing the collected data; S1.3, extracting data features, including load change rate, response time and energy consumption fluctuation rate, to form a feature data packet; S1.4, combining the extracted feature quantities to generate a device behavior feature data packet in a unified format.

3. The intelligent online learning method for optimizing multi-device energy consumption according to claim 2, characterized in that: Step S2 includes the following contents: S2.1, based on the multi-dimensional characteristics of the device behavior feature data packets, define the causal influence intensity between devices; S2.2, based on the causal relationship modeling, construct the feedback path matrix between devices to describe the direct action chain between devices in the entire multi-device collaborative system; S2.3, calculating the feedback strength by calculating the weighted sum of the feedback path weights between all devices; S2.4, finally, based on the feedback path matrix and feedback strength, the action chain characteristics between devices are extracted to quantify the impact of each feedback path on the overall system performance.

4. The intelligent online learning method for optimizing multi-device energy consumption according to claim 1, characterized in that: Step S3 also includes the following contents: S3.

4. First, based on the classification results and abnormality assessment, the process of judging and marking high-risk feedback patterns is as follows: by calculating the abnormal intensity index of each type of feedback pattern, it is compared with the preset abnormal intensity threshold. When the abnormal intensity index is greater than or equal to the abnormal intensity threshold, the corresponding feedback pattern is marked as a high-risk feedback pattern; at the same time, the characteristic boundary of the pattern is combined to analyze whether the characteristic value distribution exceeds the normal range. If the characteristic boundary satisfies the existence of abnormal distribution, the corresponding feedback pattern is confirmed to be a high-risk pattern; finally, a corresponding optimization template is generated for each type of high-risk feedback pattern.

5. The intelligent online learning method for optimizing multi-device energy consumption according to claim 4, characterized in that: Step S4 includes the following contents: S4.1, the weight adjustment of the feedback path is based on the original feedback weight, and the influence of the high-risk path is dynamically weakened according to the abnormal strength of the high-risk feedback pattern and the adjustment suggestions in the optimization template; the adjusted feedback path weight proportionally reduces the contribution of the high-risk path while ensuring that the low-risk path maintains its function; S4.2, generating a device adjustment instruction set according to the adjusted feedback path weight and action chain characteristics, including a power adjustment amplitude, a load distribution priority, and a response correction value; The power adjustment amplitude is based on the rate of change of feedback intensity. By analyzing the sensitivity of feedback intensity to the power output of the equipment, the power output direction and amplitude that needs to be adjusted for each equipment are determined. The load allocation priority is based on the adjusted feedback path weight and the action chain characteristics of the device. The load priority is allocated by calculating the ratio of the feedback path weight and the action chain characteristics of each device in the feedback path network; The response correction value is based on the current response delay of the device, and optimizes its adaptability to load changes by dynamically adjusting the weight ratio of its self-feedback path and the paths associated with other devices; S4.3, sends the adjustment instruction set to the device and monitors the adjustment effect in real time.

6. An intelligent online learning system for optimizing energy consumption of multiple devices, used to implement an intelligent online learning method for optimizing energy consumption of multiple devices as described in any one of claims 1 to 5, characterized in that: include: Behavior capture module, interaction analysis module, pattern recognition module and feedback control module; Behavior capture module: For multi-device systems with collaborative tasks and sharing and predicting behaviors through physical or logical connections, the module collects various data and signals of the devices respectively, extracts corresponding features, generates device behavior feature data packets in a unified format, and transmits the device behavior feature data packets to the interaction analysis module as input for causal relationship analysis; Interaction analysis module: Based on the device behavior feature data package, it defines the causal influence intensity between devices and builds the feedback interaction matrix between devices; Extract feedback path features and action chain features, quantify the impact of each feedback path on the overall system performance, generate feature inputs for high-risk feedback pattern recognition, and pass feedback path features to the pattern recognition module; Pattern recognition module: Based on the feedback path characteristics, clustering algorithm combined with reinforcement learning is used to optimize the classification boundary, classify the feedback pattern and evaluate the abnormality, and calculate the abnormal intensity of the feedback pattern; when the abnormal intensity is greater than the preset threshold and there is an abnormal distribution in the feature boundary, the corresponding feedback pattern is marked as a high-risk feedback pattern, and an optimization template containing the abnormal intensity, feature boundary and optimization suggestions is generated, and the optimization template and recognition results are sent to the feedback control module; Feedback control module: According to the risk feedback labeling and optimization template, the feedback path weight is dynamically adjusted to weaken the impact of high-risk paths; based on the adjusted feedback path weight and action chain characteristics, a device adjustment instruction set is generated and sent to the device for execution.

Images