A power dynamic distribution control system for building energy consumption-oriented photoelectric curtain wall

Abstract

This invention discloses a dynamic power allocation control system for photovoltaic curtain walls in building energy consumption, specifically relating to the fields of building energy conservation and distributed energy management. It includes: a multi-source heterogeneous sensing and acquisition module, a source-load characteristic sensing and prediction module, a dynamic priority iterative power allocation core calculation module, a zoned power execution and control module, a closed-loop feedback and self-learning optimization module, and a human-computer interaction and early warning module. This invention achieves dynamic source-load prediction through refined data acquisition and analysis of building and curtain wall zones; it calculates the optimal power allocation command that balances photovoltaic absorption, economic efficiency, and energy storage losses by constructing a dynamic priority and multi-objective optimization model; and it ensures accurate and reliable command execution and continuous system optimization through smooth execution, closed-loop feedback, and self-learning optimization. This invention solves the problems of inaccurate source-load matching, singular optimization, and weak adaptability, achieving intelligent and efficient building energy management.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation and distributed energy management technology, and more specifically, to a dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption. Background Technology

[0002] Currently, in the fields of building energy conservation and distributed energy management, incorporating building-integrated photovoltaic (BIPV) systems into building energy management systems has become an important development direction. Existing technologies typically employ centralized energy management systems, using fixed scheduling strategies or simple rule-based power allocation methods to uniformly regulate building-integrated photovoltaic power generation, energy storage units, grid power supply, and building load. These systems, to some extent, achieve localized consumption of renewable energy and preliminary optimization of electricity costs.

[0003] However, in practical use, it still has some shortcomings. For example, the system generally lacks the ability to perceive and dynamically analyze the power generation characteristics of each physical zone of the photovoltaic curtain wall, and fails to fully consider the spatiotemporal correlation and dynamic changes of energy consumption demand in each functional area inside the building, resulting in the separation of "source-load" analysis and difficulty in achieving accurate matching; the power allocation strategy is mostly based on static or semi-static priority settings, and cannot be dynamically iteratively optimized according to multiple factors such as real-time electricity price, energy storage status, and energy urgency, and the optimization target is often singular, with insufficient global collaborative optimization capability; the power execution and control link lacks effective smooth adjustment and fault redundancy mechanism, which can easily lead to power fluctuations and affect the safe and stable operation of the power grid and equipment; the system generally does not have the ability to perform closed-loop feedback and self-learning optimization based on actual operating data, and it is difficult to continuously adapt to dynamic changes in building energy consumption patterns, lighting conditions, power grid policies, etc., and there is a risk of long-term operational efficiency decline. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption, which solves the problems mentioned in the background art through the following scheme.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption, comprising: Multi-source heterogeneous sensing and acquisition module: Deployed in the functional zones of the building and the physical zones of the photovoltaic curtain wall, it collects building energy consumption parameters, photovoltaic curtain wall parameters, environmental parameters, power grid status parameters and energy storage unit parameters, and preprocesses the collected raw data to generate standardized preprocessed data; Source-load characteristic perception and prediction module: Based on the preprocessed data, analyze the effective adjustable power of each physical zone of the photovoltaic curtain wall, and perform spatiotemporal coupling prediction of the energy consumption demand of each functional area of ​​the building, generating an analysis report containing source-end power characteristics and load-end prediction data. The core calculation module for dynamic priority iterative power allocation integrates the analysis report, grid status data, and energy storage operation data. After consistency verification, it constructs and solves a multi-objective optimization model based on the functional area energy consumption priority calculated dynamically and iteratively, with photovoltaic absorption rate, grid electricity purchase cost, and energy storage loss as optimization objectives. The optimal power allocation value is obtained, and standardized power allocation instructions are generated. The partitioned power execution control module receives and parses the power allocation command, controls the distributed intelligent execution nodes deployed in the corresponding physical partitions and functional areas, drives the power electronic actuators to complete the smooth adjustment of power, and feeds back the actual execution results. Closed-loop feedback and self-learning optimization module: Collects actual data of the entire link of source, load, grid and storage after power allocation is executed, calculates the deviation from the theoretical value and performs hierarchical correction, and at the same time performs self-learning iterative optimization of the core strategy and parameters of the system based on reinforcement learning algorithm; Human-computer interaction and early warning module: Provides multi-dimensional visual monitoring of system operation status, supports manual / automatic mode switching and strategy simulation, triggers graded early warnings based on full-link anomalies and performs coordinated handling, and realizes collaborative management of system data.

[0006] The technical effects and advantages of this invention are as follows: This invention achieves panoramic, high-precision, and high-reliability data acquisition of the operational status of each functional area of ​​a building and each physical zone of a photovoltaic curtain wall through a multi-source heterogeneous sensing and acquisition module deployed in a zoned manner. Combined with the source-load characteristic sensing and prediction module, it provides spatiotemporal coupling prediction of photovoltaic output characteristics and building energy consumption demand, thereby solving the problems of source-load analysis separation and ignoring zone and spatiotemporal differences in traditional systems, and laying a solid data foundation for accurate dynamic power allocation. This invention constructs a multi-objective optimization model with photovoltaic absorption rate, grid electricity purchase cost and energy storage loss as the core through a dynamic priority iterative power allocation core calculation module. It also introduces priority weights dynamically calculated based on factors such as real-time electricity price, energy storage status and energy urgency, to achieve the comprehensive optimization of economic benefits and clean energy absorption, and dynamically adapt to changes in external conditions. This invention achieves fast and shock-free execution of power commands by using a partitioned power execution control module, employing a smooth adjustment algorithm based on power electronic devices and a millisecond-level fault self-diagnosis and redundancy switching mechanism. This invention establishes a complete closed loop of "monitoring-analysis-correction-optimization" through closed-loop feedback and self-learning optimization modules, and uses reinforcement learning algorithms to continuously iterate and optimize core strategies and parameters, thereby enabling it to adapt to changes in building energy consumption patterns, equipment aging and policy adjustments in the long term, maintain the system in an efficient and optimal state, and reduce dependence on external manual intervention and long-term operation and maintenance costs. This invention enables multi-dimensional visualization of system operation status and historical data tracing through human-computer interaction and early warning modules, and supports strategy simulation and safe manual intervention. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0008] Figure 2 This is a schematic diagram of the source load characteristic sensing and prediction module of the present invention.

[0009] Figure 3 This is a schematic diagram of the core operation module for dynamic priority iterative power allocation of the present invention.

[0010] Figure 4 This is a schematic diagram of the closed-loop feedback and self-learning optimization module structure of the present invention. Detailed Implementation

[0011] 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.

[0012] refer to Figures 1-4 The photovoltaic curtain wall power dynamic distribution control system shown includes: Multi-source heterogeneous sensing and acquisition module: Deployed in the functional zones of the building and the physical zones of the photovoltaic curtain wall, it collects building energy consumption parameters, photovoltaic curtain wall parameters, environmental parameters, power grid status parameters and energy storage unit parameters, and preprocesses the collected raw data to generate standardized preprocessed data; Source-load characteristic perception and prediction module: Based on the preprocessed data, analyze the effective adjustable power of each physical zone of the photovoltaic curtain wall, and perform spatiotemporal coupling prediction of the energy consumption demand of each functional area of ​​the building, generating an analysis report containing source-end power characteristics and load-end prediction data. The core calculation module for dynamic priority iterative power allocation integrates the analysis report, grid status data, and energy storage operation data. After consistency verification, it constructs and solves a multi-objective optimization model based on the functional area energy consumption priority calculated dynamically and iteratively, with photovoltaic absorption rate, grid electricity purchase cost, and energy storage loss as optimization objectives. The optimal power allocation value is obtained, and standardized power allocation instructions are generated. The partitioned power execution control module receives and parses the power allocation command, controls the distributed intelligent execution nodes deployed in the corresponding physical partitions and functional areas, drives the power electronic actuators to complete the smooth adjustment of power, and feeds back the actual execution results. Closed-loop feedback and self-learning optimization module: Collects actual data of the entire link of source, load, grid and storage after power allocation is executed, calculates the deviation from the theoretical value and performs hierarchical correction, and at the same time performs self-learning iterative optimization of the core strategy and parameters of the system based on reinforcement learning algorithm; Human-computer interaction and early warning module: Provides multi-dimensional visual monitoring of system operation status, supports manual / automatic mode switching and strategy simulation, triggers graded early warnings based on full-link anomalies and performs coordinated handling, and realizes collaborative management of system data.

[0013] The multi-source heterogeneous sensing and acquisition module, through partitioned deployment, high-precision acquisition, and redundant verification design, solves problems such as data distortion and poor scenario adaptability caused by "holistic, low-precision, and non-redundant" approaches, providing the entire system with a real, complete, and usable data source. The specific process is as follows: Data Acquisition Node Deployment and Networking: The building is divided into functional zones such as office area, computer room area, HVAC area, and lighting area, based on building function attributes. It is also physically divided into zones based on the orientation of the photovoltaic curtain wall, floor level, and component batch. Each zone independently deploys one distributed data acquisition node. For core parameters such as active power and photovoltaic output power, dual-sensor redundancy is configured to improve data acquisition reliability. Each data acquisition node is networked with the edge computing unit and core processing unit via Profinet industrial Ethernet, with a preset communication rate of ≥100Mbps, ensuring real-time data acquisition and transmission at the transmission level and meeting the data timeliness requirements of the system's high-frequency operations.

[0014] Operating condition identification and acquisition frequency switching: After the system starts, it defaults to normal operating condition, controlling all acquisition nodes to perform normal data acquisition at a frequency of 1 second / time; at the same time, it monitors the system operating condition in real time. When special operating conditions such as the switching of power grid peak and valley periods, sudden changes in light intensity of ±200lx / 10s, and sudden changes in regional energy consumption of ±10% / 5s are identified, the acquisition frequency is automatically triggered to switch, and all nodes are synchronously adjusted to a high-frequency acquisition mode of 100ms / time; after the special operating conditions return to normal, the acquisition frequency automatically drops back to 1 second / time.

[0015] Each distributed data acquisition node synchronously acquires five categories of parameters according to a unified time sequence, and strictly controls the acquisition accuracy of each category of parameters to ensure the accuracy of the raw data. Building energy consumption parameters: real-time active power (accuracy ≤ ±0.5%), reactive power, voltage, current, and total energy consumption of the building in each functional area; Photovoltaic curtain wall parameters: real-time output power (accuracy ≤ ±0.5%), voltage, current, component temperature of each physical zone, and total output power of the photovoltaic curtain wall; Environmental parameters: outdoor light intensity (accuracy ≤ ±20lx), temperature, wind speed, rainfall, and indoor temperature and humidity of the building; Power grid status parameters: real-time voltage (accuracy ≤ ±0.2%), current, frequency, peak-valley-flat electricity price information, and power purchase / sale limits; Energy storage unit parameters: energy storage battery state of charge (SOC, accuracy ≤ ±1%), charge and discharge power, battery temperature, and remaining capacity.

[0016] After the collected raw data is transmitted to the edge computing unit, a standardized three-step preprocessing process is executed to eliminate data interference, fill in missing data, and unify the data format: A moving average method is used for data denoising to filter out invalid interference caused by instantaneous fluctuations and restore the true characteristics of the data; the mean of three adjacent data sets is used to fill in missing data to avoid downstream analysis errors caused by data breaks; and dimensional normalization is performed. Simultaneously, redundancy verification is performed on the core parameter data collected by the dual sensors: when the deviation between the two sensor data is ≤±1%, the mean of the two data sets is taken as valid data; when the deviation is >±1%, the faulty sensor is immediately marked and the backup sensor data is activated, ultimately generating distortion-free, standardized preprocessed data.

[0017] The source-load characteristic sensing and prediction module, through its design of partitioned analysis, spatiotemporal characteristic coupling, and source-load data linkage, solves problems such as source-load matching lag and poor power allocation adaptability caused by the separation of source-load analysis and the neglect of partition differences and spatiotemporal variation characteristics. It is a key data analysis link for realizing source-grid-load-storage coordinated optimization. The specific process is as follows: Power Characteristic Analysis of Photovoltaic Curtain Wall Zones: This analysis provides a refined and real-time analysis of the power output characteristics of each physical zone of the photovoltaic curtain wall. A multi-factor correction model is used to calculate the actual effective adjustable power of each zone, quantifying the power adjustment capability and complementarity of each zone, thus providing accurate source-end data for power allocation. Basic parameter import and model initialization: Read the nominal maximum output power of each physical zone of the photovoltaic curtain wall from the system's local database. The system initializes the full life cycle power characteristic model for each partition by taking into account the component model, factory parameters, and actual installation time, combined with the historical degradation records of the photovoltaic modules in that partition stored in the system, ensuring that the model matches the actual operating status of the partition modules.

[0018] Real-time calculation of multiple correction factors: Real-time reception of real-time illuminance in each photovoltaic curtain wall zone Component operating temperature Combined with the service life of the partition components The system calculates three core correction factors—illuminance, temperature, and aging—to achieve multi-dimensional real-time correction of nominal power. The calculation rules for each factor are as follows: Illuminance Correction Factor Calculated based on a standard test light intensity of 1000 lx normalized; temperature correction factor. With a standard operating temperature of 25℃, it adapts to the characteristics of component power changing with temperature; aging correction factor , The calculation rule, which is based on the actual service life of the module, conforms to the industry standard that the annual degradation rate of photovoltaic modules is ≤0.8%, and is consistent with the degradation characteristics of the module throughout its entire life cycle.

[0019] Effective adjustable power solution and optimization: Substitute the initial nominal maximum output power and the three real-time calculated correction factors into the formula. Calculate the actual maximum output power of each photovoltaic curtain wall zone; take 5% of the nominal maximum output power of the zone as the minimum output power. ,pass Calculate the effective adjustable power of each zone and sum them up to obtain the total effective adjustable power of the photovoltaic curtain wall. Simultaneously, the power regulation dead zone for each zone is quantified to ±2%. By combining the differences in operating conditions such as illumination and temperature in each zone, the power complementarity of the zones is analyzed, and finally a power characteristic report of the photovoltaic curtain wall zones containing the power characteristics, adjustment capabilities, and complementarity of each zone is generated.

[0020] Building Energy Consumption Spatiotemporal Coupling Prediction: This method performs spatiotemporal coupling prediction of energy consumption demands in various functional areas of a building. Through feature selection, model training, and dynamic correction, it achieves accurate predictions of energy consumption at the minute, hour, and day levels, providing reliable load-side data for power allocation. Its core principle aligns with the functional area correlations and time-period patterns of building energy consumption. Input Feature Filtering and Weighting: Standardized multi-dimensional data is received in real time. Through data feature extraction and correlation analysis, core features affecting building energy consumption are filtered out, including historical energy consumption data of each functional area of ​​the building in the past 7 days, real-time operating conditions of the system, outdoor and indoor environmental parameters, and peak-valley-flat electricity price information. Differentiated weights are assigned to each core feature based on the attention mechanism. Among them, indoor temperature and peak-valley electricity price have the most significant impact on energy consumption, and their weights are set to 30%. The remaining features are assigned the remaining weights according to their correlation, ensuring that the input data of the prediction model closely matches the actual factors affecting energy consumption.

[0021] Spatiotemporal coupling factor calculation and model training: Based on the energy consumption correlation of various functional areas of the building (such as the linkage between energy consumption in the HVAC area and the office area) and the energy consumption patterns at different times (such as the energy consumption difference between peak and off-peak periods), the spatiotemporal coupling factor of each functional area and each time period is calculated. The value range of this factor is [0.8, 1.2]. For functional areas with stable energy consumption, such as computer rooms, the value is limited to... The value fluctuation is ≤±0.05, ensuring the prediction accuracy under stable operating conditions; the improved attention mechanism LSTM model is called, and the core features selected and the calculated spatiotemporal coupling factor are used as model input. The model is trained based on the historical energy consumption data stored in the system until the model loss rate is ≤3%, ensuring that the model has high-precision energy consumption prediction capability.

[0022] Prediction and Dynamic Correction: The core feature data collected in real time is input into the trained spatiotemporal coupled prediction model, using the formula: Complete the calculation of predicted energy consumption values ​​for each functional area of ​​the building and for each time period, including For historical energy consumption data, For real-time energy consumption data, For historical data weighting coefficients, Environmental correction factor; To offset prediction errors caused by changes in real-time operating conditions, the system adjusts the weighting coefficients every 5 minutes based on the transmitted actual energy consumption data. With environmental correction factors Dynamic adjustments are made to ensure that the model's prediction accuracy for energy consumption within one hour is ≥95%, meeting the real-time and accuracy requirements of downstream power allocation.

[0023] The core computation module for dynamic priority iterative power allocation completes the entire process of optimal power allocation through five consecutive steps: multi-source data integration and verification, dynamic priority iterative calculation, multi-objective optimization model construction, optimal solution solving, and standardized instruction output. The specific process is as follows: Multi-source data integration and consistency verification; real-time reception of source load analysis and prediction reports (including effective adjustable power of each zone of the photovoltaic curtain wall). Total effective adjustable power Predicted energy consumption values ​​per minute / hour for each functional area of ​​the building ), and simultaneously receive real-time grid status data (peak-valley-flat electricity price). Power purchase / sale limits (Grid voltage / frequency) and real-time operating data of energy storage units (SOC, charge / discharge power limits) (Battery temperature), and all data are integrated into a standardized computational dataset according to the four dimensions of "source end - load end - grid end - storage end". Consistency checks are performed on the dataset to remove invalid data with incorrect format or exceeding the range, and missing key data is supplemented with the latest collected data to ensure the authenticity, completeness and validity of the computational data source, laying the data foundation for subsequent algorithm calculations.

[0024] The dynamic iterative calculation of real-time energy consumption priority for functional areas is based on the integrated computational dataset. It takes each functional area of ​​the building as the priority calculation object and calculates the energy consumption priority of each functional area in real time through a dynamic priority formula. The priority is iteratively updated every 100ms to ensure that the priority matches the real-time status of the source, grid, load, and storage systems. The priority calculation formula is as follows: The definitions and value rules for each factor are as follows: Basic Priority Fixed values ​​are set according to the energy importance of functional areas: 0.9 for the computer room area, 0.7 for the HVAC area, 0.6 for the office area, and 0.5 for the lighting area. These values ​​can be flexibly configured as needed. Electricity price factor : It is linked with the power grid during peak, valley and normal periods, with peak power period of 1.2, normal power period of 1.0, and valley power period of 0.8; Energy storage SOC factor It is linearly related to the energy storage state of charge, and the calculation formula is as follows: The lower the SOC, the higher the factor value, and priority is given to energy storage for power replenishment; the higher the SOC, the lower the factor value, and priority is given to meeting the energy needs of building loads. Energy urgency factor The value is dynamically adjusted based on the deviation between real-time and predicted energy consumption in the functional area, with a range of 0.9. 1.1 When the actual energy consumption is higher than the predicted energy consumption, the higher value is used; otherwise, the lower value is used. After each iteration of calculation, a real-time energy consumption priority ranking table for each functional area is generated, which serves as an important weighting basis for subsequent power allocation.

[0025] The multi-objective optimization model for source-grid-load-storage coordination is constructed with global optimization as the core principle. It includes optimization objectives, comprehensive objective functions, and multi-dimensional constraints to achieve a comprehensive balance between photovoltaic consumption, electricity purchase costs, and energy storage losses.

[0026] Optimization objectives are defined: three core optimization objectives are set, namely, maximizing the photovoltaic grid integration rate (…). Minimize the cost of purchasing electricity from the power grid. Minimize energy storage charging and discharging losses ( The formula for calculating the photovoltaic grid integration rate is as follows: The formula for calculating energy storage loss is: )( , (This is the energy storage charging and discharging loss coefficient, which is matched with the energy storage battery model).

[0027] Construction of the comprehensive objective function: A linear weighted method is used to integrate the three single objective functions into a comprehensive objective function, achieving synergistic optimization of multiple objectives. The formula is as follows: Preset weighting coefficients =0.5、 =0.3、 =0.2, prioritizing the photovoltaic absorption rate while taking into account electricity purchase costs and energy storage losses, the weighting coefficient can be flexibly configured according to operation and maintenance needs.

[0028] Multi-dimensional constraint settings: To ensure the security, compliance, and feasibility of power allocation commands, five hard constraints are set, covering the entire link from the source end, storage end, network end, and load end: Photovoltaic curtain wall zoned power constraints: That is, the power allocated to the functional area by each partition does not exceed its effective adjustable power, and is non-negative; Energy storage charging and discharging power constraints: , That is, the energy storage charging and discharging power is within the rated limit range of the equipment; Energy storage SOC constraint: 20%≤SOC≤90% to avoid battery damage caused by overcharging or over-discharging of energy storage; Power grid interaction constraints: , That is, the power purchased and sold by the power grid shall not exceed the limit approved by the power grid; Load-end power balance constraints: The total energy demand of each functional area of ​​the building is met by the coordinated power supply of the photovoltaic curtain wall, the energy storage power supply, and the power grid, so as to achieve real-time power balance.

[0029] The optimal power allocation value is determined based on the constructed comprehensive objective function and five major constraints. The Lagrange multiplier method is used to solve for the optimal solution of the objective function with constraints, and the optimal power allocation value from each zone of the photovoltaic curtain wall to each functional area of ​​the building is accurately calculated. The core solution steps are as follows: Transform the five major constraints into a general constraint form. or ; Constructing the Lagrange function: ,in It is a Lagrange multiplier, and ≥0; For the power distribution variable in the Lagrange function Find the first-order partial derivatives and set them to 0 to obtain a system of partial derivative equations; Solving the partial derivative equations, and considering the boundary values ​​of the constraints, invalid solutions are eliminated, ultimately yielding the optimal power allocation value that satisfies all constraints. At the same time, the optimal charging and discharging power of energy storage and the optimal power purchase / sale power of the power grid are calculated.

[0030] Standardized instruction generation and high-speed output will solve for the optimal power allocation value. The optimal charging and discharging power of energy storage and the optimal power purchase / sale power of the grid are converted into standardized power allocation instructions. These instructions contain core information: physical zoning identifiers for the photovoltaic curtain wall, functional zone identifiers for the building, target adjustment power values, power adjustment time limits, and energy storage / grid collaborative action parameters. After generation, the instructions are synchronously pushed to the zoned power execution and control module and the closed-loop feedback and self-learning optimization module via high-speed industrial Ethernet. The instruction transmission delay is strictly controlled to ≤20ms, ensuring that the execution layer can receive and execute allocation instructions in real time. Simultaneously, it provides instruction baseline data to the feedback layer, achieving high-speed linkage between calculation, execution, and feedback.

[0031] The partitioned power control module, with its core logic of receiving, parsing, executing, verifying, and providing feedback on power allocation commands, completes the entire power control process through five continuous and closed-loop steps: command reception and parsing, actuator status detection and activation, smooth power adjustment execution, fault self-diagnosis and redundancy switching, and execution result feedback and log recording. The specific process is as follows: Command reception and precise parsing: Receives standardized commands for optimal power allocation in real time via high-speed industrial Ethernet, with command transmission latency strictly controlled to ≤20ms to ensure timely execution; Upon receiving the instruction, the module's built-in parsing unit immediately performs structured parsing, extracting the core key information, including the unique identifier of the photoelectric curtain wall's physical partition, the unique identifier of the building's functional area, and the target adjustment power value. The system includes parameters such as power regulation time limits and energy storage / grid coordinated action parameters. It also performs format verification and validity judgment on the parsed information, eliminating invalid commands and correcting commands with abnormal formats. At the same time, it accurately locates the corresponding distributed intelligent execution node based on the zoning identifier and functional area identifier. These nodes are deployed at the power connection terminals of each physical zone of the photovoltaic curtain wall and each functional area of ​​the building to achieve one-to-one precise control of zoning power regulation.

[0032] Actuator status detection and startup: After completing instruction parsing and node location, the module immediately triggers the startup process of the corresponding distributed intelligent execution node, and the core starts the combined power electronic actuator of solid-state relay (SSR) + frequency converter, eliminating mechanical wear and regulation delay problems from the hardware level; Before startup, the execution node first performs a comprehensive self-check of its own operating status. The check includes whether the node's input / output voltage and current are within the rated range, whether the frequency converter's conduction angle adjustment function is normal, whether the solid-state relay's on / off state is controllable, and whether the communication link between the node and other modules is smooth. After the self-test is successful, the actuator officially enters the power regulation ready state and waits for the regulation command to be triggered. If the self-test detects a fault, the device number, fault type and fault occurrence time of the fault node are immediately marked and pushed to the human-machine interaction and early warning module to trigger the corresponding level of early warning. At the same time, the redundancy switching preparation process is started.

[0033] Smooth power adjustment and precise execution: After the actuator enters the ready state, it immediately adjusts the power value according to the analyzed target. The system initiates a smooth power regulation process. Its core functionality involves precise control of the frequency converter to achieve millisecond-level smooth power transitions, avoiding voltage and current fluctuations caused by sudden power increases or decreases, and preventing damage to building electrical equipment and photovoltaic curtain wall components. The regulation process strictly follows a preset smooth regulation formula: ,in The time constant is set to 0.5s by default and can be flexibly configured according to the anti-interference capabilities of the building's electrical equipment and photoelectric curtain wall components. The module collects the actual output power during the adjustment process in real time. The frequency converter dynamically adjusts the conduction angle and switching time to precisely control the smooth rise / fall of power from the current value to the target value. The entire adjustment process achieves millisecond-level response, and the power curve has no obvious fluctuations, ensuring the stability of power transmission.

[0034] Fault self-diagnosis and redundancy switching: During the power regulation process, the module performs millisecond-level real-time status monitoring on each distributed intelligent execution node. The monitoring scope includes the operating temperature of the actuator, voltage / current changes, solid-state relay on / off status, frequency converter regulation accuracy, communication link connectivity, etc., to achieve real-time self-diagnosis of faults. If any node is detected to have a solid-state relay failure, frequency converter adjustment accuracy exceeding the standard, communication interruption, or abnormal power output, the module will immediately trigger a redundancy switching mechanism to quickly transfer the power allocation adjustment task of the faulty node to a pre-deployed adjacent redundant execution node. The redundant node will continue to complete the power adjustment work according to the original instruction's target power value and adjustment time limit. The entire switching process is seamless, ensuring that the power adjustment task is not interrupted and guaranteeing the power demand of the building's functional areas and the power output stability of the photovoltaic curtain wall. At the same time, the module will push the updated fault details (including fault duration, fault impact range, and redundancy switching results) to the human-machine interaction and early warning module again to complete the real-time synchronization of fault information and provide accurate data for the subsequent handling by operation and maintenance personnel.

[0035] Execution result feedback and log recording: During the power adjustment process, the module continuously collects the actual output power data of the execution node. When the actual output power reaches the target adjustment power value, the module will adjust the output power accordingly. When the power state remains stable for 3 acquisition cycles (a total of 300ms), the module determines that the power adjustment task has been completed. Upon completion of the task, the module immediately feeds back the actual adjustment results to the closed-loop feedback and self-learning optimization module in real time via industrial Ethernet, providing real execution-level data support for the four-dimensional deviation analysis. Simultaneously, the module automatically generates a detailed execution log for this adjustment. The log includes the partition / functional area identifier corresponding to the adjustment task, the target power value, the actual power value, the adjustment time, the deviation value, the status of the actuator equipment, whether a fault occurred, and the redundancy switching situation. The execution log is also synchronously uploaded to the dedicated database of the human-machine interaction and early warning module for storage, realizing full traceability of the execution process and providing data basis for subsequent equipment operation and maintenance and parameter optimization.

[0036] All distributed intelligent execution nodes perform power adjustment in parallel, with each node independently completing the power adjustment of its corresponding zone and functional area without interference. This ensures that the power distribution of the photovoltaic curtain wall throughout the building is carried out efficiently and orderly, achieving a precise match between the power source and the load demand.

[0037] The closed-loop feedback and self-learning optimization module works by using real-time closed-loop feedback as a foundation and iterative self-learning strategies as an advanced approach. These two stages operate in parallel with data sharing. The deviation data and execution results from the closed-loop feedback provide real training samples for self-learning, while the optimization strategies from self-learning provide more accurate correction criteria for the closed-loop feedback. The specific process unfolds in two core sections, with logical connections and progressive steps at each stage to ensure effective optimization. Step 1: Four-dimensional closed-loop feedback control of power generation, grid, load and storage: Through data acquisition, deviation calculation, graded correction and command synchronization, the prediction model and calculation strategy are accurately calibrated to ensure that the actual effect of power allocation is highly matched with the theoretical optimal value.

[0038] End-to-end data acquisition and cross-validation: The module acquires real-time end-to-end operational data after the power allocation command is executed. The data sources are divided into two parts: Receive power execution result data, including the actual power output to the functional area from each zone of the photovoltaic curtain wall and the actual adjustment parameters of the actuator; Retrieve synchronized real-time acquisition data to achieve full coverage of core four-dimensional data, specifically including: Source end: Actual total output power of each zone of the photovoltaic curtain wall , Actual effective output power of each zone; Load end: Actual energy consumption power of each functional area of ​​the building Real-time energy consumption changes in each functional area; Grid end: Actual power purchased / sold by the power grid Actual electricity price settlement power; Storage side: Actual charging / discharging power of energy storage units Actual SOC value after charging and discharging, and battery temperature changes; After data collection, the module cross-validates the actual four-dimensional data with the original collected data and the optimal calculation data, removes abnormal data and fills in missing data to ensure that the data used for deviation analysis is true, complete and effective, and accurately reflects the actual execution effect of power allocation.

[0039] Multi-dimensional deviation calculation and problem location are based on cross-validated data. The module calculates the deviation value and direction for four dimensions: source, load, network, and storage. The deviation value is the absolute difference between the actual data and the theoretical / predicted data, and the deviation direction is used to determine the trend of data deviation. Simultaneously, the module locates the core link causing the deviation through deviation source analysis. If the deviation originates from the source end, it is determined to be a deviation in the parameters of the power characteristic analysis model of the photovoltaic curtain wall; If the deviation originates from the load end, it is determined that the energy consumption prediction model is not adapted to the real-time operating conditions. If the deviation originates from the network / storage end, it is determined that the constraints or weight settings of the multi-objective optimization model are unreasonable. If the deviation is a local single-point deviation, it is determined to be a problem with the adjustment accuracy of the actuator.

[0040] To achieve a refined correction strategy, the module calculates the comprehensive deviation coefficient according to preset weights. The overall impact of the four-dimensional bias is comprehensively evaluated using the following formula: The load end is the core of building energy consumption, with a preset weight. =0.4, with equal weights for the source, network, and storage ends. = = =0.2, the weight can be flexibly configured according to system operation and maintenance needs.

[0041] The module uses a comprehensive deviation coefficient With the predicted total building energy consumption The ratio is used as the basis for judgment, triggering a three-level graded correction strategy. Different deviation levels correspond to different correction methods, balancing correction accuracy and system computational efficiency. Micro deviation ( ≤5% ): Only online fine-tuning is performed, without recalculation, directly correcting the environmental correction factor. Spatiotemporal coupling factor Local parameters can be used to offset deviations caused by minor changes in operating conditions. Small deviation (5% < ≤10% ): Perform local recalculation, send correction instructions to the dynamic priority iterative power allocation core calculation module, requiring it to recalculate the power allocation values ​​only for the photovoltaic curtain wall zones and building functional zones involved in the deviation, while maintaining the original strategy for the remaining zones, thereby achieving accurate correction while controlling the amount of calculation; Large deviation ( >10% ): Initiate a global recalculation, trigger the dynamic priority iterative power allocation core operation module to rebuild the multi-objective optimization model, adjust the core parameters such as objective function weights and constraint thresholds, and resolve the optimal power allocation value of the entire system, thus completely resolving the global deviation caused by major sudden changes in operating conditions and model parameter mismatch.

[0042] Correction instruction synchronization and execution effect tracking: The module generates corresponding standardized correction instructions according to the hierarchical correction strategy. The instructions contain core information such as correction object, correction parameters, correction threshold, and recalculation range. They are pushed in real time to the corresponding source load characteristic sensing and prediction module or dynamic priority iterative power allocation core calculation module via high-speed industrial Ethernet to ensure that the correction instructions are received and executed within milliseconds. After the correction command is executed, the module continuously tracks the subsequent execution results of the partition power control module, collects the corrected four-dimensional actual data, recalculates the deviation coefficient, and verifies the correction effect. If the deviation returns to the range of minor deviation, the correction is considered successful. If the deviation is not effectively improved, the deviation source should be traced again, and the correction strategy should be adjusted until precise matching of power distribution is achieved, forming a true closed-loop feedback control.

[0043] Step 2: Iterative Reinforcement Learning Self-Learning Strategy: Using deviation data generated by closed-loop feedback control and system operation data throughout the entire chain as training samples, the core strategies and parameters of the system are autonomously iterated and optimized based on the Deep Q-Network (DQN) reinforcement learning algorithm, so as to achieve adaptive matching of the system to different working conditions and different operating states.

[0044] System operational state modeling and action space definition: The module first models the real-time operational state of the system as the state space of a Markov decision process (MDP). state space The core operating parameters include four dimensions: source, load, grid, and storage, specifically: The system incorporates key data such as the effective adjustable power of each zone of the photovoltaic curtain wall, the energy consumption characteristics of each functional area of ​​the building, the peak-valley-flat electricity price and interactive power limit of the power grid, the SOC and charging and discharging power limit of the energy storage unit, and the deviation coefficient of the closed-loop feedback and the operating status of the actuators, to ensure that the state space can comprehensively and accurately reflect the actual operation of the system. Meanwhile, the power allocation strategy adjustment of the system is defined as action space A. Action space A contains all the core operations that can be optimized, specifically: parameter adjustment of the prediction model, coefficient adjustment of dynamic priority, weight adjustment of the multi-objective optimization model, and hierarchical correction threshold adjustment of closed-loop feedback, etc., to achieve full coverage optimization of the core strategy of the system.

[0045] Comprehensive Reward Function Construction and Real-Time Reward Calculation: To guide the reinforcement learning algorithm towards the global optimum, a comprehensive reward function that is highly aligned with the system optimization objective is constructed, and the reward value is calculated in real time. To determine the merits of each strategy adjustment, the reward value is positively correlated with system operating efficiency and negatively correlated with operating costs and equipment wear and tear. The calculation formula is as follows: in: The photovoltaic grid integration rate is the core optimization objective and has the following weight. =10, the highest percentage; Weighting of electricity purchase cost to the power grid =5; Weighting of energy storage unit charging and discharging losses =3.

[0046] The weighting coefficients are preset to prioritize photovoltaic grid integration rate, while also considering electricity purchase cost and energy storage loss, and can be flexibly configured according to the core needs of system operation and maintenance. The module calculates the reward value in real time at a frequency of 100ms / time. : When the photovoltaic absorption rate increases, the cost of electricity purchase decreases, and energy storage losses decrease, the reward value... Increase; conversely, decrease. This reduces the burden and provides a clear value orientation for algorithm iteration.

[0047] DQN Model Training and Q-Value Iterative Update: The module calls the Deep Q-Network (DQN) algorithm to update the state space. Action space Comprehensive reward value As input to the model, continuous model training and Q-value iterative updates are carried out; The core hyperparameter of the model is preset as follows: learning rate =0.05, ensuring the efficiency and stability of model learning; discount factor =0.95, focusing on the long-term rewards of strategy adjustments; Exploration rate The Q-value was linearly reduced from 0.9 to 0.1, realizing the transition of the model from random exploration to precise utilization. In the early stage, the optimal solution was explored by randomly adjusting the strategy, and in the later stage, the strategy was precisely optimized based on the trained Q-value. During training, the module stores each state change, policy adjustment, and reward value change of the system as a sample in the experience replay pool. It then trains the model by randomly sampling samples from the experience replay pool, avoiding overfitting caused by sample correlation and ensuring the model can adapt to different operating conditions. Training continues until the Q-value converges, i.e., the reward value after policy adjustment. There was no significant improvement, and the system's operating efficiency tended to stabilize.

[0048] Optimal Strategy Generation and System-wide Parameter Update: After the DQN model converges, the module generates the corresponding optimal power allocation strategy based on the optimal Q-value output by the model. This strategy includes a source-load characteristic perception and prediction module, a dynamic priority iterative power allocation core computation module, and all core optimization parameters and thresholds of this module. The module then translates the optimal strategy into standardized parameter update instructions and updates the core parameters of the entire system via high-speed industrial Ethernet. Update the parameters of the source-load characteristic sensing and prediction model, including the calculation rules for the illumination / temperature correction factor and the spatiotemporal coupling factor. The range of values, the weight coefficients of the prediction model, etc., improve the accuracy of source load analysis and prediction; Update the coefficients of the dynamic priority iterative algorithm and the weights of the multi-objective optimization model, including the basic priority. SOC factor Objective function weights The system improves the global optimality of power allocation calculations; thirdly, it updates the weights of the hierarchical correction threshold and comprehensive deviation coefficient of the closed-loop feedback in this module to improve the adaptability of feedback correction.

[0049] The human-computer interaction and early warning module enables real-time linkage between all work processes and modules, with zero-delay data synchronization, precise issuance of operation commands, and instant push of early warning information, forming a complete operation and maintenance interaction management system. Multi-dimensional dynamic visualization monitoring: Based on the B / S architecture, a multi-terminal compatible interactive platform is built to integrate the core operating data of the modules in real time. It is classified by source / load / grid / storage dimensions and displays core indicators such as photovoltaic absorption rate, energy storage SOC, and power matching degree and equipment operating status in the form of digital dashboards, dynamic power distribution flow diagrams, and equipment topology diagrams. It supports accurate query of historical data in multiple time dimensions, and data updates are synchronized with the system in real time.

[0050] Dual-mode manual / automatic operation with strategy simulation interaction: A seamless switching entry point is set for manual / automatic operation modes. In automatic mode, the system runs autonomously, while in manual mode, core parameters can be adjusted online. The built-in strategy simulation function can simulate the system operation effect after parameter adjustment, predict changes in indicators such as photovoltaic absorption rate and electricity purchase cost, and issue standardized manual instructions and record them throughout the process to achieve scientific and risk-free manual intervention.

[0051] Three-tiered early warning and module-based coordinated response: Comprehensive monitoring of anomalies across the entire system chain, including equipment, data, and computational execution. Based on the impact of the anomaly on system operation and the urgency of the response, early warnings are divided into three levels: Level 1 (Urgent), Level 2 (Important), and Level 3 (General). Each level corresponds to specific triggering conditions. Level 1 Warning (Emergency): The triggering condition is an anomaly that will directly affect the safe operation of the system, cause equipment damage or building energy interruption, such as overcharging / over-discharging of energy storage (SOC < 20% or SOC > 90%), grid interaction power exceeding the approved limit, failure of core equipment of execution node, and interruption of the entire communication link. Level 2 Warning (Important): The triggering condition is an anomaly that affects the system's operating efficiency but does not directly lead to equipment damage or power interruption, such as a power distribution comprehensive deviation coefficient exceeding 10%, a photovoltaic absorption rate decrease of <80%, missing data from multiple acquisition nodes, or failure of self-learning model training. Level 3 warning (general): The triggering condition is a minor anomaly that has little impact on system operating efficiency and can be handled gradually later, such as high temperature of a single device, data deviation of a single data acquisition node, failure to push warning information, etc.

[0052] Tiered early warning system and coordinated response: Different levels of early warning trigger corresponding early warning mechanisms, and simultaneously coordinate with relevant modules for initial handling to ensure timely response to abnormal situations. The early warning mechanisms and coordinated response strategies are as follows: Level 1 Warning (Emergency): Immediately trigger audible and visual alarms (platform terminal + on-site operation and maintenance room) + SMS / APP push (all operation and maintenance personnel), and simultaneously coordinate with relevant modules to carry out emergency handling, such as immediately suspending energy storage charging and discharging commands when energy storage is overcharged / overdischarged, and immediately cutting off photovoltaic power supply to some non-critical functional areas when the grid power exceeds the limit, until operation and maintenance personnel intervene to handle the situation; Level 2 Warning (Important): Immediately triggers an audible and visual alarm (platform-side) + a platform pop-up notification (operation and maintenance personnel login terminal). The pop-up includes anomaly details, scope of impact, and handling suggestions. At the same time, the anomaly information is synchronized to the closed-loop feedback and self-learning optimization module for initial correction through self-learning optimization. Level 3 Warning (General): Only triggers a platform pop-up notification. The pop-up contains details of the anomaly and handling suggestions. It is handled uniformly by maintenance personnel during routine inspections and does not trigger any linked operations to avoid affecting the normal operation of the system.

[0053] Closed-loop management of early warnings: All early warning information is automatically entered into the early warning log, including information such as early warning level, trigger time, trigger reason, scope of impact, handling method, handling person, handling time, and handling result. After the operation and maintenance personnel complete the handling, they need to confirm on the platform. After confirmation, the early warning status is automatically lifted. Unhandled early warnings will continue to remind you on the platform until the handling is completed, forming a closed-loop management of early warning triggering - early warning handling - early warning confirmation, ensuring that no early warnings are missed.

[0054] Data Collaborative Management and Self-Learning Empowerment: Build a distributed structured database to retain data from each module in a differentiated manner, and permanently store core operation and optimization data; set up a three-level permission system (administrator / operation staff / visitor) to achieve secure control over data operations; store full lifecycle data to provide training samples for self-learning optimization, and visualize the optimization effect to achieve deep data linkage with the self-learning module.

[0055] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption, characterized in that, include: Multi-source heterogeneous sensing and acquisition module: Deployed in the functional zones of the building and the physical zones of the photovoltaic curtain wall, it collects building energy consumption parameters, photovoltaic curtain wall parameters, environmental parameters, power grid status parameters and energy storage unit parameters, and preprocesses the collected raw data to generate standardized preprocessed data; Source-load characteristic perception and prediction module: Based on the preprocessed data, analyze the effective adjustable power of each physical zone of the photovoltaic curtain wall, and perform spatiotemporal coupling prediction of the energy consumption demand of each functional area of ​​the building, generating an analysis report containing source-end power characteristics and load-end prediction data. The core calculation module for dynamic priority iterative power allocation integrates the analysis report, grid status data, and energy storage operation data. After consistency verification, it constructs and solves a multi-objective optimization model based on the functional area energy consumption priority calculated dynamically and iteratively, with photovoltaic absorption rate, grid electricity purchase cost, and energy storage loss as optimization objectives. The optimal power allocation value is obtained, and standardized power allocation instructions are generated. The partitioned power execution control module receives and parses the power allocation command, controls the distributed intelligent execution nodes deployed in the corresponding physical partitions and functional areas, drives the power electronic actuators to complete the smooth adjustment of power, and feeds back the actual execution results. Closed-loop feedback and self-learning optimization module: Collects actual data from the entire link of the photovoltaic curtain wall power supply, building energy load, power grid and energy storage unit after power allocation is executed, calculates the deviation from the theoretical value and performs hierarchical correction, and at the same time performs self-learning iterative optimization of the system's core strategies and parameters based on reinforcement learning algorithm; Human-computer interaction and early warning module: Provides multi-dimensional visual monitoring of system operation status, supports manual / automatic mode switching and strategy simulation, triggers graded early warnings based on full-link anomalies and performs coordinated handling, and realizes collaborative management of system data.

2. The photovoltaic curtain wall power dynamic distribution control system for building energy consumption according to claim 1, characterized in that, The original data includes: Real-time active power, reactive power, voltage, and current of each functional area of ​​the building; real-time output power, voltage, current, and component temperature of each physical zone of the photovoltaic curtain wall; outdoor light intensity, temperature, wind speed, rainfall, and indoor temperature and humidity of the building; real-time voltage, current, frequency, peak-valley-flat electricity price information, and power purchase / sale limits of the power grid; state of charge, charging and discharging power, battery temperature, and remaining capacity of the energy storage battery.

3. The photovoltaic curtain wall power dynamic distribution control system for building energy consumption according to claim 1, characterized in that, The effective adjustable power includes: the actual maximum output power of each physical zone, obtained by correcting it based on the nominal maximum output power and through real-time calculated light correction factor, temperature correction factor and aging correction factor; the adjustable power range of each zone obtained by subtracting the preset minimum output power of the zone from the actual maximum output power; and the total effective adjustable power of the photovoltaic curtain wall obtained by summing the adjustable power ranges of each zone.

4. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The spatiotemporal coupling prediction includes: The core input features affecting building energy consumption are screened and each feature is assigned a differentiated weight. The core features include at least historical energy consumption data of each functional area of ​​the building, real-time operating conditions, environmental parameters and grid electricity price information during the time period. Based on the energy consumption correlation of each functional area of ​​the building and the energy consumption patterns at different times, the spatiotemporal coupling factor of each functional area at each time period is calculated. An improved attention mechanism LSTM model is invoked and trained, using the core input features and spatiotemporal coupling factors as inputs, and the model is trained based on historical data. The core feature data collected in real time is substituted into the trained model to calculate the predicted energy consumption of each functional area of ​​the building at each time period. Based on actual energy consumption data, the weighting coefficients and environmental correction factors in the prediction model are dynamically adjusted.

5. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The calculation of the energy consumption priority of the functional area includes: Based on the basic priority of each functional area of ​​the building, the electricity price factor determined by the real-time electricity price period of the power grid, the energy storage SOC factor determined by the state of charge of the energy storage unit, and the energy urgency factor determined by the deviation between the real-time energy consumption and the predicted energy consumption of the functional area. By multiplying the basic priority, electricity price factor, energy storage SOC factor and energy urgency factor, the real-time energy consumption priority of each functional area is dynamically calculated. The real-time energy consumption priority is updated according to a preset iteration frequency.

6. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The construction of the multi-objective optimization model includes: The core optimization objectives are to maximize the photovoltaic absorption rate, minimize the grid power purchase cost, and minimize energy storage losses. A comprehensive objective function is constructed, and the core optimization objectives are integrated using a linear weighting method, with corresponding weight coefficients configured. Multi-dimensional constraints are set, including at least the power constraints of the photovoltaic curtain wall partition, the power and SOC constraints of energy storage charging and discharging, the power constraints of grid interaction, and the power balance constraints of the load end.

7. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The smoothing adjustment includes: Based on the target adjustment power value parsed from the power distribution instruction, the frequency converter is controlled to make the actual output power smoothly transition to the target adjustment power value according to the exponential function curve. In this process, by adjusting the conduction angle and switching time of the frequency converter, the rise or fall of the actual output power is controlled to ensure that the power curve does not change abruptly.

8. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The graded correction of the deviation includes: Calculate the comprehensive deviation coefficient of the four-dimensional data of source, load, grid, and storage; and trigger different levels of correction strategies based on the ratio of the comprehensive deviation coefficient to the predicted total building energy consumption. When the ratio does not exceed the first threshold, online fine-tuning is performed, and only local parameters are corrected; When the ratio exceeds the first threshold but does not exceed the second threshold, a local recalculation is performed to recalculate the power allocation values ​​for the partitions and functional areas involved in the deviation; when the ratio exceeds the second threshold, a global recalculation is performed to reconstruct and solve the multi-objective optimization model.

9. A dynamic power distribution control system for photovoltaic curtain walls oriented towards building energy consumption according to claim 1, characterized in that, The tiered early warning system includes: Based on the impact of the abnormal situation on system operation and the urgency of the response, the early warning is divided into three levels: Level 1, Level 2, and Level 3, each with clearly defined triggering conditions. Level 1 warning: The trigger condition is an anomaly that will directly affect the safe operation of the system, cause equipment damage, or interrupt the building's energy supply; Level 2 warning: The trigger condition is an anomaly that affects system operating efficiency but does not directly cause equipment damage or power interruption; Level 3 warning: The trigger condition is a minor anomaly that has little impact on system efficiency and can be dealt with gradually.

Images